CN111028253A - Iron concentrate powder segmentation method and segmentation device - Google Patents
Iron concentrate powder segmentation method and segmentation device Download PDFInfo
- Publication number
- CN111028253A CN111028253A CN201911164824.1A CN201911164824A CN111028253A CN 111028253 A CN111028253 A CN 111028253A CN 201911164824 A CN201911164824 A CN 201911164824A CN 111028253 A CN111028253 A CN 111028253A
- Authority
- CN
- China
- Prior art keywords
- attention
- convolution
- module
- segmentation
- iron powder
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
-
- 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
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/20084—Artificial neural networks [ANN]
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a method and a device for segmenting iron fine powder, which can completely and accurately extract the iron fine powder from an iron fine powder image. The method comprises the following steps: acquiring a fine iron powder image and a corresponding label graph in a transmission process as a training sample set; establishing a split network SegNet based on an attention mechanism module, wherein the attention mechanism comprises: spatial attention and channel attention; training a segmentation network SegNet by using the obtained training sample set; and (4) segmenting the iron fine powder image to be segmented by utilizing the trained segmentation network SegNet, and outputting the category of each pixel point of the iron fine powder image. The invention relates to the technical field of ore screening.
Description
Technical Field
The invention relates to the technical field of ore screening, in particular to a method and a device for dividing fine iron powder.
Background
Fine iron powder is a raw material for producing pellets, and the iron content of the fine iron powder determines the quality of the fine iron powder. In China, mineral powder is rich in resources and large in reserves, but a lot of ores with small particles and complex varieties are mixed in the mineral powder, so that the mineral powder resources in China are rich in yield, but unsatisfactory in quality, and the balling property and the stability of components are influenced. Therefore, most of mineral powder adopted by many domestic enterprises is imported, but the quantity of imported ore is more and more along with the expansion of the demand. Due to the limited iron ore resources, the supply of raw materials is insufficient, and the domestic market has to be switched to. Meanwhile, the steel industry in China develops towards high quality and finish machining, and the quality of the iron concentrate powder used as the raw material must adapt to the change.
The traditional quality and taste analysis technology of the fine iron powder generally uses a chemical or physical method, and the time and labor cost are higher. The deep learning is a widely applied and novel machine learning method and is an efficient algorithm, the classification is obtained by taking the features extracted from different levels as the basis of identification, the defects of time consumption, labor consumption and the like of the traditional fine iron powder can be effectively overcome, and meanwhile, the high accuracy rate and the maintenance cost can be well guaranteed to be achieved at low cost.
Because the shape, the color, the texture and the volume of the iron concentrate powder and other ores mixed in the iron concentrate powder have many similarities, some iron concentrate powder is difficult to distinguish by naked eyes; in addition, some fine iron powder and ore have very small volumes, and the area of the fine iron powder and ore presented in the picture occupies extremely few pixels, which belongs to a small target. In addition, small targets contain less information, so that the contained information is easy to ignore in the process of pooling and convolution, and meanwhile, the existing semantic segmentation method is relatively difficult to extract edge features (due to inaccurate edge positioning) and small target information, so that the classification accuracy of edge pixels and small-area pixels is poor, and the classification accuracy of pixels in a large range is not favorable for completely extracting the target object of fine iron powder.
Disclosure of Invention
The invention aims to provide a method and a device for dividing powdered iron so as to solve the problems that powdered iron cannot be completely and accurately extracted due to small characteristic difference between the powdered iron and other ores mixed in the powdered iron, small target volume or inaccurate edge positioning in the prior art.
To solve the above technical problem, an embodiment of the present invention provides a method for dividing fine iron powder, including:
acquiring a fine iron powder image and a corresponding label graph in a transmission process as a training sample set;
establishing a split network SegNet based on an attention mechanism module, wherein the attention mechanism comprises: spatial attention and channel attention;
training a segmentation network SegNet by using the obtained training sample set;
and (4) segmenting the iron fine powder image to be segmented by utilizing the trained segmentation network SegNet, and outputting the category of each pixel point of the iron fine powder image.
