CN115049836A - Image segmentation method, device, equipment and storage medium - Google Patents

Image segmentation method, device, equipment and storage medium Download PDF

Info

Publication number
CN115049836A
CN115049836A CN202210978696.XA CN202210978696A CN115049836A CN 115049836 A CN115049836 A CN 115049836A CN 202210978696 A CN202210978696 A CN 202210978696A CN 115049836 A CN115049836 A CN 115049836A
Authority
CN
China
Prior art keywords
image
image segmentation
confidence coefficient
activation
label
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202210978696.XA
Other languages
Chinese (zh)
Other versions
CN115049836B (en
Inventor
郑喜民
肖京
舒畅
陈又新
胡浩楠
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Priority to CN202210978696.XA priority Critical patent/CN115049836B/en
Publication of CN115049836A publication Critical patent/CN115049836A/en
Application granted granted Critical
Publication of CN115049836B publication Critical patent/CN115049836B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20212—Image combination
    • G06T2207/20221—Image fusion; Image merging
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/08—Detecting or categorising vehicles

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Multimedia (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computing Systems (AREA)
  • Databases & Information Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Image Analysis (AREA)

Abstract

The invention relates to the field of artificial intelligence, and discloses an image segmentation method, which comprises the following steps: segmenting the unmarked image set by using the trained image segmentation model to obtain a first image confidence coefficient; performing category activation operation on the confidence coefficient of the first image to obtain a category activation image; the category activation image is subjected to up-sampling to obtain a second image confidence coefficient; fusing the first image confidence coefficient and the second image confidence coefficient to obtain a pseudo label of the unmarked image set; training a preset target image segmentation model by using a training data set and a final label to obtain a trained target image segmentation model; and segmenting the image to be segmented by utilizing the trained target image segmentation model to obtain an image segmentation result. The invention also relates to a block chain technology, and the image to be segmented can be stored in the block chain node. The invention also provides an image segmentation device, equipment and a medium. The invention can improve the accuracy of image segmentation.