Further, before training the segmentation network SegNet by using the acquired training sample set, the method further includes:
carrying out consistent segmentation processing on the obtained fine iron powder image and the corresponding label graph;
and carrying out image enhancement treatment on the iron fine powder image subjected to the segmentation treatment and the corresponding annotation graph.
Further, the image enhancement processing includes: one or more of rotation, flipping, zooming in, zooming out, translation.
Further, the segmentation network SegNet includes: an encoder and an attention mechanism module-based decoder;
the encoder is used for performing convolution, batch normalization and pooling on the input image and extracting a characteristic diagram;
the decoder is used for extracting space attention features from the feature diagram output by the encoder, up-sampling, extracting channel attention features, splicing, convolving and carrying out batch normalization processing on the extracted space attention features and the channel attention features, and outputting the category of each pixel point of the fine iron powder image.
Further, the decoder includes: the system comprises a first attention mechanism module, a first convolution normalization module connected with the first attention mechanism module, a second attention mechanism module connected with the first convolution normalization module, a second convolution normalization module connected with the second attention mechanism module, a third attention mechanism module connected with the second convolution normalization module, a third convolution normalization module connected with the third attention mechanism module, a fourth attention mechanism module connected with the third convolution normalization module, a fourth convolution normalization module connected with the fourth attention mechanism module and a classifier module connected with the fourth convolution normalization module;
wherein each attention mechanism module comprises: the system comprises a space attention unit, an upper sampling layer connected with the space attention unit, a channel attention unit connected with the upper sampling layer and a splicing unit connected with the channel attention unit;
each convolution normalization module includes: a number of network structures connected by convolutional layers and a bulk normalization layer, wherein each convolutional layer is followed by a linear rectification function ReLU.
Further, the spatial attention unit is configured to perform convolution and deconvolution operations on received feature maps output by adjacent batch normalization layers, add the deconvolution result and the feature maps output by the batch normalization layers with the same number of features and corresponding to the deconvolution result in the encoder after performing convolution operations, perform multiplication on the upsampled result and the feature maps output by the batch normalization layers with the same number of features and corresponding to the upsampled result in the encoder after performing linear rectification function ReLU, convolution, sigmoid and upsampling operations, perform convolution and batch normalization operations on the multiplication results, and extract spatial attention features, where sigmoid represents a hyperbolic tangent S-type function.
Further, the channel attention unit is used for performing convolution operation on the output result of the upper sampling layer, performing compression operation on the convolution result by using the global pooling layer, restoring the characteristic dimension by using the correlation between the two fully-connected layer modeling channels and adjusting the dimension and the shape of the matrix according to the compression result, multiplying the characteristic diagram of the restored dimension and the convolution result, and extracting the channel attention feature.
Further, the training of the segmentation network SegNet by using the acquired training sample set includes:
and training the segmentation network SegNet by using the obtained training sample set, and optimizing the segmentation network SegNet by taking a loss function of the minimized segmentation network SegNet as an objective function.
Further, the loss function of the segmentation network SegNet is:
wherein L isflRepresenting the loss value, α representing the balance factor, gamma representing the adjustment factor, y' representing the predicted output value, y representing the label of the training sample.
The embodiment of the present invention further provides a device for dividing refined iron powder, including:
the acquisition module is used for acquiring the fine iron powder image and the corresponding annotation graph in the transmission process as a training sample set;
the establishing module is used for establishing the segmentation network SegNet based on the attention mechanism module, wherein the attention mechanism comprises: spatial attention and channel attention;
the training module is used for training the segmentation network SegNet by utilizing the obtained training sample set;
and the segmentation module is used for segmenting the iron fine powder image to be segmented by utilizing the trained segmentation network SegNet and outputting the category of each pixel point of the iron fine powder image.
The technical scheme of the invention has the following beneficial effects:
in the scheme, in order to obtain the characteristics contained in the small target and more characteristic information which is different from other ores, the channel attention is added on the basis of the original segmentation network SegNet, and in order to obtain more edge information and better position a target object, the space attention is added on the basis of the original segmentation network SegNet, so that the trained segmentation network SegNet based on the space attention and the channel attention is utilized to completely and accurately extract the iron fine powder from the obtained iron fine powder image to be segmented in the conveying process, thereby improving the extraction precision and the working efficiency of the iron fine powder and achieving the purpose of generating high-quality pellet ore raw materials.
Drawings
Fig. 1 is a schematic flow chart of a fine iron powder segmentation method according to an embodiment of the present invention;
fig. 2 is a schematic structural diagram of a segmentation network SegNet based on an attention mechanism module according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a workflow of a spatial attention unit according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a workflow of a channel attention unit according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of a fine iron powder dividing device according to an embodiment of the present invention.
Detailed Description
In order to make the technical problems, technical solutions and advantages of the present invention more apparent, the following detailed description is given with reference to the accompanying drawings and specific embodiments.
The invention provides a method and a device for dividing iron concentrate powder, aiming at the problem that the iron concentrate powder cannot be completely and accurately extracted due to small characteristic difference, small target volume or inaccurate edge positioning between the conventional iron concentrate powder and other ores mixed in the iron concentrate powder.
Example one
As shown in fig. 1, the method for dividing refined iron powder according to the embodiment of the present invention includes:
s101, acquiring a fine iron powder image and a corresponding annotation graph in a transmission process as a training sample set;
s102, establishing a segmentation network SegNet based on an attention mechanism module, wherein the attention mechanism module comprises: spatial attention and channel attention;
s103, training a segmentation network SegNet by using the obtained training sample set;
and S104, segmenting the iron fine powder image to be segmented by utilizing the trained segmentation network SegNet, and outputting the category of each pixel point of the iron fine powder image.
According to the fine iron powder segmentation method disclosed by the embodiment of the invention, in order to obtain characteristics contained in a small target and more characteristic information different from other ores, the channel attention is increased on the basis of the original segmentation network SegNet, and in order to obtain more edge information and better position a target object, the space attention is increased on the basis of the original segmentation network SegNet, so that the fine iron powder can be completely and accurately extracted from an obtained fine iron powder image to be segmented in the conveying process by utilizing the trained segmentation network SegNet based on the space attention and the channel attention, the fine iron powder extraction precision and the working efficiency are improved, and the purpose of generating a high-quality pellet ore raw material is achieved.
In this embodiment, SegNet is a deep convolutional encoder-decoder architecture for image segmentation.
In an embodiment of the foregoing fine iron powder segmentation method, further before training the segmentation network SegNet by using the acquired training sample set, the method further includes:
carrying out consistent segmentation processing on the obtained fine iron powder image and the corresponding label graph;
and carrying out image enhancement treatment on the iron fine powder image subjected to the segmentation treatment and the corresponding annotation graph.
In the embodiment, a high-definition industrial camera can be used for obtaining a fine iron powder image in a transmission process, then a part of the fine iron powder image is selected, manual marking is carried out by a domain expert to obtain a marking image of the selected fine iron powder image, and the selected fine iron powder image and the corresponding marking image are used as a training sample set; and taking other fine iron powder images as a test sample set, wherein the test sample set is used for testing the accuracy of the SegNet segmentation of the trained segmentation network.
In this embodiment, it is assumed that the size of the obtained original fine iron powder image is 2592 × 1944, the size of the image after image segmentation is 256 × 256, and the corresponding label graph of each training sample image is also uniformly segmented.
In this embodiment, the image enhancement processing includes: one or more of rotation, flipping, zooming in, zooming out, and translation; for example, randomly rotating the image by a certain angle, changing the orientation of the image content; flipping the image in either the horizontal or vertical direction; enlarging or reducing the image according to a certain proportion; the image is translated in the image plane.