Description

Image segmentation method, device, equipment and storage medium
Technical Field
The present invention relates to the field of artificial intelligence, and in particular, to an image segmentation method, apparatus, device, and storage medium.
Background
Currently, with the rapid development of the deep learning technology, the image segmentation technology based on the deep learning is widely applied to the fields of insurance, finance and the like with the rapid and accurate image segmentation capability. Most of the existing image segmentation algorithms train a fully supervised learning model by adopting an artificially labeled image set, and a prediction label obtained by segmenting and predicting an unlabeled image is directly used as a pseudo label of the unlabeled image by using the model. However, the prediction result of the fully supervised learning model is not completely accurate, and some pseudo labels with poor accuracy influence poor image segmentation results, so that the accuracy of image segmentation is low.
Disclosure of Invention
The invention provides an image segmentation method, an image segmentation device, image segmentation equipment and a storage medium, and aims to improve the accuracy of image segmentation.
In order to achieve the above object, the present invention provides an image segmentation method, comprising:
acquiring an unlabeled image set, an labeled image set and a real image label corresponding to the labeled image set, and segmenting the unlabeled image set by using a trained image segmentation model to obtain a first image confidence coefficient;
performing category activation operation on the first image confidence coefficient to obtain a category activation graph of the unmarked image set;
the category activation image is subjected to up-sampling to obtain a second image confidence coefficient;
fusing the first image confidence coefficient and the second image confidence coefficient to obtain a pseudo label of the unmarked image set;
integrating the real image labels and the pseudo labels to obtain final labels, taking the unmarked image sets and the marked image sets as training data sets, and training a preset target image segmentation model by using the training data sets and the final labels to obtain a trained target image segmentation model;
and acquiring an image to be segmented, and segmenting the image to be segmented by using the trained target image segmentation model to obtain an image segmentation result of the image to be segmented.
Optionally, performing a category activation operation on the first image confidence to obtain a category activation map of the unlabeled image set, including:
acquiring a convolution characteristic diagram of the unmarked image set, and setting the activation value of the convolution characteristic diagram to zero to obtain the activation image confidence of the unmarked image set;
calculating a difference set of the first image confidence coefficient and the activation image confidence coefficient, screening out a positive difference value larger than zero from the difference set, and dividing the positive difference value by the number of pixel points corresponding to the category activation confidence coefficient to obtain a fusion weight;
and weighting the convolution characteristic diagram by using the fusion weight, and inputting the weighted characteristic diagram into a preset activation function to obtain the category activation diagram.
Optionally, the fusing the first image confidence coefficient and the second image confidence coefficient to obtain a pseudo label of the unlabeled image set includes:
acquiring a first weight coefficient of the first image confidence coefficient, and multiplying the first weight coefficient by the first image confidence coefficient to obtain a first parameter;
acquiring a second weight coefficient of the second image confidence coefficient, and multiplying the second weight coefficient by the second image confidence coefficient to obtain a second parameter;
summing the first parameter and the second parameter to obtain a fusion parameter;
judging whether the fusion parameter is smaller than a preset threshold value or not;
when the fusion parameter is smaller than a preset threshold value, screening out a label corresponding to the fusion parameter;
and when the fusion parameter is not less than a preset threshold value, taking the label corresponding to the fusion parameter as the pseudo label.
Optionally, the upsampling the class activation map to obtain a second image confidence level includes:
and sequentially selecting four adjacent pixel values of the category activation graph, and performing linear interpolation on each corresponding pixel value in the unmarked image in the horizontal direction and the vertical direction by using the four adjacent pixel values to obtain the confidence coefficient of the second image.
Optionally, the training a preset target image segmentation model by using the training data set and the final label to obtain a trained target image segmentation model includes:
extracting the characteristics of the training data set by using a cavity convolution layer in a preset target image segmentation model and carrying out characteristic fusion to obtain a characteristic fusion image set;
performing target object identification on the feature fusion image set by using a pyramid pooling layer in the preset target image segmentation model to obtain a target object identification image set;
inputting the target object identification image set into an activation function in the preset target image segmentation model to obtain a predicted image segmentation label of the training data set output by the activation function;
calculating loss values of a predicted image segmentation label and the final label by using a loss function in the preset target image segmentation model, and adjusting parameters of the preset target image segmentation model according to the loss values until the loss values meet preset conditions to obtain the trained target image segmentation model.
Optionally, the segmenting the unlabeled image set by using the trained image segmentation model to obtain a first image confidence coefficient includes:
performing convolution operation on the unmarked image set by using the trained image segmentation model to obtain an image characteristic data set;
performing pooling operation on the image characteristic data set to obtain a pooled data set;
and performing activation operation on the pooled data set by using an activation function to obtain a first image confidence corresponding to the pooled data in the pooled data set.
In order to solve the above problem, the present invention also provides an image segmentation apparatus comprising:
the first image segmentation module is used for acquiring an unlabeled image set, an labeled image set and a real image label corresponding to the labeled image set, and segmenting the unlabeled image set by using a trained image segmentation model to obtain a first image confidence coefficient;
the image category activation module is used for carrying out category activation operation on the first image confidence coefficient to obtain a category activation graph of the unmarked image set;
the image up-sampling module is used for up-sampling the category activation map to obtain a second image confidence coefficient;
the image pseudo label generating module is used for fusing the first image confidence coefficient and the second image confidence coefficient to obtain a pseudo label of the unmarked image set;
the target image segmentation model training module is used for integrating the real image labels and the pseudo labels to obtain final labels, taking the unmarked image sets and the marked image sets as training data sets, and training a preset target image segmentation model by using the training data sets and the final labels to obtain a trained target image segmentation model;
and the image segmentation module is used for acquiring an image to be segmented, and segmenting the image to be segmented by using the trained target image segmentation model to obtain an image segmentation result of the image to be segmented.
In order to solve the above problem, the present invention also provides an electronic device, including:
a memory storing at least one computer program; and
and a processor executing the computer program stored in the memory to implement the image segmentation method described above.
In order to solve the above problem, the present invention also provides a computer-readable storage medium, in which at least one computer program is stored, the at least one computer program being executed by a processor in an electronic device to implement the image segmentation method described above.
In the embodiment of the invention, firstly, the unmarked image set is segmented by utilizing the trained image segmentation model to obtain the confidence coefficient of the first image, which is convenient for subsequently judging which image areas in the unmarked image set are close to the real label distribution of the unmarked image; secondly, performing a class activation operation on the first image confidence coefficient to determine the importance degree of each image pixel in the unmarked segmentation image, further performing upsampling on the class activation image to obtain a second image confidence coefficient, recovering the size of the class activation image to the size consistent with that of the unmarked image, further determining the importance degree of each image pixel, improving the accuracy of subsequent pseudo label screening, then obtaining the pseudo label of the unmarked image set by fusing the first image confidence coefficient and the second image confidence coefficient, screening out the image pixel label with lower confidence coefficient to remove the pseudo label with poorer accuracy, and simultaneously ensuring the integrity of image information and improving the accuracy of subsequent image segmentation; and finally, training a target image segmentation model through the real image label, the pseudo label, the unmarked image set and the marked image set, training the target image segmentation model by using complete and accurate image information, improving the accuracy of the model in image segmentation, and performing image segmentation on an image to be segmented by using the model to obtain an image segmentation result, so that the accuracy of image segmentation can be improved. Therefore, the image segmentation method, the image segmentation device, the image segmentation equipment and the storage medium provided by the embodiment of the invention can improve the accuracy of image segmentation.
Drawings
Fig. 1 is a schematic flowchart of an image segmentation method according to an embodiment of the present invention;
FIG. 2 is a detailed flowchart illustrating a step of an image segmentation method according to an embodiment of the present invention;
FIG. 3 is a detailed flowchart illustrating a step of an image segmentation method according to an embodiment of the present invention;
FIG. 4 is a block diagram of an image segmentation apparatus according to an embodiment of the present invention;
fig. 5 is a schematic internal structural diagram of an electronic device for implementing an image segmentation method according to an embodiment of the present invention;
the implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The embodiment of the invention provides an image segmentation method. The execution subject of the image segmentation method includes, but is not limited to, at least one of electronic devices such as a server and a terminal, which can be configured to execute the method provided by the embodiments of the present application. In other words, the image segmentation method may be performed by software or hardware installed in a terminal device or a server device, and the software may be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, and the like.
Referring to a flowchart of an image segmentation method provided in an embodiment of the present invention shown in fig. 1, in the embodiment of the present invention, the image segmentation method includes the following steps S1-S6:
s1, obtaining an unmarked image set, an marked image set and a real image label corresponding to the marked image set, and segmenting the unmarked image set by using the trained image segmentation model to obtain a first image confidence coefficient.
In the embodiment of the invention, the unlabeled image set refers to a label-free segmented image set; the labeling image set is a segmented image set with labels; the real image label is a real result of image segmentation obtained after the labeling image set is manually labeled; the first image confidence level refers to a probability that the target object exists in the unlabeled image set, for example, in a car insurance scene, the unlabeled image set may be a probability that a scratch, i.e., a concave-convex portion, of the target vehicle is identified.