In an embodiment of the fine iron powder dividing method, the dividing network SegNet further includes: an encoder (encoder) and an attention mechanism module-based decoder (encoder);
the encoder is used for performing convolution, batch normalization and pooling on the input image and extracting a characteristic diagram;
the decoder is used for extracting space attention features from the feature diagram output by the encoder, up-sampling, extracting channel attention features, splicing, convolving and carrying out batch normalization processing on the extracted space attention features and the channel attention features, and outputting the category of each pixel point of the fine iron powder image.
In this embodiment, the encoder performs feature extraction, and the decoder is configured to analyze the extracted features.
In this embodiment, as shown in fig. 2, the encoder is briefly explained:
1) the encoder comprises: 5 convolution normalization modules (1, 2, 3, 4, 5) and 4 maximum pooling layers max pool (1, 2, 3, 4), each convolution normalization module comprising: a plurality of network structures connected by convolutional layers (conv) and batch normalization layers (BN), each convolutional layer being followed by a Linear rectifying function (ReLU), wherein ReLU is an activation function;
2) the convolution normalization modules 1 and 2 respectively comprise 2 network structures connected by convolution layers and batch normalization layers; the convolution normalization modules 3, 4 and 5 respectively comprise 3 network structures connected by convolution layers and batch normalization layers; thus, the encoder has a total of 13 convolutional layers;
3) the convolution normalization module 1 is connected with the maximum pooling layer 1, the maximum pooling layer 1 is connected with the convolution normalization module 2, the convolution normalization module 2 is connected with the maximum pooling layer 2, the maximum pooling layer 2 is connected with the convolution normalization module 3, the convolution normalization module 3 is connected with the maximum pooling layer 3, the maximum pooling layer 3 is connected with the convolution normalization module 4, the convolution normalization module 4 is connected with the maximum pooling layer 4, and the maximum pooling layer 4 is connected with the convolution normalization module 5;
4) the convolutional layer is used for extracting features of the images, the batch normalization layer is used for relieving the problem of gradient dispersion in a deep network, and the pooling layer can gradually reduce network training parameters and avoid overfitting of the segmentation network SegNet. Due to the convolution and pooling operations in the encoder, the number of feature channels is gradually increased and the size of the feature map is gradually decreased in order to extract more accurate features.
In this embodiment, as shown in fig. 2, the decoder is briefly described:
1) the decoder includes: the system comprises a first attention mechanism module, a first convolution normalization module connected with the first attention mechanism module, a second attention mechanism module connected with the first convolution normalization module, a second convolution normalization module connected with the second attention mechanism module, a third attention mechanism module connected with the second convolution normalization module, a third convolution normalization module connected with the third attention mechanism module, a fourth attention mechanism module connected with the third convolution normalization module, a fourth convolution normalization module connected with the fourth attention mechanism module, and a classifier module (conv + softmax) connected with the fourth convolution normalization module; that is, the decoder includes 4 attention mechanism modules (first, second, third, fourth), 4 convolution normalization modules (first, second, third, fourth), and 1 classifier module, wherein each convolution normalization module includes: a plurality of network structures connected by convolutional layers and batch normalization layers, each convolutional layer being followed by a linear rectification function ReLU;
2) each attention mechanism module includes: a Spatial Attention (Spatial Attention) unit, an upsampling layer (upsampling) connected to the Spatial Attention unit, a Channel-wise Attention (Channel-wise Attention) unit connected to the upsampling layer, a concatenating unit (concatenate) connected to the Channel Attention unit, and thus, the decoder uses 4 upsampling layers;
in this embodiment, Spatial attribute may increase the extraction of Spatial features, and Channel-wise attribute may increase the extraction of Channel features. Adding a Spatial Attention unit in front of each upper sampling layer, adding a Channel-wise Attention unit behind each upper sampling layer, splicing the features extracted by the space Attention unit and the Channel Attention unit through a splicing unit, and transmitting the spliced features to a subsequent convolutional layer; therefore, the network can obtain more context information, and the segmentation result is more accurate. It should be noted that the size and the number of channels of the two feature maps to be spliced need to be consistent;
3) the first convolution normalization module, the second convolution normalization module and the third convolution normalization module respectively comprise 3 network structures connected by convolution layers and batch normalization layers; the fourth convolution normalization module comprises 2 network structures connected by convolution layers and batch normalization layers; thus, the encoder has a total of 11 convolutional layers;
4) the characteristic size of setting is up-sampled to the successive layer of upsampling layer, and in decoder, after each up-sampling, characteristic channel quantity all can halve, and the size of characteristic map all can be doubled, can resume the image of original resolution ratio size like this.
5) And the classifier module consists of conv and softmax and is used for processing the output of the fourth convolution normalization module, generating nonlinear transformation and outputting the category of each pixel point of the fine iron powder image.
In the embodiment of the fine iron powder division method, further, as shown in fig. 3, the spatial attention unit, the method is used for performing convolution and inverse convolution (ConvTranspose) operation on a received feature graph (y) output by an adjacent batch normalization layer, performing convolution operation on an inverse convolution result and a feature graph (z) output by the batch normalization layer with the same number of features and corresponding to the inverse convolution result in an encoder, then adding (add), extracting more features from the addition result through a linear rectification function ReLU, convolution, a hyperbolic tangent S-shaped function (sigmoid) and an upsampling operation, multiplying (Multiply) the upsampling result with a feature map (z) output by a batch normalization layer with the same number of features corresponding to the upsampling result in an encoder, namely adding a weight to each feature, so that the local important information of the picture can be extracted through transformation; and then, performing convolution and batch normalization operation on the multiplication result to extract spatial attention features.
In a specific embodiment of the foregoing fine iron powder segmentation method, as shown in fig. 4, the channel attention unit is configured to perform a convolution operation on an output result (y) of the upsampling layer, perform a compression operation on a convolution result by using the global pooling layer, model a correlation between channels by using the full connection layer (dense)1 and the full connection layer 2 according to the compression result, and restore the characteristic dimension by adjusting the dimension and the shape (Reshape) of the matrix (i.e., the characteristic dimension is reduced to 1/16 that is input through the full connection layer 1, and then is increased back to the original dimension through the full connection layer 2 and Reshape operations after ReLU activation), output weights having the same number as that of the input features, multiply the feature diagram of the restored dimension and the convolution result, and weight the weights to the features of each channel, thereby extracting the channel attention feature. This allows more non-linearity of the extracted channel attention characteristics, which results in a better fit to the complex correlations between channels, while greatly reducing the amount of parameters and computations.
In an embodiment of the foregoing fine iron powder segmentation method, the training the segmentation network SegNet using the obtained training sample set further includes:
and training the segmentation network SegNet by using the obtained training sample set, and optimizing the segmentation network SegNet by taking a loss function of the minimized segmentation network SegNet as an objective function.
In this embodiment, the loss function of the segmentation network SegNet is:
wherein y represents a fine iron powder identifier.
Wherein L isflRepresenting a loss value, α representing a balance factor for balancing the number scale unevenness of the positive and negative samples themselves, and gamma representing an adjustment factor to make the segmentation network more focused on difficult, misclassified samples(ii) a y 'represents the predicted output value of the segmentation network SegNet, and the value of y' is between 0 and 1; y represents the label of the training sample, 1 represents the fine iron powder sample, and 0 represents the background sample.
In this embodiment, the positive sample refers to a sample belonging to a certain category, the negative sample refers to a sample not belonging to a certain category, and at this time, the positive sample is a fine iron powder sample, and the negative sample is a background sample.
In this embodiment, α and γ are both adjustable hyper-parameters, y ' is the prediction result of the segmentation network SegNet, and has a value between (0-1), when y is 1, y ' tends to 1, which represents an easily recognizable positive sample, and its contribution to the weight tends to 0, and when y is 0, y ' tends to 0, which represents an easily recognizable negative sample, and its contribution to the weight tends to 0flThe problem of class imbalance can be solved, and not only the weight of the background class is reduced, but also the weight of the easily recognized positive/negative sample (easy positive/negative) is reduced.