In the embodiment of the invention, the trained image segmentation model is a model trained by a labeled image set, and the model is mainly used for carrying out image segmentation on an unlabeled image set by a deep learning algorithm, so that the first image confidence coefficient of a target object in the unlabeled image set is identified. Wherein the trained image segmentation model comprises: convolutional layers, pooling layers, and activation functions.
According to the embodiment of the invention, the unmarked image set is segmented by utilizing the trained image segmentation model to obtain the confidence coefficient of the first image, which is convenient for subsequently judging which image areas in the unmarked image set are close to the real label distribution of the unmarked image, and the accuracy of subsequent image segmentation is improved.
As an embodiment of the present invention, the segmenting the unlabeled image set by using the trained image segmentation model to obtain a first image confidence coefficient includes:
performing convolution operation on the unmarked image set by using the trained image segmentation model to obtain an image characteristic data set; performing pooling operation on the image characteristic data set to obtain a pooled data set; and performing activation operation on the pooled data set by using an activation function to obtain a first image confidence corresponding to the pooled data in the pooled data set.
The convolution operation can extract image characteristic data in an unmarked image set; the pooling operation can perform dimension reduction operation on the image feature data set, so that the image segmentation efficiency can be improved by reducing the calculated amount while maintaining the key information of the image; the activation function may be a Sigmoid function, and may activate the unlabeled image set to obtain a first image confidence of the unlabeled image set output by the activation function.
And S2, carrying out category activation operation on the first image confidence coefficient to obtain a category activation graph of the unlabeled image set.
In the embodiment of the present invention, the category activation map refers to a binary map having a size consistent with that of the unmarked segmented image. The category activation operation may be handled by an association-CAM (Ablation-visualization technique).
Furthermore, the embodiment of the invention performs the class activation operation on the first image confidence coefficient to obtain the class activation map of the unlabeled image set, so that the importance degree of each image pixel in the unlabeled segmentation map can be determined, and the accuracy of the subsequent image segmentation can be improved conveniently.
As an embodiment of the present invention, referring to fig. 2, the performing a category activation operation on the first image confidence to obtain a category activation map of the unlabeled image set includes the following steps S21-S23:
s21, acquiring a convolution feature map of the unlabeled image set, and setting the activation value of the convolution feature map to zero to obtain the activation image confidence of the unlabeled image set;
s22, calculating a difference set of the first image confidence coefficient and the activation image confidence coefficient, screening out a positive difference value larger than zero from the difference set, and dividing the positive difference value by the number of pixel points corresponding to the category activation confidence coefficient to obtain a fusion weight;
s23, weighting the convolution characteristic diagram by using the fusion weight, and inputting the weighted characteristic diagram into a preset activation function to obtain the category activation diagram.
The obtaining of the convolution feature acquisition graph can extract a feature graph corresponding to the last layer of convolution from an image feature data set obtained after the convolution of the image segmentation model; and setting the activation value of the convolution characteristic diagram to zero, namely setting the activation value of a convolution channel to 0, so that distortion of the unmarked segmentation image can be avoided.
In an embodiment of the present invention, the fusion weight is used to determine the importance degree of each image position in the unlabeled segmentation graph, and the fusion weight can be implemented by the following formula:
Figure 220351DEST_PATH_IMAGE001
wherein the content of the first and second substances,
Figure 864959DEST_PATH_IMAGE002
representing a fusion weight; m pred Representing a first image confidence;
Figure 462294DEST_PATH_IMAGE003
representing a class activation confidence for an ith image pixel in a kth convolution channel;
Figure 217760DEST_PATH_IMAGE004
representing a positive difference between the first image confidence and the activation image confidence; p i Representing the ith image pixel value.
Further, the category activation map may be implemented by the following formula:
Figure 669470DEST_PATH_IMAGE005
wherein L is i Representing a category activation graph; ReLU denotes activation function; k refers to the convolution channel;
Figure 484979DEST_PATH_IMAGE002
representing the fusion weight of the ith image pixel in the kth convolution channel;
Figure 304031DEST_PATH_IMAGE006
a feature map representing the ith image pixel in the kth channel.
And S3, performing up-sampling on the class activation map to obtain a second image confidence.
In this embodiment of the present invention, the second image confidence level refers to a probability that the target object exists in the class activation map. Further, since the class activation map is obtained by convolving the feature map, the size of the feature map is often smaller than that of the original map, the image needs to be restored to the original size for subsequent image segmentation, and the process of restoring the image to the original size is called upsampling.
According to the method and the device, the class activation image is up-sampled to obtain the confidence coefficient of the second image, the size of the class activation image can be restored to the size consistent with the size of the unmarked image, the importance degree of each image pixel is further determined, and the accuracy of subsequent pseudo label screening is improved.
As an embodiment of the present invention, the upsampling the class activation map to obtain a second image confidence level includes:
and sequentially selecting four adjacent pixel values of the category activation graph, and performing linear interpolation on each corresponding pixel value in the unmarked image in the horizontal direction and the vertical direction by using the four adjacent pixel values to obtain the confidence coefficient of the second image.
The four adjacent pixel values are used for carrying out linear interpolation on each corresponding pixel value in the unmarked image in the horizontal direction and the vertical direction respectively, namely, the corresponding pixel values are subjected to linear interpolation in the horizontal direction twice through the four adjacent pixel values, the interpolation obtained in the horizontal direction is substituted into the linear interpolation in the vertical direction, weights corresponding to the four adjacent pixel points can be obtained, and each pixel point in the four adjacent pixel points is multiplied by the weight and summed respectively, so that the confidence corresponding to the pixel value is obtained.
Specifically, one pixel value in the unlabeled image set is f (x, y), and the four adjacent pixel values in the category activation map are respectively f (Q) 11 )=(x 1 ,y 1 )，f(Q 12 )=(x 1 ,y 2 )，f(Q 21 )=(x 2 ,y 1 ) And f (Q) 22 )=(x 2 ,y 2 )。
In an embodiment of the present invention, a linear interpolation value in the horizontal direction can be obtained according to the pixel values:
Figure 863188DEST_PATH_IMAGE007
Figure 700563DEST_PATH_IMAGE008
a linear interpolation value in the vertical direction can be obtained from the pixel values:
Figure 421394DEST_PATH_IMAGE009
Figure 727742DEST_PATH_IMAGE010
wherein, the x 2 -x 1 =1，y 2 -y 1 =1, the pixel value f (x, y) can be expressed as:
Figure 825011DEST_PATH_IMAGE011
wherein (x) 2 -x)(y 2 -y) can be represented as f (Q) 11 ) Corresponding weights, (x-x) 1 )(y 2 -y) can be represented as f (Q) 21 ) Corresponding weight, (x) 2 -x)(y-y 1 ) Is expressed as f (Q) 12 ) Corresponding weights, (x-x) 1 )(y-y 1 ) Is expressed as f (Q) 22 ) The corresponding weight.
Wherein f (x, y) represents any pixel point in the unmarked image set, x represents the abscissa of the pixel point, and y represents the ordinate of the pixel point; f (Q) 11 ) Representing pixel points coinciding with the pixel positions of the top left image in the class activation map, and f (Q) 11 ) Selection of pixel pointsTaking the image to be horizontally moved to the right or vertically moved to the bottom right corner in the category activation map until the pixel position is coincident with the pixel position of the image at the bottom right corner in the category activation map, and pixel point x 1 Abscissa, y, representing pixel point 1 Representing the vertical coordinate of the pixel point; f (Q) 12 ) Is shown in the class activation graph with f (Q) 11 ) Horizontally adjacent pixel points, x 1 Abscissa, y, representing pixel point 1 Representing the ordinate of the pixel; f (Q) 21 ) Is shown in the class activation graph with f (Q) 11 ) Vertically adjacent pixel points, x 2 Abscissa, y, representing pixel point 1 Representing the vertical coordinate of the pixel point; f (Q) 22 ) Is shown in the class activation graph with f (Q) 21 ) Horizontally adjacent pixel points, x 2 Abscissa, y, representing pixel point 2 Indicating the ordinate of the pixel.
And S4, fusing the first image confidence coefficient and the second image confidence coefficient to obtain a pseudo label of the unmarked image set.
In the embodiment of the invention, the pseudo label refers to a segmented image label corresponding to an unlabeled image set.
According to the embodiment of the invention, the first image confidence coefficient and the second image confidence coefficient are fused to obtain the pseudo label of the unmarked image set, the image pixel label with lower confidence coefficient can be screened out to remove the pseudo label with poorer accuracy, and the pseudo label corresponding to the confidence coefficient of each image pixel can be finally determined according to the fused confidence coefficient, so that the useful information in the image is avoided being lost while the low confidence coefficient pixel is screened out, the integrity of the image information is ensured, and the accuracy of the subsequent image segmentation is improved.
As an embodiment of the present invention, referring to fig. 3, the fusing the first image confidence and the second image confidence to obtain the pseudo label of the unlabeled image set includes the following steps S41-S46:
s41, obtaining a first weight coefficient of the first image confidence coefficient, and multiplying the first weight coefficient by the first image confidence coefficient to obtain a first parameter;
s42, obtaining a second weight coefficient of the second image confidence coefficient, and multiplying the second weight coefficient by the second image confidence coefficient to obtain a second parameter;
s43, summing the first parameter and the second parameter to obtain a fusion parameter;
s44, judging whether the fusion parameters are smaller than a preset threshold value;
s45, when the fusion parameter is smaller than a preset threshold value, screening out the label corresponding to the fusion parameter;
and S46, when the fusion parameter is not less than a preset threshold value, taking the label corresponding to the fusion parameter as the pseudo label.
Wherein the first weight coefficient represents a degree of importance of the first image confidence; the second weight coefficient represents the degree of importance of the second image confidence, and the value of the sum of the first weight coefficient and the second weight coefficient is 1.
In an embodiment of the present invention, the preset threshold may be customized according to a specific scene, for example, in an identification scene of a vehicle picture in the field of vehicle insurance, the preset threshold may be 0.7, and if the fusion parameter exceeds 0.7, it is identified that a vehicle exists in an undivided image set; if the fusion parameter does not exceed 0.7, it indicates that no vehicle information in the set of undivided images is identified.
S5, integrating the real image labels and the pseudo labels to obtain final labels, taking the unmarked image sets and the marked image sets as training data sets, and training preset target image segmentation models by using the training data sets and the final labels to obtain trained target image segmentation models.
In the embodiment of the present invention, the preset target image segmentation model may be a deep learning model CNN or RNN.
In an embodiment of the present invention, the pseudo label in the final label is a label corresponding to the unlabeled image set, and since the image pixel label with poor accuracy has been removed, the accuracy of the pseudo label, that is, the accuracy of the final label, can be ensured, and when the preset target image segmentation model is trained by using the training data set and the final label, the accuracy of the model is improved.
According to the embodiment of the invention, the real image labels and the pseudo labels are integrated to obtain the final labels, the unmarked image set and the marked image set are used as training data sets, and the training data sets are used for training the preset target image segmentation model to obtain the trained target image segmentation model, so that the semi-supervised image segmentation model training can be realized, the model training efficiency is improved, the complete and accurate image information is used for training the target image segmentation model, and the accuracy of the model in image segmentation is improved.
As an embodiment of the present invention, the training a preset target image segmentation model by using the training data set and the final label to obtain a trained target image segmentation model includes:
extracting the characteristics of the training data set by using a cavity convolution layer in a preset target image segmentation model and carrying out characteristic fusion to obtain a characteristic fusion image set; performing target object identification on the feature fusion image set by using a pyramid pooling layer in the preset target image segmentation model to obtain a target object identification image set; inputting the target object identification image set into an activation function in the preset target image segmentation model to obtain a predicted image segmentation label of the training data set output by the activation function; calculating loss values of a predicted image segmentation label and the final label by using a loss function in the preset target image segmentation model, and adjusting parameters of the preset target image segmentation model according to the loss values until the loss values meet preset conditions to obtain the trained target image segmentation model.
Compared with the common convolutional layer, the hole convolutional layer has more special holes and wider visual field, so that when the features are extracted, the global information of more images can be mastered through feature fusion under the condition that the images are reduced by the same multiple. Further, the pyramid pooling layer may replace a general pooling layer, and a divided image having a uniform image size may be output regardless of the size of the input image. Further, in this embodiment of the present invention, the activation function may be a Sigmoid function, and the target object recognition image set may be activated to obtain a predictive image segmentation label of the training data set output by the activation function.
In the embodiment of the present invention, the preset condition may be set according to the actual model training scenario, for example, the preset condition may be that the loss value is smaller than a preset threshold.
In detail, the calculating the loss values of the predicted image segmentation tag and the final tag by using the loss function in the preset target image segmentation model comprises:
calculating a loss value for a predicted image partition label and the final label using the following loss function:
Figure 779542DEST_PATH_IMAGE012
wherein L(s) represents a loss value; k represents the number of predicted image partition tags; j represents a variable of the number of predicted image segmentation tags; y is i Representing the ith prediction image partition tag;
Figure 671274DEST_PATH_IMAGE013
indicating the ith final label.
And S6, acquiring an image to be segmented, and segmenting the image to be segmented by using the trained target image segmentation model to obtain an image segmentation result of the image to be segmented.
In the embodiment of the invention, the image to be segmented can be a damaged image of a vehicle with a vehicle risk, and can be acquired from a database of a client platform.
In the embodiment of the invention, the trained target image segmentation model can be used for segmenting the vehicle information in the vehicle damage image and the vehicle scratch set vehicle impact position to obtain the vehicle damage detail image.
In the embodiment of the invention, firstly, the unmarked image set is segmented by utilizing the trained image segmentation model to obtain the confidence coefficient of the first image, which is convenient for subsequently judging which image areas in the unmarked image set are close to the real label distribution of the unmarked image; secondly, performing a class activation operation on the first image confidence coefficient to determine the importance degree of each image pixel in the unmarked segmentation image, further performing upsampling on the class activation image to obtain a second image confidence coefficient, recovering the size of the class activation image to the size consistent with that of the unmarked image, further determining the importance degree of each image pixel, improving the accuracy of subsequent pseudo label screening, then obtaining the pseudo label of the unmarked image set by fusing the first image confidence coefficient and the second image confidence coefficient, screening out the image pixel label with lower confidence coefficient to remove the pseudo label with poorer accuracy, and simultaneously ensuring the integrity of image information and improving the accuracy of subsequent image segmentation; and finally, training a target image segmentation model through the real image label, the pseudo label, the unmarked image set and the marked image set, training the target image segmentation model by using complete and accurate image information, improving the accuracy of the model in image segmentation, and performing image segmentation on an image to be segmented by using the model to obtain an image segmentation result, so that the accuracy of image segmentation can be improved. Therefore, the image segmentation method provided by the embodiment of the invention can improve the accuracy of image segmentation.
The image segmentation apparatus 100 according to the present invention may be installed in an electronic device. According to the implemented functions, the image segmentation apparatus may include a first image segmentation module 101, an image class activation module 102, an image upsampling module 103, an image pseudo tag generation module 104, a target image segmentation model training module 105, and an image segmentation module 106, which may also be referred to as a unit, and refer to a series of computer program segments capable of being executed by a processor of an electronic device and performing a fixed function, and which are stored in a memory of the electronic device.
In the present embodiment, the functions regarding the respective modules/units are as follows:
the first image segmentation module 101 is configured to obtain an unlabeled image set, an labeled image set, and a real image label corresponding to the labeled image set, and segment the unlabeled image set by using a trained image segmentation model to obtain a first image confidence.
In the embodiment of the invention, the unlabeled image set refers to a label-free segmented image set; the labeling image set is a segmented image set with labels; the real image label is a real result of image segmentation obtained after the labeling image set is manually labeled; the first image confidence level refers to a probability that the target object exists in the unlabeled image set, for example, in a car insurance scene, the unlabeled image set may be a probability that a scratch, i.e., a concave-convex portion, of the target vehicle is identified.
In the embodiment of the invention, the trained image segmentation model is a model trained by a labeled image set, and the model is mainly used for carrying out image segmentation on an unlabeled image set by a deep learning algorithm, so that the first image confidence coefficient of a target object in the unlabeled image set is identified. Wherein the trained image segmentation model comprises: convolutional layers, pooling layers, and activation functions.
According to the embodiment of the invention, the unmarked image set is segmented by utilizing the trained image segmentation model to obtain the confidence coefficient of the first image, which is convenient for subsequently judging which image areas in the unmarked image set are close to the real label distribution of the unmarked image, and the accuracy of subsequent image segmentation is improved.
As an embodiment of the present invention, the first image segmentation module 101 segments the unlabeled image set by using a trained image segmentation model by performing the following operations to obtain a first image confidence level, including:
performing convolution operation on the unmarked image set by using the trained image segmentation model to obtain an image characteristic data set; performing pooling operation on the image characteristic data set to obtain a pooled data set; and performing activation operation on the pooled data set by using an activation function to obtain a first image confidence corresponding to the pooled data in the pooled data set.
The convolution operation can extract image characteristic data in the unmarked image set; the pooling operation can perform dimension reduction operation on the image feature data set, so that the image segmentation efficiency can be improved by reducing the calculated amount while maintaining the key information of the image; the activation function may be a Sigmoid function, and may activate the unlabeled image set to obtain a first image confidence of the unlabeled image set output by the activation function.
The image category activation module 102 is configured to perform a category activation operation on the first image confidence level to obtain a category activation map of the unlabeled image set.
In the embodiment of the present invention, the category activation map refers to a binary map having a size consistent with that of the unmarked segmented image. The category activation operation may be handled by an association-CAM (Ablation-visualization technique).
Furthermore, the embodiment of the invention performs the class activation operation on the first image confidence coefficient to obtain the class activation map of the unlabeled image set, so that the importance degree of each image pixel in the unlabeled segmentation map can be determined, and the accuracy of the subsequent image segmentation can be improved conveniently.
As an embodiment of the present invention, the image category activation module 102 performs a category activation operation on the first image confidence level by performing the following operations to obtain a category activation map of the unlabeled image set, including:
acquiring a convolution characteristic diagram of the unmarked image set, and setting the activation value of the convolution characteristic diagram to zero to obtain the activation image confidence of the unmarked image set;
calculating a difference set of the first image confidence coefficient and the activation image confidence coefficient, screening out a positive difference value larger than zero from the difference set, and dividing the positive difference value by the number of pixel points corresponding to the category activation confidence coefficient to obtain a fusion weight;
and weighting the convolution characteristic diagram by using the fusion weight, and inputting the weighted characteristic diagram into a preset activation function to obtain the category activation diagram.
The obtaining of the convolution feature acquisition graph can extract a feature graph corresponding to the last layer of convolution from an image feature data set obtained after the convolution of the image segmentation model; and setting the activation value of the convolution characteristic diagram to zero, namely setting the activation value of a convolution channel to 0, so that distortion of the unmarked segmentation image can be avoided.