In this embodiment, the segmentation network SegNet may be optimized by using an optimizer random gradient (SGD), where the learning rate is 0.1, the value of the learning rate attenuation value (decay) is 1e to 4, the value of the momentum parameter (momentum) is 0.9, and momentum indicates how far the original update direction is to be kept, and the value is between 0 and 1.
In this embodiment, the fine iron powder image to be segmented is segmented and image-enhanced, then the trained model is used to segment the fine iron powder image to be segmented after image enhancement, and the class of each pixel point is output, where the class is represented by 0 and 1, 0 represents a background, and 1 represents fine iron powder.
Example two
The present invention also provides a specific embodiment of a fine iron powder dividing apparatus, which corresponds to the specific embodiment of the fine iron powder dividing method, and the fine iron powder dividing apparatus can achieve the object of the present invention by executing the flow steps in the specific embodiment of the method, so the explanation in the specific embodiment of the fine iron powder dividing method is also applicable to the specific embodiment of the fine iron powder dividing apparatus provided by the present invention, and will not be described in detail in the following specific embodiment of the present invention.
As shown in fig. 5, an embodiment of the present invention further provides a fine iron powder dividing device, including:
the acquisition module 11 is used for acquiring the fine iron powder image and the corresponding annotation graph in the transmission process as a training sample set;
an establishing module 12, configured to establish a split network SegNet based on an attention mechanism module, where the attention mechanism includes: spatial attention and channel attention;
the training module 13 is configured to train the segmentation network SegNet by using the obtained training sample set;
and the segmentation module 14 is configured to segment the fine iron powder image to be segmented by using the trained segmentation network SegNet, and output a category of each pixel point of the fine iron powder image.
According to the fine iron powder dividing device disclosed by the embodiment of the invention, in order to acquire characteristics contained in a small target and more characteristic information different from other ores, the channel attention is increased on the basis of the original dividing network SegNet, and in order to acquire more edge information and better position a target object, the space attention is increased on the basis of the original dividing network SegNet, so that the fine iron powder can be completely and accurately extracted from the obtained fine iron powder image to be divided in the conveying process by utilizing the trained dividing network SegNet based on the space attention and the channel attention, the fine iron powder extraction precision and the working efficiency are improved, and the purpose of generating high-quality pellet ore raw materials is achieved.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions.
While the foregoing is directed to the preferred embodiment of the present invention, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (10)
1. A method for dividing fine iron powder is characterized by comprising the following steps:
acquiring a fine iron powder image and a corresponding label graph in a transmission process as a training sample set;
establishing a split network SegNet based on an attention mechanism module, wherein the attention mechanism comprises: spatial attention and channel attention;
training a segmentation network SegNet by using the obtained training sample set;
and (4) segmenting the iron fine powder image to be segmented by utilizing the trained segmentation network SegNet, and outputting the category of each pixel point of the iron fine powder image.
2. The fine iron powder segmentation method according to claim 1, wherein before training the segmentation network SegNet using the acquired training sample set, the method further comprises:
carrying out consistent segmentation processing on the obtained fine iron powder image and the corresponding label graph;
and carrying out image enhancement treatment on the iron fine powder image subjected to the segmentation treatment and the corresponding annotation graph.
3. The method of claim 2, wherein the image enhancement process comprises: one or more of rotation, flipping, zooming in, zooming out, translation.
4. The fine iron powder partitioning method according to claim 1, wherein the partitioning network SegNet comprises: an encoder and an attention mechanism module-based decoder;
the encoder is used for performing convolution, batch normalization and pooling on the input image and extracting a characteristic diagram;
the decoder is used for extracting space attention features from the feature diagram output by the encoder, up-sampling, extracting channel attention features, splicing, convolving and carrying out batch normalization processing on the extracted space attention features and the channel attention features, and outputting the category of each pixel point of the fine iron powder image.