In an embodiment of the present invention, the fusion weight is used to determine the importance degree of each image position in the unlabeled segmentation graph, and the fusion weight can be implemented by the following formula:
Figure 464918DEST_PATH_IMAGE001
wherein the content of the first and second substances,
Figure 365878DEST_PATH_IMAGE002
representing a fusion weight; m pred Representing a first image confidence;
Figure 522053DEST_PATH_IMAGE003
representing a class activation confidence for an ith image pixel in a kth convolution channel;
Figure 709320DEST_PATH_IMAGE004
representing a positive difference between the first image confidence and the activation image confidence; p i Representing the ith image pixel value.
Further, the category activation map may be implemented by the following formula:
Figure 849315DEST_PATH_IMAGE005
wherein L is i Representing a category activation graph; ReLU denotes the activation function; k refers to the convolution channel;
Figure 429332DEST_PATH_IMAGE002
representing the ith of the kth convolution channelA fusion weight of an image pixel;
Figure 705592DEST_PATH_IMAGE006
a feature map representing the ith image pixel in the kth channel.
The image upsampling module 103 is configured to upsample the class activation map to obtain a second image confidence.
In this embodiment of the present invention, the second image confidence level refers to a probability that the target object exists in the class activation map. Further, since the class activation map is obtained by convolving the feature map, the size of the feature map is often smaller than that of the original map, the image needs to be restored to the original size for subsequent image segmentation, and the process of restoring the image to the original size is called upsampling.
According to the method and the device, the class activation image is up-sampled to obtain the confidence coefficient of the second image, the size of the class activation image can be restored to the size consistent with the size of the unmarked image, the importance degree of each image pixel is further determined, and the accuracy of subsequent pseudo label screening is improved.
As an embodiment of the present invention, the image upsampling module 103 upsamples the class activation map to obtain a second image confidence level by performing the following operations, including:
and sequentially selecting four adjacent pixel values of the category activation graph, and performing linear interpolation on each corresponding pixel value in the unmarked image in the horizontal direction and the vertical direction by using the four adjacent pixel values to obtain the confidence coefficient of the second image.
The four adjacent pixel values are used for carrying out linear interpolation on each corresponding pixel value in the unmarked image in the horizontal direction and the vertical direction respectively, namely, the corresponding pixel values are subjected to linear interpolation in the horizontal direction twice through the four adjacent pixel values, the interpolation obtained in the horizontal direction is substituted into the linear interpolation in the vertical direction, weights corresponding to the four adjacent pixel points can be obtained, and each pixel point in the four adjacent pixel points is multiplied by the weight and summed respectively, so that the confidence corresponding to the pixel value is obtained.
Specifically, one pixel value in the unlabeled image set is f (x, y), and the four adjacent pixel values in the category activation map are respectively f (Q) 11 )=(x 1 ,y 1 )，f(Q 12 )=(x 1 ,y 2 )，f(Q 21 )=(x 2 ,y 1 ) And f (Q) 22 )=(x 2 ,y 2 )。
In an embodiment of the present invention, a linear interpolation value in the horizontal direction can be obtained according to the pixel values:
Figure 204707DEST_PATH_IMAGE007
Figure 691052DEST_PATH_IMAGE008
a linear interpolation value in the vertical direction can be obtained from the pixel values:
Figure 933814DEST_PATH_IMAGE009
Figure 64581DEST_PATH_IMAGE010
wherein, the x 2 -x 1 =1，y 2 -y 1 =1, the pixel value f (x, y) can be expressed as:
Figure 609963DEST_PATH_IMAGE011
wherein (x) 2 -x)(y 2 -y) can be represented as f (Q) 11 ) Corresponding weights, (x-x) 1 )(y 2 -y) can be represented as f (Q) 21 ) Corresponding weight, (x) 2 -x)(y-y 1 ) Is expressed as f (Q) 12 ) Corresponding weights, (x-x) 1 )(y-y 1 ) To representIs f (Q) 22 ) The corresponding weight.
Wherein f (x, y) represents any pixel point in the unmarked image set, x represents the abscissa of the pixel point, and y represents the ordinate of the pixel point; f (Q) 11 ) Representing pixel points coinciding with the pixel positions of the top left image in the class activation map, and f (Q) 11 ) The selection of the pixel point can be gradually shifted to the right horizontally or vertically until the pixel point coincides with the pixel position of the lower right image in the category activation map, and the pixel point x 1 Abscissa, y, representing pixel point 1 Representing the vertical coordinate of the pixel point; f (Q) 12 ) Is shown in the class activation graph with f (Q) 11 ) Horizontally adjacent pixel points, x 1 Abscissa, y, representing pixel point 1 Representing the vertical coordinate of the pixel point; f (Q) 21 ) Is shown in the class activation graph with f (Q) 11 ) Vertically adjacent pixel points, x 2 Abscissa, y, representing pixel point 1 Representing the ordinate of the pixel; f (Q) 22 ) Is shown in the class activation graph with f (Q) 21 ) Horizontally adjacent pixel points, x 2 Abscissa, y, representing pixel point 2 Indicating the ordinate of the pixel.
The image pseudo tag generation module 104 is configured to fuse the first image confidence coefficient and the second image confidence coefficient to obtain a pseudo tag of the unlabeled image set.
In the embodiment of the invention, the pseudo label refers to a segmented image label corresponding to an unlabeled image set.
According to the embodiment of the invention, the first image confidence coefficient and the second image confidence coefficient are fused to obtain the pseudo label of the unmarked image set, the image pixel label with lower confidence coefficient can be screened out to remove the pseudo label with poorer accuracy, and the pseudo label corresponding to the confidence coefficient of each image pixel can be finally determined according to the fused confidence coefficient, so that the useful information in the image is avoided being lost while the low confidence coefficient pixel is screened out, the integrity of the image information is ensured, and the accuracy of the subsequent image segmentation is improved.
As an embodiment of the present invention, the image pseudo tag generating module 104 fuses the first image confidence coefficient and the second image confidence coefficient to obtain a pseudo tag of the unlabeled image set by performing the following operations, including:
acquiring a first weight coefficient of the first image confidence coefficient, and multiplying the first weight coefficient by the first image confidence coefficient to obtain a first parameter;
acquiring a second weight coefficient of the second image confidence coefficient, and multiplying the second weight coefficient by the second image confidence coefficient to obtain a second parameter;
summing the first parameter and the second parameter to obtain a fusion parameter;
judging whether the fusion parameter is smaller than a preset threshold value or not;
when the fusion parameter is smaller than a preset threshold value, screening out a label corresponding to the fusion parameter;
and when the fusion parameter is not less than a preset threshold value, taking the label corresponding to the fusion parameter as the pseudo label.
Wherein the first weight coefficient represents a degree of importance of the first image confidence; the second weight coefficient represents the degree of importance of the second image confidence, and the value of the sum of the first weight coefficient and the second weight coefficient is 1.
In an embodiment of the present invention, the preset threshold may be customized according to a specific scene, for example, in an identification scene of a vehicle picture in the field of vehicle insurance, the preset threshold may be 0.7, and if the fusion parameter exceeds 0.7, it is identified that a vehicle exists in an undivided image set; if the fusion parameter does not exceed 0.7, it indicates that no vehicle information in the set of undivided images is identified.
The target image segmentation model training module 105 is configured to integrate the real image labels and the pseudo labels to obtain final labels, use the unlabeled image set and the labeled image set as training data sets, and train a preset target image segmentation model by using the training data sets and the final labels to obtain a trained target image segmentation model.
In the embodiment of the present invention, the preset target image segmentation model may be a deep learning model CNN or RNN.
In an embodiment of the present invention, the pseudo label in the final label is a label corresponding to the unlabeled image set, and since the image pixel label with poor accuracy has been removed, the accuracy of the pseudo label, that is, the accuracy of the final label, can be ensured, and when the preset target image segmentation model is trained by using the training data set and the final label, the accuracy of the model is improved.
According to the embodiment of the invention, the real image label and the pseudo label are integrated to obtain a final label, the unmarked image set and the marked image set are used as training data sets, a preset target image segmentation model is trained by using the training data sets to obtain a trained target image segmentation model, the semi-supervised image segmentation model training can be realized, the model training efficiency is improved, the complete and accurate image information is used for training the target image segmentation model, and the accuracy of the model in image segmentation is improved.
As an embodiment of the present invention, the target image segmentation model training module 105 performs the following operations to train a preset target image segmentation model by using the training data set and the final label, so as to obtain a trained target image segmentation model, including:
extracting the characteristics of the training data set by using a cavity convolution layer in a preset target image segmentation model and carrying out characteristic fusion to obtain a characteristic fusion image set; performing target object identification on the feature fusion image set by using a pyramid pooling layer in the preset target image segmentation model to obtain a target object identification image set; inputting the target object identification image set into an activation function in the preset target image segmentation model to obtain a predicted image segmentation label of the training data set output by the activation function; calculating loss values of a predicted image segmentation label and the final label by using a loss function in the preset target image segmentation model, and adjusting parameters of the preset target image segmentation model according to the loss values until the loss values meet preset conditions to obtain the trained target image segmentation model.
Compared with the common convolutional layer, the hole convolutional layer has more special holes and wider visual field, so that when the features are extracted, the global information of more images can be mastered through feature fusion under the condition that the images are reduced by the same multiple. Further, the pyramid pooling layer may replace a general pooling layer, and a divided image having a uniform image size may be output regardless of the size of the input image. Further, in the embodiment of the present invention, the activation function may be a Sigmoid function, and the target object recognition image set may be activated to obtain a predictive image segmentation label of the training data set output by the activation function.
In the embodiment of the present invention, the preset condition may be set according to the actual model training scene, for example, the preset condition may be that the loss value is smaller than a preset threshold.
In detail, the calculating the loss values of the predicted image segmentation tag and the final tag by using the loss function in the preset target image segmentation model comprises:
calculating a loss value for a predicted image partition label and the final label using the following loss function:
Figure 911501DEST_PATH_IMAGE012
wherein L(s) represents a loss value; k represents the number of predicted image partition tags; j represents a variable of the number of predicted image segmentation tags; y is i Representing the ith prediction image partition tag;
Figure 567741DEST_PATH_IMAGE013
indicating the ith final label.
The image segmentation module 106 is configured to obtain an image to be segmented, and segment the image to be segmented by using the trained target image segmentation model to obtain an image segmentation result of the image to be segmented.
In the embodiment of the invention, the image to be segmented can be a damaged image of a vehicle with a vehicle risk, and can be acquired from a database of a client platform.
In the embodiment of the invention, the trained target image segmentation model can be used for segmenting the vehicle information in the vehicle damage image and the vehicle scratch set vehicle impact position to obtain the vehicle damage detail image.
In the embodiment of the invention, firstly, the unmarked image set is segmented by utilizing the trained image segmentation model to obtain the confidence coefficient of the first image, which is convenient for subsequently judging which image areas in the unmarked image set are close to the real label distribution of the unmarked image; secondly, performing a class activation operation on the first image confidence coefficient to determine the importance degree of each image pixel in the unmarked segmentation image, further performing upsampling on the class activation image to obtain a second image confidence coefficient, recovering the size of the class activation image to the size consistent with that of the unmarked image, further determining the importance degree of each image pixel, improving the accuracy of subsequent pseudo label screening, then obtaining the pseudo label of the unmarked image set by fusing the first image confidence coefficient and the second image confidence coefficient, screening out the image pixel label with lower confidence coefficient to remove the pseudo label with poorer accuracy, and simultaneously ensuring the integrity of image information and improving the accuracy of subsequent image segmentation; and finally, training a target image segmentation model through the real image label, the pseudo label, the unmarked image set and the marked image set, training the target image segmentation model by using complete and accurate image information, improving the accuracy of the model in image segmentation, and performing image segmentation on an image to be segmented by using the model to obtain an image segmentation result, so that the accuracy of image segmentation can be improved. Therefore, the image segmentation device provided by the embodiment of the invention can improve the accuracy of image segmentation.
Fig. 5 is a schematic structural diagram of an electronic device for implementing the image segmentation method according to the present invention.
The electronic device may comprise a processor 10, a memory 11, a communication bus 12 and a communication interface 13, and may further comprise a computer program, such as an image segmentation program, stored in the memory 11 and executable on the processor 10.
The memory 11 includes at least one type of media, which includes flash memory, removable hard disk, multimedia card, card type memory (e.g., SD or DX memory, etc.), magnetic memory, local disk, optical disk, etc. The memory 11 may in some embodiments be an internal storage unit of the electronic device, for example a removable hard disk of the electronic device. The memory 11 may also be an external storage device of the electronic device in other embodiments, such as a plug-in mobile hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), and the like, which are provided on the electronic device. Further, the memory 11 may also include both an internal storage unit and an external storage device of the electronic device. The memory 11 may be used not only to store application software installed in the electronic device and various types of data, such as codes of an image segmentation program, but also to temporarily store data that has been output or is to be output.
The processor 10 may be composed of an integrated circuit in some embodiments, for example, a single packaged integrated circuit, or may be composed of a plurality of integrated circuits packaged with the same or different functions, including one or more Central Processing Units (CPUs), microprocessors, digital Processing chips, graphics processors, and combinations of various control chips. The processor 10 is a Control Unit (Control Unit) of the electronic device, connects various components of the whole electronic device by using various interfaces and lines, and executes various functions and processes data of the electronic device by running or executing programs or modules (e.g., image segmentation programs, etc.) stored in the memory 11 and calling data stored in the memory 11.
The communication bus 12 may be a PerIPheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus may be divided into an address bus, a data bus, a control bus, etc. The communication bus 12 is arranged to enable connection communication between the memory 11 and at least one processor 10 or the like. For ease of illustration, only one thick line is shown, but this does not mean that there is only one bus or one type of bus.
Fig. 5 shows only an electronic device having components, and those skilled in the art will appreciate that the structure shown in fig. 5 does not constitute a limitation of the electronic device, and may include fewer or more components than those shown, or some components may be combined, or a different arrangement of components.
For example, although not shown, the electronic device may further include a power supply (such as a battery) for supplying power to each component, and preferably, the power supply may be logically connected to the at least one processor 10 through a power management device, so that functions of charge management, discharge management, power consumption management and the like are realized through the power management device. The power supply may also include any component of one or more dc or ac power sources, recharging devices, power failure detection circuitry, power converters or inverters, power status indicators, and the like. The electronic device may further include various sensors, a bluetooth module, a Wi-Fi module, and the like, which are not described herein again.
Optionally, the communication interface 13 may include a wired interface and/or a wireless interface (e.g., WI-FI interface, bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device and other electronic devices.
Optionally, the communication interface 13 may further include a user interface, which may be a Display (Display), an input unit (such as a Keyboard (Keyboard)), and optionally, a standard wired interface, or a wireless interface. Alternatively, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, or the like. The display, which may also be referred to as a display screen or display unit, is suitable, among other things, for displaying information processed in the electronic device and for displaying a visualized user interface.
It is to be understood that the described embodiments are for purposes of illustration only and that the scope of the appended claims is not limited to such structures.
The image segmentation program stored in the memory 11 of the electronic device is a combination of a plurality of computer programs, which when executed in the processor 10, enable:
acquiring an unlabeled image set, an labeled image set and a real image label corresponding to the labeled image set, and segmenting the unlabeled image set by using a trained image segmentation model to obtain a first image confidence coefficient;
performing category activation operation on the first image confidence coefficient to obtain a category activation graph of the unmarked image set;
the category activation image is subjected to up-sampling to obtain a second image confidence coefficient;
fusing the first image confidence coefficient and the second image confidence coefficient to obtain a pseudo label of the unmarked image set;
integrating the real image labels and the pseudo labels to obtain final labels, taking the unmarked image sets and the marked image sets as training data sets, and training a preset target image segmentation model by using the training data sets and the final labels to obtain a trained target image segmentation model;
and acquiring an image to be segmented, and segmenting the image to be segmented by using the trained target image segmentation model to obtain an image segmentation result of the image to be segmented.
Specifically, the processor 10 may refer to the description of the relevant steps in the embodiment corresponding to fig. 1 for a specific implementation method of the computer program, which is not described herein again.
Further, the electronic device integrated module/unit, if implemented in the form of a software functional unit and sold or used as a separate product, may be stored in a computer readable medium. The computer readable medium may be non-volatile or volatile. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, removable hard disk, magnetic disk, optical disk, computer Memory, Read-Only Memory (ROM).
Embodiments of the present invention may also provide a computer-readable storage medium, where the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor of an electronic device, the computer program may implement:
acquiring an unlabeled image set, an labeled image set and a real image label corresponding to the labeled image set, and segmenting the unlabeled image set by using a trained image segmentation model to obtain a first image confidence coefficient;
performing category activation operation on the first image confidence coefficient to obtain a category activation graph of the unmarked image set;
the category activation image is subjected to up-sampling to obtain a second image confidence coefficient;
fusing the first image confidence coefficient and the second image confidence coefficient to obtain a pseudo label of the unmarked image set;
integrating the real image labels and the pseudo labels to obtain final labels, taking the unmarked image sets and the marked image sets as training data sets, and training a preset target image segmentation model by using the training data sets and the final labels to obtain a trained target image segmentation model;
and acquiring an image to be segmented, and segmenting the image to be segmented by using the trained target image segmentation model to obtain an image segmentation result of the image to be segmented.
Further, the computer-readable storage medium may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function, and the like; the storage data area may store data created according to the use of the blockchain node, and the like.
In the several embodiments provided in the present invention, it should be understood that the disclosed medium, apparatus, device and method may be implemented in other manners. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the modules is only one logical functional division, and other divisions may be realized in practice.
The modules described as separate parts may or may not be physically separate, and parts displayed as modules may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment.
In addition, functional modules in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, or in a form of hardware plus a software functional module.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof.
The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference signs in the claims shall not be construed as limiting the claim concerned.
The block chain is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, a consensus mechanism, an encryption algorithm and the like. A block chain (Blockchain), which is essentially a decentralized database, is a series of data blocks associated by using a cryptographic method, and each data block contains information of a batch of network transactions, so as to verify the validity (anti-counterfeiting) of the information and generate a next block. The blockchain may include a blockchain underlying platform, a platform product service layer, an application service layer, and the like.
Furthermore, it is obvious that the word "comprising" does not exclude other elements or steps, and the singular does not exclude the plural. A plurality of units or means recited in the system claims may also be implemented by one unit or means in software or hardware. The terms second, etc. are used to denote names, but not any particular order.
Finally, it should be noted that the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting, and although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions may be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims (10)