5. The fine iron powder division method according to claim 4, wherein the decoder comprises: the system comprises a first attention mechanism module, a first convolution normalization module connected with the first attention mechanism module, a second attention mechanism module connected with the first convolution normalization module, a second convolution normalization module connected with the second attention mechanism module, a third attention mechanism module connected with the second convolution normalization module, a third convolution normalization module connected with the third attention mechanism module, a fourth attention mechanism module connected with the third convolution normalization module, a fourth convolution normalization module connected with the fourth attention mechanism module and a classifier module connected with the fourth convolution normalization module;
wherein each attention mechanism module comprises: the system comprises a space attention unit, an upper sampling layer connected with the space attention unit, a channel attention unit connected with the upper sampling layer and a splicing unit connected with the channel attention unit;
each convolution normalization module includes: a number of network structures connected by convolutional layers and a bulk normalization layer, wherein each convolutional layer is followed by a linear rectification function ReLU.
6. The fine iron powder segmentation method according to claim 5, wherein the spatial attention unit is configured to perform convolution and deconvolution operations on received feature maps output by adjacent batch normalization layers, add a deconvolution result and a feature map output by a batch normalization layer with the same number of features corresponding to the deconvolution result in an encoder after performing convolution operations, multiply an upsampled result and a feature map output by a batch normalization layer with the same number of features corresponding to the upsampled result in an encoder after performing a linear rectification function ReLU, convolution, sigmoid and upsampling operations, and perform convolution and batch normalization operations on the multiplication results to extract spatial attention features, where sigmoid represents a hyperbolic tangent S-type function.
7. The fine iron powder segmentation method according to claim 5, wherein the channel attention unit is configured to perform convolution operation on an output result of the upsampling layer, perform compression operation on a convolution result by using a global pooling layer, restore a feature dimension by using correlation between two fully-connected layer modeling channels and by adjusting a dimension and a shape of a matrix according to the compression result, and multiply a feature map of the restored dimension and the convolution result to extract a channel attention feature.
8. The method of claim 1, wherein training the segmentation network SegNet using the obtained training sample set comprises:
and training the segmentation network SegNet by using the obtained training sample set, and optimizing the segmentation network SegNet by taking a loss function of the minimized segmentation network SegNet as an objective function.
9. The fine iron powder partitioning method according to claim 8, wherein a loss function of the partitioning network SegNet is:
wherein L isflRepresenting the loss value, α representing the balance factor, gamma representing the adjustment factor, y' representing the predicted output value, y representing the label of the training sample.
10. A fine iron powder dividing device is characterized by comprising:
the acquisition module is used for acquiring the fine iron powder image and the corresponding annotation graph in the transmission process as a training sample set;
the establishing module is used for establishing the segmentation network SegNet based on the attention mechanism module, wherein the attention mechanism comprises: spatial attention and channel attention;
the training module is used for training the segmentation network SegNet by utilizing the obtained training sample set;
and the segmentation module is used for segmenting the iron fine powder image to be segmented by utilizing the trained segmentation network SegNet and outputting the category of each pixel point of the iron fine powder image.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911164824.1A CN111028253B (en) | 2019-11-25 | 2019-11-25 | Method and device for dividing fine iron powder |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911164824.1A CN111028253B (en) | 2019-11-25 | 2019-11-25 | Method and device for dividing fine iron powder |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111028253A true CN111028253A (en) | 2020-04-17 |
CN111028253B CN111028253B (en) | 2023-05-30 |
Family
ID=70206469
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911164824.1A Active CN111028253B (en) | 2019-11-25 | 2019-11-25 | Method and device for dividing fine iron powder |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111028253B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112950586A (en) * | 2021-03-02 | 2021-06-11 | 攀钢集团攀枝花钢铁研究院有限公司 | LF furnace steel slag infrared identification method and system |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109325911A (en) * | 2018-08-27 | 2019-02-12 | 北京航空航天大学 | A kind of space base rail detection method based on attention enhancing mechanism |
CN110059717A (en) * | 2019-03-13 | 2019-07-26 | 山东大学 | Convolutional neural networks automatic division method and system for breast molybdenum target data set |
CN110070073A (en) * | 2019-05-07 | 2019-07-30 | 国家广播电视总局广播电视科学研究院 | Pedestrian's recognition methods again of global characteristics and local feature based on attention mechanism |
WO2019153908A1 (en) * | 2018-02-11 | 2019-08-15 | 北京达佳互联信息技术有限公司 | Image recognition method and system based on attention model |
CN110197182A (en) * | 2019-06-11 | 2019-09-03 | 中国电子科技集团公司第五十四研究所 | Remote sensing image semantic segmentation method based on contextual information and attention mechanism |
-
2019
- 2019-11-25 CN CN201911164824.1A patent/CN111028253B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2019153908A1 (en) * | 2018-02-11 | 2019-08-15 | 北京达佳互联信息技术有限公司 | Image recognition method and system based on attention model |
CN109325911A (en) * | 2018-08-27 | 2019-02-12 | 北京航空航天大学 | A kind of space base rail detection method based on attention enhancing mechanism |
CN110059717A (en) * | 2019-03-13 | 2019-07-26 | 山东大学 | Convolutional neural networks automatic division method and system for breast molybdenum target data set |
CN110070073A (en) * | 2019-05-07 | 2019-07-30 | 国家广播电视总局广播电视科学研究院 | Pedestrian's recognition methods again of global characteristics and local feature based on attention mechanism |
CN110197182A (en) * | 2019-06-11 | 2019-09-03 | 中国电子科技集团公司第五十四研究所 | Remote sensing image semantic segmentation method based on contextual information and attention mechanism |
Non-Patent Citations (1)
Title |
---|
赵欣欣;钱胜胜;刘晓光;: "基于卷积神经网络的铁路桥梁高强螺栓缺失图像识别方法" * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112950586A (en) * | 2021-03-02 | 2021-06-11 | 攀钢集团攀枝花钢铁研究院有限公司 | LF furnace steel slag infrared identification method and system |
Also Published As
Publication number | Publication date |
---|---|
CN111028253B (en) | 2023-05-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107292333B (en) | A kind of rapid image categorization method based on deep learning | |
CN107194872B (en) | Remote sensed image super-resolution reconstruction method based on perception of content deep learning network | |
CN106875381B (en) | Mobile phone shell defect detection method based on deep learning | |
CN110009028B (en) | Microscopic image data enhancement method and device | |
CN109087303A (en) | The frame of semantic segmentation modelling effect is promoted based on transfer learning | |
CN110827312A (en) | Learning method based on cooperative visual attention neural network | |
CN113837193A (en) | Zinc flotation froth image segmentation algorithm based on improved U-Net network | |
CN104063713A (en) | Semi-autonomous on-line studying method based on random fern classifier | |
CN111709901A (en) | Non-multiple multi/hyperspectral remote sensing image color homogenizing method based on FCM cluster matching and Wallis filtering | |
CN103810522A (en) | Counting method and device for corn ear grains | |
CN111127454A (en) | Method and system for generating industrial defect sample based on deep learning | |
CN112200797B (en) | Effective training method based on PCB noise labeling data | |
CN113591866A (en) | Special job certificate detection method and system based on DB and CRNN | |
CN114612664A (en) | Cell nucleus segmentation method based on bilateral segmentation network | |
Wang et al. | Automatic segmentation of concrete aggregate using convolutional neural network | |
CN115601330A (en) | Colonic polyp segmentation method based on multi-scale space reverse attention mechanism | |
CN111028253A (en) | Iron concentrate powder segmentation method and segmentation device | |
CN107316063A (en) | Multiple labeling sorting technique, device, medium and computing device | |
CN112784894B (en) | Automatic labeling method for rock slice microscopic image | |
CN110348339B (en) | Method for extracting handwritten document text lines based on case segmentation | |
CN104123723A (en) | Structure compensation based image quality evaluation method | |
CN108197663A (en) | Based on the calligraphy work image classification method to pairing set Multi-label learning | |
CN113470035A (en) | Image segmentation method based on sequence image error network correction | |
CN112950655A (en) | Land use information automatic extraction method based on deep learning | |
Xue et al. | An efficient method of casting defects detection based on deep learning |
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 |