1. A method of image segmentation, the method comprising:
acquiring an unlabeled image set, an labeled image set and a real image label corresponding to the labeled image set, and segmenting the unlabeled image set by using a trained image segmentation model to obtain a first image confidence coefficient;
performing category activation operation on the first image confidence coefficient to obtain a category activation graph of the unmarked image set;
the category activation image is subjected to up-sampling to obtain a second image confidence coefficient;
fusing the first image confidence coefficient and the second image confidence coefficient to obtain a pseudo label of the unmarked image set;
integrating the real image labels and the pseudo labels to obtain final labels, taking the unmarked image sets and the marked image sets as training data sets, and training a preset target image segmentation model by using the training data sets and the final labels to obtain a trained target image segmentation model;
and acquiring an image to be segmented, and segmenting the image to be segmented by using the trained target image segmentation model to obtain an image segmentation result of the image to be segmented.
2. The image segmentation method according to claim 1, wherein performing a class activation operation on the first image confidence level to obtain a class activation map of the unlabeled image set includes:
acquiring a convolution characteristic diagram of the unmarked image set, and setting the activation value of the convolution characteristic diagram to zero to obtain the activation image confidence of the unmarked image set;
calculating a difference set of the first image confidence coefficient and the activation image confidence coefficient, screening out a positive difference value larger than zero from the difference set, and dividing the positive difference value by the number of pixel points corresponding to the category activation confidence coefficient to obtain a fusion weight;
and weighting the convolution characteristic diagram by using the fusion weight, and inputting the weighted characteristic diagram into a preset activation function to obtain the category activation diagram.
3. The image segmentation method according to claim 1, wherein the fusing the first image confidence level and the second image confidence level to obtain the pseudo label of the unlabeled image set comprises:
acquiring a first weight coefficient of the first image confidence coefficient, and multiplying the first weight coefficient by the first image confidence coefficient to obtain a first parameter;
acquiring a second weight coefficient of the second image confidence coefficient, and multiplying the second weight coefficient by the second image confidence coefficient to obtain a second parameter;
summing the first parameter and the second parameter to obtain a fusion parameter;
judging whether the fusion parameter is smaller than a preset threshold value or not;
when the fusion parameter is smaller than a preset threshold value, screening out a label corresponding to the fusion parameter;
and when the fusion parameter is not less than a preset threshold value, taking the label corresponding to the fusion parameter as the pseudo label.
4. The image segmentation method of claim 1, wherein the upsampling the class activation map to obtain a second image confidence level comprises:
and sequentially selecting four adjacent pixel values of the category activation graph, and performing linear interpolation on each corresponding pixel value in the unmarked image in the horizontal direction and the vertical direction by using the four adjacent pixel values to obtain the confidence coefficient of the second image.
5. The image segmentation method according to claim 1, wherein the training a preset target image segmentation model by using the training data set and the final label to obtain a trained target image segmentation model comprises:
extracting the characteristics of the training data set by using a cavity convolution layer in a preset target image segmentation model and carrying out characteristic fusion to obtain a characteristic fusion image set;
performing target object identification on the feature fusion image set by using a pyramid pooling layer in the preset target image segmentation model to obtain a target object identification image set;
inputting the target object identification image set into an activation function in the preset target image segmentation model to obtain a predicted image segmentation label of the training data set output by the activation function;
calculating loss values of a predicted image segmentation label and the final label by using a loss function in the preset target image segmentation model, and adjusting parameters of the preset target image segmentation model according to the loss values until the loss values meet preset conditions to obtain the trained target image segmentation model.
6. The image segmentation method of claim 1, wherein the segmenting the set of unlabeled images using the trained image segmentation model to obtain a first image confidence level comprises:
performing convolution operation on the unmarked image set by using the trained image segmentation model to obtain an image characteristic data set;
performing pooling operation on the image characteristic data set to obtain a pooled data set;
and performing activation operation on the pooled data set by using an activation function to obtain a first image confidence corresponding to the pooled data in the pooled data set.
7. An image segmentation apparatus, characterized in that the apparatus comprises:
the first image segmentation module is used for acquiring an unlabeled image set, an labeled image set and a real image label corresponding to the labeled image set, and segmenting the unlabeled image set by using a trained image segmentation model to obtain a first image confidence coefficient;
the image category activation module is used for carrying out category activation operation on the first image confidence coefficient to obtain a category activation graph of the unmarked image set;
the image up-sampling module is used for up-sampling the category activation map to obtain a second image confidence coefficient;
the image pseudo label generating module is used for fusing the first image confidence coefficient and the second image confidence coefficient to obtain a pseudo label of the unmarked image set;
the target image segmentation model training module is used for integrating the real image labels and the pseudo labels to obtain final labels, taking the unmarked image sets and the marked image sets as training data sets, and training a preset target image segmentation model by using the training data sets and the final labels to obtain a trained target image segmentation model;
and the image segmentation module is used for acquiring an image to be segmented, and segmenting the image to be segmented by using the trained target image segmentation model to obtain an image segmentation result of the image to be segmented.
8. The image segmentation apparatus according to claim 7, wherein the fusing the first image confidence level and the second image confidence level to obtain the pseudo label of the unlabeled image set comprises:
acquiring a first weight coefficient of the first image confidence coefficient, and multiplying the first weight coefficient by the first image confidence coefficient to obtain a first parameter;
acquiring a second weight coefficient of the second image confidence coefficient, and multiplying the second weight coefficient by the second image confidence coefficient to obtain a second parameter;
summing the first parameter and the second parameter to obtain a fusion parameter;
judging whether the fusion parameter is smaller than a preset threshold value or not;
when the fusion parameter is smaller than a preset threshold value, screening out a label corresponding to the fusion parameter;
and when the fusion parameter is not less than a preset threshold value, taking the label corresponding to the fusion parameter as the pseudo label.
9. An electronic device, characterized in that the electronic device comprises:
at least one processor; and the number of the first and second groups,
a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,
the memory stores a computer program executable by the at least one processor to enable the at least one processor to perform the image segmentation method according to any one of claims 1 to 6.
10. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the image segmentation method according to any one of claims 1 to 6.
CN202210978696.XA 2022-08-16 2022-08-16 Image segmentation method, device, equipment and storage medium Active CN115049836B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210978696.XA CN115049836B (en) 2022-08-16 2022-08-16 Image segmentation method, device, equipment and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210978696.XA CN115049836B (en) 2022-08-16 2022-08-16 Image segmentation method, device, equipment and storage medium

Publications (2)

Publication Number Publication Date
CN115049836A true CN115049836A (en) 2022-09-13
CN115049836B CN115049836B (en) 2022-10-25

Family

ID=83168095

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210978696.XA Active CN115049836B (en) 2022-08-16 2022-08-16 Image segmentation method, device, equipment and storage medium

Country Status (1)

Country Link
CN (1) CN115049836B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115501050A (en) * 2022-10-31 2022-12-23 江苏理工学院 Wheelchair user health monitoring system based on computer vision and training method of detection network thereof

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110930417A (en) * 2019-11-26 2020-03-27 腾讯科技（深圳）有限公司 Training method and device of image segmentation model, and image segmentation method and device
CN112884770A (en) * 2021-04-28 2021-06-01 腾讯科技（深圳）有限公司 Image segmentation processing method and device and computer equipment
CN112990218A (en) * 2021-03-25 2021-06-18 北京百度网讯科技有限公司 Optimization method and device of image semantic segmentation model and electronic equipment
CN113096138A (en) * 2021-04-13 2021-07-09 西安电子科技大学 Weak supervision semantic image segmentation method for selective pixel affinity learning
CN113435522A (en) * 2021-06-30 2021-09-24 平安科技（深圳）有限公司 Image classification method, device, equipment and storage medium
CN114821063A (en) * 2022-05-11 2022-07-29 北京百度网讯科技有限公司 Semantic segmentation model generation method and device and image processing method
WO2022160449A1 (en) * 2021-01-28 2022-08-04 平安科技（深圳）有限公司 Text classification method and apparatus, electronic device, and storage medium

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110930417A (en) * 2019-11-26 2020-03-27 腾讯科技（深圳）有限公司 Training method and device of image segmentation model, and image segmentation method and device
WO2022160449A1 (en) * 2021-01-28 2022-08-04 平安科技（深圳）有限公司 Text classification method and apparatus, electronic device, and storage medium
CN112990218A (en) * 2021-03-25 2021-06-18 北京百度网讯科技有限公司 Optimization method and device of image semantic segmentation model and electronic equipment
CN113096138A (en) * 2021-04-13 2021-07-09 西安电子科技大学 Weak supervision semantic image segmentation method for selective pixel affinity learning
CN112884770A (en) * 2021-04-28 2021-06-01 腾讯科技（深圳）有限公司 Image segmentation processing method and device and computer equipment
CN113435522A (en) * 2021-06-30 2021-09-24 平安科技（深圳）有限公司 Image classification method, device, equipment and storage medium
CN114821063A (en) * 2022-05-11 2022-07-29 北京百度网讯科技有限公司 Semantic segmentation model generation method and device and image processing method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
黄伟等: "基于多分类器投票集成的半监督情感分类方法研究", 《中文信息学报》 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115501050A (en) * 2022-10-31 2022-12-23 江苏理工学院 Wheelchair user health monitoring system based on computer vision and training method of detection network thereof

Also Published As

Publication number Publication date
CN115049836B (en) 2022-10-25

Similar Documents

Publication Publication Date Title
CN112257774A (en) Target detection method, device, equipment and storage medium based on federal learning
CN111783982B (en) Method, device, equipment and medium for acquiring attack sample
CN111814962A (en) Method and device for acquiring parameters of recognition model, electronic equipment and storage medium
CN112699775A (en) Certificate identification method, device and equipment based on deep learning and storage medium
CN112137591B (en) Target object position detection method, device, equipment and medium based on video stream
CN112132216B (en) Vehicle type recognition method and device, electronic equipment and storage medium
CN112749653A (en) Pedestrian detection method, device, electronic equipment and storage medium
CN111639704A (en) Target identification method, device and computer readable storage medium
CN112541902A (en) Similar area searching method, similar area searching device, electronic equipment and medium
CN115049836B (en) Image segmentation method, device, equipment and storage medium
CN114494800B (en) Predictive model training method and device, electronic equipment and storage medium
CN115471775A (en) Information verification method, device and equipment based on screen recording video and storage medium
CN111985449A (en) Rescue scene image identification method, device, equipment and computer medium
CN112101481B (en) Method, device, equipment and storage medium for screening influence factors of target object
CN112269875A (en) Text classification method and device, electronic equipment and storage medium
CN113706019B (en) Service capability analysis method, device, equipment and medium based on multidimensional data
CN112215336B (en) Data labeling method, device, equipment and storage medium based on user behaviors
CN114463685A (en) Behavior recognition method and device, electronic equipment and storage medium
CN113343882A (en) Crowd counting method and device, electronic equipment and storage medium
CN112561889A (en) Target detection method and device, electronic equipment and storage medium
CN114359645B (en) Image expansion method, device, equipment and storage medium based on characteristic area
CN111652226B (en) Picture-based target identification method and device and readable storage medium
CN114240935B (en) Space-frequency domain feature fusion medical image feature identification method and device
CN114742828B (en) Intelligent analysis method and device for workpiece damage assessment based on machine vision
CN112101279B (en) Target object abnormality detection method, target object abnormality detection device, electronic equipment and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant