CN111860674B - Sample category identification method, sample category identification device, computer equipment and storage medium - Google Patents
Sample category identification method, sample category identification device, computer equipment and storage medium Download PDFInfo
- Publication number
- CN111860674B CN111860674B CN202010738650.1A CN202010738650A CN111860674B CN 111860674 B CN111860674 B CN 111860674B CN 202010738650 A CN202010738650 A CN 202010738650A CN 111860674 B CN111860674 B CN 111860674B
- Authority
- CN
- China
- Prior art keywords
- sample
- result
- training
- semantic
- model
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- 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
- G06N3/08—Learning methods
- G06N3/088—Non-supervised learning, e.g. competitive learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to the field of artificial intelligence, and discloses a sample category identification method, a sample category identification device, computer equipment and a storage medium, wherein the method comprises the following steps: acquiring a sample to be identified; inputting a first sample detection model to extract sample characteristics and identify semantic characteristics to obtain a sample identification result to be processed; when the sample feature space result and the sample semantic space result are not matched with all the first sample categories, inputting the sample feature space result and the sample semantic space result into a second sample detection model; clustering the sample feature space results to obtain a first abnormal result, and performing similarity matching on the sample semantic space results to obtain a second abnormal result; and outputting the sample category of the sample to be identified. The invention realizes automatic identification of sample category, saves labor cost and improves identification accuracy. The invention is suitable for the fields of intelligent traffic or intelligent medical treatment, and the like, can further promote the construction of intelligent cities, and also relates to a blockchain technology, wherein a first sample detection model can be stored in the blockchain.
Description
Technical Field
The present invention relates to the field of image classification of artificial intelligence, and in particular, to a method and apparatus for identifying a sample class, a computer device, and a storage medium.
Background
Currently, with the development of technology, the application of pattern recognition in life is becoming more and more common. In order to obtain a good recognition effect in image recognition, a large number of sample images of each type are required for training. However, it is often difficult to collect a large number of training samples of each category in practical application, especially when identifying an abnormal category or a new category that contains a known description and is difficult to collect, it is difficult to collect images of the categories as samples for training, because the abnormal category or the new category is manually identified, and the probability of occurrence is extremely low (because the model that can accurately identify the category is required to be obtained through destructive experiments or is required to be obtained at high cost, etc.), and the model that can accurately identify the category is not trained, and high labor cost is required to identify and classify the model, so that the problem of identifying zero samples is very important, and the problem is more and more focused in industry. For example, in the quality inspection of parts, images of normal parts and broken parts are easy to collect, images of parts with abnormal conditions such as slight cracks, slight depressions, internal breakage and the like are difficult to collect, and the recognition result is poor; in the aspect of vehicle damage assessment and identification, vehicle damage photos of scratches, tears and pits are easy to collect, vehicle damage photos with serious damage such as wrinkles, dead folds and deletions are difficult to collect, and damage assessment and identification are poor; in medical image inspection, common X-ray films such as normal lung, pneumonia, hydrocele and the like are easy to collect, but rare X-ray films such as early stage of tuberculosis, early stage of lung cancer and the like are difficult to collect, and pulmonary medical detection is poor.
Disclosure of Invention
The invention provides a sample type identification method, a sample type identification device, computer equipment and a storage medium, which can automatically identify the sample type of a sample to be identified through a first sample detection model based on end-to-end and a second sample detection model based on zero sample learning.
A sample class identification method, comprising:
acquiring a sample to be identified;
inputting the sample to be identified into a first sample detection model for sample feature extraction and semantic feature identification to obtain a sample identification result to be processed; the sample identification result to be processed comprises a first sample detection result, a sample feature space result and a sample semantic space result;
when the first sample detection result is not matched with all the first sample categories, inputting the sample feature space result and the sample semantic space result into a second sample detection model;
clustering the sample feature space results through the second sample detection model to obtain a first abnormal result, and performing similarity matching on the sample semantic space results to obtain a second abnormal result;
And determining the sample category of the sample to be identified according to the first abnormal result and the second abnormal result and outputting the sample category.
A sample class identification device, comprising:
the acquisition module is used for acquiring a sample to be identified;
the identification module is used for inputting the sample to be identified into a first sample detection model to extract sample characteristics and identify semantic characteristics so as to obtain an identification result of the sample to be processed; the sample identification result to be processed comprises a first sample detection result, a sample feature space result and a sample semantic space result;
the matching module is used for inputting the sample characteristic space result and the sample semantic space result into a second sample detection model when the first sample detection result is not matched with all the first sample categories;
the anomaly module is used for carrying out clustering treatment on the sample feature space results through the second sample detection model to obtain a first anomaly result, and carrying out similarity matching on the sample semantic space results to obtain a second anomaly result;
and the output module is used for determining the sample category of the sample to be identified according to the first abnormal result and the second abnormal result and outputting the sample category.
A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the sample class identification method described above when the computer program is executed.
A computer readable storage medium storing a computer program which, when executed by a processor, implements the steps of the sample class identification method described above.
The invention provides a sample category identification method, a sample category identification device, computer equipment and a storage medium, wherein a sample to be identified is obtained; inputting the sample to be identified into a first sample detection model for sample feature extraction and semantic feature identification to obtain a sample identification result to be processed; the sample identification result to be processed comprises a first sample detection result, a sample feature space result and a sample semantic space result; when the first sample detection result is not matched with all the first sample categories, inputting the sample feature space result and the sample semantic space result into a second sample detection model; clustering the sample feature space results through the second sample detection model to obtain a first abnormal result, and performing similarity matching on the sample semantic space results to obtain a second abnormal result; according to the first abnormal result and the second abnormal result, the sample category of the sample to be identified is determined and output, so that the sample to be identified is identified through the end-to-end first sample detection model, when the first sample detection result output by the first sample detection model is not matched with the first sample category, the sample category of the sample to be identified is identified through the second sample detection model based on zero sample learning according to the association relation between the sample feature space result and the sample semantic space result, and therefore, the sample category of the sample difficult to collect or the sample category can be automatically identified accurately, the cost of manual identification is reduced, the sample category is automatically marked for the sample difficult to collect or the sample category, the labor cost is saved, and the identification accuracy is improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments of the present invention will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic view of an application environment of a sample class identification method according to an embodiment of the invention;
FIG. 2 is a flow chart of a sample class identification method according to an embodiment of the invention;
FIG. 3 is a flow chart of a sample class identification method according to another embodiment of the invention;
FIG. 4 is a flowchart of step S20 of a sample class identification method according to an embodiment of the present invention;
FIG. 5 is a flowchart of step S206 of the sample class identification method according to an embodiment of the present invention;
FIG. 6 is a flowchart of step S2063 of the sample class identification method in one embodiment of the present invention;
FIG. 7 is a flowchart of step S40 of a sample class identification method according to an embodiment of the present invention;
FIG. 8 is a schematic block diagram of a sample class identification device according to an embodiment of the present invention;
FIG. 9 is a schematic diagram of a computer device in accordance with an embodiment of the invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The sample category identification method provided by the invention can be applied to an application environment as shown in fig. 1, wherein a client (computer equipment) communicates with a server through a network. Among them, clients (computer devices) include, but are not limited to, personal computers, notebook computers, smartphones, tablet computers, cameras, and portable wearable devices. The server may be implemented as a stand-alone server or as a server cluster composed of a plurality of servers.
In one embodiment, as shown in fig. 2, a sample class identification method is provided, and the technical scheme mainly includes the following steps S10-S50:
s10, acquiring a sample to be identified.
It is understood that, the sample recognition request is triggered under the application scenario of recognizing the sample category of the collected sample, the sample to be recognized in the sample recognition request is obtained, and the application scenario can be set according to requirements, for example, the collected part sample image is recognized on the quality inspection of the automobile part, or the sample for training the damage-assessment recognition model is recognized on the vehicle damage, or the sample for training the lung recognition model is recognized on the medical image inspection, and the sample to be recognized is a sample image or a photo file which needs to be recognized.
S20, inputting the sample to be identified into a first sample detection model for sample feature extraction and semantic feature identification to obtain a sample identification result to be processed; the sample identification result to be processed comprises a first sample detection result, a sample feature space result and a sample semantic space result.
The method comprises inputting the sample to be identified into the first sample detection model, wherein the first sample detection model is a multi-branch convolutional neural network model which is identified by a plurality of branches and trained through training samples of a plurality of first sample categories, the first sample categories are conventional and easy to collect, a large number of sample categories such as non-destructive, scratch, tearing, dent and the like of a vehicle, the network structure of the first sample detection model can be set according to requirements, such as the network structure of a Resnet50, the network structure of a CNN, the network structure of a VGG and the like, preferably, the network structure of the first sample detection model is the network structure of the Resnet50, the sample characteristics are characteristics related to the first sample category identification of a plurality of branch dimensions, the sample characteristics comprise high-dimension dominant characteristics and low-dimension implicit characteristics, such as in vehicle damage identification, the high-dimension characteristics comprise deformation characteristics, color difference characteristics and the like, the low-dimension implicit characteristics comprise smoothness characteristics and the like, the characteristic of the sample characteristics correspond to the characteristic vector to the sample characteristics of the sample to the sample category, the sample characteristics can be obtained by a vector according to the characteristic of the sample category, the sample characteristics of the sample category identification vector is obtained after the sample is processed according to the required by the characteristic vector of the sample category, the sample detection model is set according to the characteristic vector of the sample category of the sample detection model, the sample feature space result can embody each feature in the sample feature through multiple dimensions, the semantic feature recognition further comprises the steps of obtaining a semantic feature vector corresponding to the input of the final mapping function output by converting the final mapping function in the embedded space model after the input is converged, wherein the final mapping function is a mapping function after the convergence, and recording all the semantic feature vectors output by the embedded space model after the convergence according to the sample feature space result as the sample semantic space result.
Identifying the sample to be identified through the first sample detection model, acquiring the identified probability value corresponding to each first sample category, screening the largest value in all probability values corresponding to each first sample category, and determining the first sample category corresponding to the largest value in all probability values as the first sample detection result corresponding to the sample to be identified if the largest value in all probability values is greater than or equal to a preset probability threshold; and if the maximum value in all the probability values is smaller than the probability threshold value, confirming the first sample detection result corresponding to the sample to be identified as null or abnormal type.
In an embodiment, as shown in fig. 3, after the step S20, that is, after the sample to be identified is input into the first sample detection model to perform sample feature extraction and semantic feature recognition, the method further includes:
and S60, when the first sample detection result is matched with any one of the first sample categories, determining the matched first sample category as the sample category of the sample to be identified.
It may be appreciated that if the first sample detection result matches one of all the first sample categories, the matching manner may be set according to requirements, for example, the matching manner is completely consistent with the content of the first sample category, or the matching is determined when the similarity with the content of the first sample category reaches a preset category probability, and the matching first sample category is recorded as the sample category of the sample to be identified.
Therefore, whether the sample to be identified is one of the first sample categories can be accurately identified through the end-to-end first sample detection model, and the identification accuracy is improved.
In an embodiment, as shown in fig. 4, before the step S20, that is, before the sample to be identified is input into the first sample detection model for sample feature extraction and semantic feature recognition, the method includes:
s201, acquiring a training sample set; the training sample set comprises a plurality of training samples, one of the training samples being associated with a first sample class and a first sample description.
It is understood that the training sample set is a set of the training samples, the training samples are images of various first sample categories which are easy to collect, one of the training samples is associated with one first sample category, and is associated with one of the first sample descriptions, the first sample descriptions are text descriptions of the first sample category in the training samples, such as a vehicle damage photograph with a scraped vehicle body is taken as the training sample, the associated first sample category is "scratch", and the associated first sample descriptions are "damage areas with no obvious deformation and depression of a scraped vehicle body middle part with a length of 7 cm and a width of 2 cm".
S202, inputting the training samples into a multi-branch convolutional neural network model containing initial parameters.
Understandably, the training samples are input to the multi-branch convolutional neural network model, which includes the initial parameters including parameters of the network structure and dimensional parameters of feature vectors, and so on.
In one embodiment, before the step S202, that is, before the training samples are input into the multi-branch convolutional neural network model with initial parameters, the method includes:
S2021, obtaining all migration parameters of the trained unsupervised domain training model through migration learning, and determining all the migration parameters as the initial parameters in the multi-branch convolutional neural network model.
Understandably, the migration learning (Transfer Learning, TL) is a multi-branch convolutional neural network model that uses parameters of training models existing in other fields to be applied in the task of the field, that is, the multi-branch convolutional neural network model acquires all model parameters of an unsupervised domain training model (unsupervised pre-training) by means of migration learning, and the unsupervised domain training model may be an unsupervised learning multi-branch convolutional neural network model in the natural field or the vehicle recognition field, and then determines all model parameters as initial parameters of the multi-branch convolutional neural network model. Therefore, the invention obtains the initial parameters from the unsupervised domain training model after training through transfer learning, can shorten the iteration times of the model, simplify the training process and improve the training efficiency.
And S203, extracting sample characteristics of the training samples through the multi-branch convolutional neural network model to obtain an image characteristic space result.
The sample feature is a feature related to sample recognition of multiple branch dimensions, the sample feature comprises a high-dimension dominant feature and a low-dimension recessive feature, the sample feature can be set according to requirements, for example, in vehicle loss recognition, the high-dimension dominant feature comprises a deformation feature, a color difference feature and the like, the low-latitude recessive feature comprises a smoothness feature and the like, the image feature space result is an image feature vector obtained by carrying out feature convolution and regularization processing on the training sample through the multi-branch convolutional neural network model, the image feature vector is an array which reflects one feature of the sample features and is measured through the feature vector, the dimension of the image feature vector can be set according to requirements, for example, the dimension of the image feature vector can be 256 dimensions, one branch task corresponds to one feature of the sample feature and also corresponds to one image feature vector, and the image feature space result can reflect each feature of the sample features through multiple dimensions.
In an embodiment, in S203, that is, the performing, by using the multi-branch convolutional neural network model, sample feature extraction on the training sample to obtain an image feature space result includes:
S2031, performing image preprocessing on the training sample to obtain a preprocessed sample image;
understandably, the preprocessing is an operation process of performing feature region recognition on the training sample, extracting an image with a preset size, and performing enhancement processing on the extracted image, in which a region with sample features is recognized in the training sample, the region with sample features can be recognized through YOLO (You Only Look Once) algorithm, an image with a preset size is extracted after the feature region with sample features is recognized, the preset size can be set according to requirements, preferably, the preset size is 224×224, and the enhancement processing is performed on the extracted image, and the enhancement processing can be set according to requirements, for example, the enhancement processing is denoising and sharpening, so as to obtain the preprocessed sample image.
S2032, extracting features of the preprocessed sample image to obtain at least one feature vector diagram;
understandably, feature extraction is performed on each channel of the preprocessed sample image, where the feature extraction is to perform convolution on each channel by using different convolution cores corresponding to different features to output a feature vector diagram, and a convolution process is determined according to a network structure of the multi-branch convolutional neural network model, for example, feature extraction of high-dimensional features is performed on each channel of the preprocessed sample image, so as to obtain a feature vector diagram corresponding to the high-dimensional features, where the feature vector diagram is composed of a plurality of feature vector values.
S2033, regularizing each feature vector graph to obtain an image feature vector corresponding to each feature vector graph;
understandably, the regularization processing is to perform N-time root regularization nonlinear processing on feature vector values in each feature vector graph, and fine adjustment can be performed on the feature vector graph through the regularization processing, so that each feature can be reflected, and an image feature vector is output.
And S2034, determining all the image feature vectors as the image feature space result.
Understandably, all of the image feature vectors are labeled as the image feature space results.
The invention realizes that the image preprocessing is carried out on the training sample to obtain a preprocessed sample image; extracting features of the preprocessed sample image to obtain at least one feature vector diagram; regularization processing is carried out on each feature vector diagram to obtain an image feature vector corresponding to each feature vector diagram; and determining all the image feature vectors as the image feature space result, so that the accuracy and the reliability of recognition can be improved through image preprocessing, feature extraction and regularization processing.
S204, carrying out semantic feature recognition on the image feature space result through an embedded space model in the multi-branch convolutional neural network model to obtain a semantic feature space result; the embedded space model is obtained by constructing an association relationship between the image feature vector and the semantic feature vector.
The embedded space model is obtained by learning a mapping relation between an image feature vector corresponding to a training sample and a first sample description corresponding to the training sample, namely, inputting the first sample description of the training sample into a Word2 vec-based semantic recognition model, performing semantic vector conversion on the first sample description to obtain a semantic feature vector corresponding to the first sample description, wherein the Word2 vec-based semantic recognition model is a trained deep neural network model, and the semantic recognition model can perform semantic vector conversion on an input text to generate vector values of semantic features corresponding to each feature of the sample features, namely, generate semantic feature vectors, for example: the first sample description is described as that 'the damage area of the middle part of the vehicle body, which is scraped out and has no obvious deformation and recess and has the length of 7 cm and the width of 2 cm', a vector value corresponding to a semantic feature related to deformation is obtained after semantic vector conversion, a vector value corresponding to a semantic feature related to chromatic aberration is obtained, a vector value corresponding to a semantic feature related to a smooth surface is obtained, a mapping function between an image feature vector corresponding to a training sample and the semantic feature vector is established, the mapping function is continuously learned through continuous iteration of the multi-branch convolutional neural network model until convergence of the multi-branch convolutional neural network model is achieved, at the moment, the mapping function learning between the image feature vector corresponding to the training sample and the semantic feature vector is completed, that is, the average variance between a value output after the image feature vector is input into the mapping function and the semantic feature vector corresponding to the image feature vector is minimum, that is completed after the embedding space model is converged, the semantic feature is identified as the feature vector corresponding to the input into the mapping function in the embedding space model, the semantic feature is obtained, and the feature vector corresponding to the input semantic feature vector is obtained according to the semantic feature result of the image space.
S205, recognizing the image feature space result through a K-means clustering algorithm to obtain a training class result, and determining the training class result, the image feature space result and the semantic feature space result as sample training results.
Understandably, the K-means clustering algorithm is also called a K-means clustering algorithm, and uses a distance as an evaluation index of similarity, determines a training class corresponding to each training class according to a distance of a cluster close to the training class, calculates a Euclidean distance between each image feature vector in the image feature space result and the center of each training class result through the K-means clustering algorithm, determines a training class result corresponding to each Euclidean distance according to each Euclidean distance, and the class contained in the training class result is identical to the class contained in the first sample class, so that end-to-end identification is achieved, the feature corresponding to the same class is extracted in a concentrated mode, and the identification accuracy is improved.
And marking the training class result, the image feature space result and the semantic feature space result corresponding to the training sample as the sample training result corresponding to the training sample.
S206, obtaining a first loss value according to the image feature space result and the first sample category corresponding to the training sample; and obtaining a second loss value according to the semantic feature space result and the first sample description corresponding to the training sample.
Understandably, the first loss value is obtained by calculating a gap between the image feature space result and a center vector corresponding to the first sample class; inputting the first sample description into a Word2 vec-based semantic recognition model, carrying out semantic vector conversion on the first sample description to obtain a semantic feature vector corresponding to the first sample description, and calculating a semantic similarity value between the semantic feature space result and the semantic feature vector to obtain the second loss value, wherein the second loss value characterizes the difference between the semantic feature space result and the semantic feature vector.
In an embodiment, as shown in fig. 5, in step S206, that is, the sample feature extraction is performed on the training sample by the multi-branch convolutional neural network model, to obtain an image feature space result, which includes:
S2061, performing vector conversion on the first sample category corresponding to the training sample through the multi-branch convolutional neural network model to obtain a center vector corresponding to the first sample category; the center vectors include an European domain center vector and an angular domain center vector.
Understandably, the first sample class is subjected to vector conversion, and the vector conversion is performed to convert the first sample class of the text type according to a preset mapping relation, so as to obtain the center vector matched with the first sample class, wherein the center vector comprises the center vector in the European domain and the center vector in the angle domain.
S2062, obtaining the European style loss value according to the image characteristic space result and the European style domain center vector by a cross entropy loss algorithm.
Understandably, a loss value between each of the image feature vectors and the euclidean domain center vector in the image feature space result is calculated by a cross entropy loss function, and the loss value is recorded as the euclidean loss value.
S2063, obtaining an angle loss value according to the image feature space result and the angle domain center vector by an ArcFace loss algorithm.
Understandably, a loss value between each of the image feature vectors and the angular domain center vector in the image feature space result is calculated by an ArcFace loss function, and the loss value is recorded as the angular loss value.
In an example, as shown in fig. 6, in the step S2063, that is, the obtaining, by an ArcFace loss algorithm, an angle loss value according to the image feature space result and the angle domain center vector includes:
s20631, regularizing the image feature space result through a regularization model in the multi-branch convolutional neural network model to obtain a regularized feature vector.
Understandably, the multi-branch convolutional neural network model includes the regularization model, where the regularization model includes a regularization function, and inputs each image feature vector into the regularization function, and outputs the regularization feature vector, where the regularization feature vector is:
wherein the mathematical notation function:
，b 1 for the first feature vector value, b, of the image feature vectors m For the mth feature vector value, b, in the image feature vector i And N is the value of the preset N times square root for the ith feature vector value in the image feature vector.
S20632, inputting the regularized feature vector and the angle domain center vector into an angle domain loss model, and obtaining an angle loss value through the ArcFace loss algorithm in the angle domain loss model.
The ArcFace loss algorithm is an algorithm for calculating a loss value of the ArcFace loss function, the regularized feature vector and the angle domain center vector are subjected to a cosine angle comparison method in the ArcFace loss function, and the difference is measured through the direction of an angle domain, so that the angle loss value is obtained.
Therefore, the angle loss value is obtained through the regularization model and the angle domain loss model in the multi-branch convolutional neural network model, so that model divergence can be limited through regularization, the fine tuning effect in the identification and classification process is achieved, and the consideration of compact distance between categories can be increased through the angle domain loss model.
And S2064, performing weighting processing on the European loss value and the angle loss value to obtain the first loss value.
It can be understood that, by inputting the euclidean loss value and the angular loss value into a first loss function, converting the euclidean loss value and the angular loss value into indexes of the same dimension and performing weighted product processing to obtain the first loss value output by the first loss function, by adding the consideration of the angular loss value on the basis of the euclidean loss value, not only the consideration of correct category classification, but also the consideration of compact distance between categories can be performed, so that the first loss function has good expressive force on classification results. In this way, the invention realizes vector conversion of the first sample category corresponding to the training sample through the multi-branch convolutional neural network model to obtain a center vector corresponding to the first sample category; obtaining an European loss value through a cross entropy loss algorithm; acquiring an angle loss value through an ArcFace loss algorithm; and carrying out weighting processing on the European loss value and the angle loss value to obtain the first loss value, so that the identification is closer to a more compact identification direction by introducing the angle loss value, and the identification precision is more accurate.
S207, obtaining a total loss value according to the first loss value and the second loss value.
It is understandable that the first loss value and the second loss value are input into a total loss model containing a total loss function in the multi-branch convolutional neural network model, the total loss function in the total loss model can be set according to requirements, the loss model is a model for generating the total loss value, and the total loss value is calculated through the total loss function.
In an embodiment, in the step S207, the obtaining a total loss value according to the first loss value and the second loss value includes:
s2071, inputting the first loss value and the second loss value into a preset total loss model, and calculating the total loss value through a total loss function in the total loss model; the total loss function is:
L＝w 1 ×X1+w 2 ×X2
wherein, the liquid crystal display device comprises a liquid crystal display device,
x1 is a first loss value;
x2 is a second loss value;
w 1 weights for the first loss value;
w 2 is the weight of the second loss value.
And S208, when the total loss value does not reach a preset convergence condition, iteratively updating initial parameters of the multi-branch convolutional neural network model, triggering the multi-branch convolutional neural network model to extract sample characteristics of the training samples to obtain image characteristic space results, and recording the multi-branch convolutional neural network model after convergence as the first sample detection model after training.
As can be appreciated, the convergence condition may be a condition that the value of the total loss value is small and will not fall down after 10000 times of calculation, that is, when the value of the total loss value is small and will not fall down again after 10000 times of calculation, training is stopped, and the multi-branch convolutional neural network model after convergence is recorded as the first sample detection model after training is completed; the convergence condition may be a condition that the total loss value is smaller than a set threshold, that is, when the total loss value is smaller than the set threshold, training is stopped, the multi-branch convolutional neural network model after convergence is recorded as the first sample detection model after training is completed, and the first sample detection model after training may be stored in a blockchain. In this way, when the total loss value does not reach the preset convergence condition, the initial parameters of the multi-branch convolutional neural network model are updated and iterated continuously, and the step of extracting the sample characteristics of the training sample through the multi-branch convolutional neural network model to obtain the image characteristic space result is triggered, so that the accurate result can be closed continuously, and the recognition accuracy is higher and higher.
It is emphasized that, to further ensure the privacy and security of the first sample detection model, the first sample detection model may also be stored in a node of the blockchain.
The blockchain is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, encryption algorithm and the like. The Blockchain (Blockchain), which is essentially a decentralised database, is a string of data blocks that are generated by cryptographic means in association, each data block containing a batch of information of network transactions for verifying the validity of the information (anti-counterfeiting) and generating the next block. The blockchain may include a blockchain underlying platform, a platform product services layer, an application services layer, and the like. The decentralized completely distributed DNS service provided by the blockchain can realize the inquiry and analysis of the domain name through the peer-to-peer data transmission service between the nodes in the network, and can be used to ensure that the operating system and firmware of some important infrastructure are not tampered, monitor the state and integrity of software, discover bad tampering, ensure that the transmitted data is not tampered, store the first sample detection model in the blockchain, and ensure the privacy and security of the first sample detection model.
In an embodiment, after the step S207, that is, after the step S207, the step of obtaining a total loss value according to the first loss value and the second loss value, the method further includes:
and S209, recording the multi-branch convolutional neural network model after convergence as the first sample detection model after training when the total loss value reaches the convergence condition.
It is understandable that, when the total loss value occurs for the first time, if the total loss value reaches the convergence condition, it is indicated that the total loss value has reached an optimal result, at which time the multi-branch convolutional neural network model has converged, and the multi-branch convolutional neural network model after convergence is recorded as the first sample detection model after training is completed.
S30, when the first sample detection result is not matched with all the first sample categories, inputting the sample feature space result and the sample semantic space result into a second sample detection model.
It is understandable that if the first sample detection result does not match any of all the first sample categories, recording the sample to be identified as the abnormal sample, the abnormal sample indicating that the result other than all the first sample categories is abnormal, inputting the sample feature space result and the sample semantic space result into the second sample detection model, wherein the second sample detection model is an adaptive model learned by manually marking the features of the known second sample category reflected in each feature of the sample features corresponding to the second sample category according to the final mapping function and the K-nearest neighbor algorithm, namely, establishing a mapping relation between the clustering range of each feature in the sample features and the range of semantic feature vectors and the known second sample category, the second sample category is a sample category which is difficult to collect (a small amount of) and is known to describe (a description related to sample features is known, and is also a description of the second sample in the whole text), and a newly found sample category which is known to describe, for example, on the basis of recognition of vehicle damage, recognition of "missing" features (having common features of "tearing" and "sinking"), a value obtained by weighting and adding vector values corresponding to deformation-related semantic features, vector values corresponding to color difference-related semantic features and vector values corresponding to semantic features related to a smooth surface falls within a neighbor range value of "tearing" and "sinking", mapping as a second sample category of "missing", and the second sample detection model automatically adjusts a mapping function in the mapping relation according to the common characteristics between each second sample category and the first sample category, and finally reaches the parameter with the highest fitting degree.
S40, clustering the sample feature space results through the second sample detection model to obtain a first abnormal result, and performing similarity matching on the sample semantic space results to obtain a second abnormal result.
The clustering method includes the steps of calculating the sample feature space result through a K-means clustering algorithm, calculating the Euclidean distance value of the sample feature space result from the center (centroid) of a cluster corresponding to each first sample category, calculating each Euclidean distance value through the second sample detection model, determining a second sample category corresponding to the area range according to the area range where each Euclidean distance value falls after calculating each Euclidean distance value, determining the second sample category as a clustering abnormal category, converting each Euclidean distance value into a probability value of each second sample category, recording the clustering abnormal category and the probability value of each second sample category as the first abnormal result, wherein the first abnormal result indicates a result identified through the dimension of the sample feature space, and simultaneously, calculating a similarity value between the sample feature space result and a semantic feature vector corresponding to each first sample category through the second sample detection model, obtaining a positive similarity value, calculating a positive similarity value, setting a algorithm for calculating the positive similarity value according to requirements, such as a similarity value, and recording the similarity value adjacent to each second sample category as a similarity value according to the first similarity value, and determining the similarity value of the second similarity value and the similarity value according to the second similarity value.
In an embodiment, as shown in fig. 7, in step S40, that is, performing similarity matching on the sample semantic space result to obtain a second abnormal result includes:
s401, calculating a similarity value between the sample semantic space result and second sample descriptions corresponding to each preset second sample category through a Word2vec model in the second sample detection model.
Understandably, the second sample detection model may further perform Word vector conversion on second sample descriptions corresponding to preset second sample categories through the Word2vec model to obtain semantic Word vectors corresponding to the second sample categories, where the second sample categories are sample categories of (a small number of) known descriptions (already known descriptions related to sample features, also referred to as second sample descriptions in the whole text) which are difficult to collect and sample categories of newly discovered and known descriptions, and then calculate similarity values between the sample semantic space results and the semantic Word vectors through a Word2vec algorithm in the Word2vec model.
The Word2vec model is a neural network model constructed based on a Word2vec algorithm, and second sample descriptions corresponding to second sample categories and related to each feature in sample features are preset in the Word2vec model.
S402, acquiring the second sample category corresponding to the maximum similarity value in all the similarity values through the second sample detection model, and determining the acquired second sample category as the second abnormal result.
Understandably, the largest similarity value is obtained from the similarity values between all the sample semantic space results and each semantic word vector, and the corresponding anomaly type is determined as the second anomaly result, wherein the second anomaly result comprises the obtained largest similarity value.
Therefore, the similarity value between the sample semantic space result and the second sample description corresponding to each preset second sample category is calculated through the Word2vec model in the second sample detection model; and then, the second sample category corresponding to the maximum similarity value in all the similarity values is acquired, and the acquired second sample category is determined to be the second abnormal result, so that the automatic identification of the second abnormal result based on a Word2vec algorithm is realized, a semantic identification method is provided, and the identification is more accurate.
S50, determining the sample category of the sample to be identified according to the first abnormal result and the second abnormal result and outputting the sample category.
Understandably, the final second sample class is obtained by weighting each probability value in the first abnormal result and each probability value in the second abnormal result, calculating according to the set weight parameters, and pulling the difference between the probability values corresponding to the second sample classes by the weight parameters, so that classification is more accurate, for example, the sum of the probability values of the "missing" second sample class weighted by the weight parameters in the first abnormal result and the second abnormal result is maximum, and the "missing" second sample class is determined as the sample class of the sample to be identified.
Wherein the sample class includes the first sample class and the second sample class, in an application scenario, samples of the first sample class may be identified from a batch of samples, and samples of the second sample class may be identified based on training samples of zero second sample class, for example: in the collected train damage photo sample set, train damage photos of a first sample category of scratches, tears, depressions can be identified, and train damage photos of a second sample category of wrinkles, dead folds, missing and the like can be identified (photos of the second sample category of wrinkles, dead folds, missing and the like are not collected before).
The invention realizes the identification of the sample by obtaining the sample to be identified; inputting the sample to be identified into a first sample detection model for sample feature extraction and semantic feature identification to obtain a sample identification result to be processed; the sample identification result to be processed comprises a first sample detection result, a sample feature space result and a sample semantic space result; when the first sample detection result is not matched with all the first sample categories, inputting the sample feature space result and the sample semantic space result into a second sample detection model; clustering the sample feature space results through the second sample detection model to obtain a first abnormal result, and performing similarity matching on the sample semantic space results to obtain a second abnormal result; according to the first abnormal result and the second abnormal result, the sample category of the sample to be identified is determined and output, so that the sample to be identified is identified through the end-to-end first sample detection model, when the first sample detection result output by the first sample detection model is not matched with the first sample category, the sample category of the sample to be identified is identified through the second sample detection model based on zero sample learning according to the association relation between the sample feature space result and the sample semantic space result, and therefore, the sample category of the sample difficult to collect or the sample category can be automatically identified accurately, the cost of manual identification is reduced, the sample category is automatically marked for the sample difficult to collect or the sample category, the labor cost is saved, and the identification accuracy is improved.
In an embodiment, a sample class identification device is provided, where the sample class identification device corresponds to the sample class identification method in the above embodiment one by one. As shown in fig. 8, the sample class recognition apparatus includes an acquisition module 11, a recognition module 12, a matching module 13, an abnormality module 14, and an output module 15.
The functional modules are described in detail as follows:
an obtaining module 11, configured to obtain a sample to be identified;
the recognition module 12 is configured to input the sample to be recognized into a first sample detection model for sample feature extraction and semantic feature recognition, so as to obtain a sample recognition result to be processed; the sample identification result to be processed comprises a first sample detection result, a sample feature space result and a sample semantic space result;
the matching module 13 is configured to input the sample feature space result and the sample semantic space result into a second sample detection model when the first sample detection result is not matched with all the first sample categories;
the anomaly module 14 is configured to perform clustering on the sample feature space result through the second sample detection model to obtain a first anomaly result, and perform similarity matching on the sample semantic space result to obtain a second anomaly result;
And the output module 15 is used for determining the sample category of the sample to be identified according to the first abnormal result and the second abnormal result and outputting the sample category.
For specific limitations of the sample class identification means, reference is made to the above description of the sample class identification method, and no further description is given here. The respective modules in the sample class identification device described above may be implemented in whole or in part by software, hardware, or a combination thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.
In one embodiment, a computer device is provided, which may be a server, and the internal structure of which may be as shown in fig. 9. The computer device includes a processor, a memory, a network interface, and a database connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a sample class identification method.
In one embodiment, a computer device is provided that includes a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the sample class identification method of the above embodiments when executing the computer program.
In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, which when executed by a processor implements the sample class identification method of the above embodiments.
Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in embodiments provided herein may include non-volatile and/or volatile memory. The nonvolatile memory can include Read Only Memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), memory bus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), among others.
The blockchain is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, encryption algorithm and the like. The Blockchain (Blockchain), which is essentially a decentralised database, is a string of data blocks that are generated by cryptographic means in association, each data block containing a batch of information of network transactions for verifying the validity of the information (anti-counterfeiting) and generating the next block. The blockchain may include a blockchain underlying platform, a platform product services layer, an application services layer, and the like.
It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-described division of the functional units and modules is illustrated, and in practical application, the above-described functional distribution may be performed by different functional units and modules according to needs, i.e. the internal structure of the apparatus is divided into different functional units or modules to perform all or part of the above-described functions.
The above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention, and are intended to be included in the scope of the present invention.
Claims (9)
1. A method for identifying a sample class, comprising:
acquiring a sample to be identified; the sample to be identified is a sample image which needs to be identified;
inputting the sample to be identified into a first sample detection model for sample feature extraction and semantic feature identification to obtain a sample identification result to be processed; the sample identification result to be processed comprises a first sample detection result, a sample feature space result and a sample semantic space result;
when the first sample detection result is not matched with all the first sample categories, inputting the sample feature space result and the sample semantic space result into a second sample detection model;
clustering the sample feature space results through the second sample detection model to obtain a first abnormal result, and performing similarity matching on the sample semantic space results to obtain a second abnormal result;
determining the sample category of the sample to be identified according to the first abnormal result and the second abnormal result and outputting the sample category;
before the sample to be identified is input into a first sample detection model to extract sample characteristics and identify semantic characteristics, the method comprises the following steps:
Acquiring a training sample set; the training sample set comprises a plurality of training samples, one of the training samples being associated with one of a first sample class and one of a first sample description;
inputting the training sample into a multi-branch convolutional neural network model containing initial parameters;
extracting sample characteristics of the training samples through the multi-branch convolutional neural network model to obtain an image characteristic space result;
carrying out semantic feature recognition on the image feature space result through an embedded space model in the multi-branch convolutional neural network model to obtain a semantic feature space result; the embedded space model is obtained by constructing an association relation between the image feature vector and the semantic feature vector;
identifying the image feature space result through a K-means clustering algorithm to obtain a training class result, and determining the training class result, the image feature space result and the semantic feature space result as sample training results;
obtaining a first loss value according to the image feature space result and the first sample category corresponding to the training sample; obtaining a second loss value according to the semantic feature space result and the first sample description corresponding to the training sample;
Obtaining a total loss value according to the first loss value and the second loss value;
and when the total loss value does not reach a preset convergence condition, iteratively updating initial parameters of the multi-branch convolutional neural network model, triggering the multi-branch convolutional neural network model to extract sample characteristics of the training samples, and obtaining an image characteristic space result, and recording the multi-branch convolutional neural network model after convergence as the first sample detection model after training is completed until the total loss value reaches the preset convergence condition.
2. The method for identifying sample category as defined in claim 1, wherein the step of inputting the sample to be identified into a first sample detection model for sample feature extraction and semantic feature identification, and further comprises the steps of:
and when the first sample detection result is matched with any first sample category, determining the matched first sample category as the sample category of the sample to be identified.
3. The method for identifying sample classes according to claim 1, wherein before inputting the training samples into the multi-branch convolutional neural network model containing initial parameters, the method comprises:
And acquiring all migration parameters of the trained unsupervised domain training model through migration learning, and determining all the migration parameters as the initial parameters in the multi-branch convolutional neural network model.
4. The method for identifying a sample class according to claim 1, wherein said obtaining a first loss value based on the image feature space result and the first sample class corresponding to the training sample comprises:
vector conversion is carried out on the first sample category corresponding to the training sample through the multi-branch convolutional neural network model, so that a center vector corresponding to the first sample category is obtained; the center vector comprises an European domain center vector and an angular domain center vector;
obtaining an European style loss value according to the image feature space result and the European style domain center vector through a cross entropy loss algorithm;
obtaining an angle loss value according to the image feature space result and the angle domain center vector by an ArcFace loss algorithm;
and carrying out weighting processing on the European loss value and the angle loss value to obtain the first loss value.
5. The method for identifying a sample class according to claim 4, wherein said obtaining an angle loss value according to the image feature space result and the angle domain center vector by an ArcFace loss algorithm comprises:
Regularizing the image feature space result through a regularization model in the multi-branch convolutional neural network model to obtain a regularized feature vector;
and inputting the regularized feature vector and the angle domain center vector into an angle domain loss model, and obtaining an angle loss value through the ArcFace loss algorithm in the angle domain loss model.
6. The method for identifying sample categories according to claim 1, wherein the performing similarity matching on the sample semantic space results to obtain second abnormal results includes:
calculating a similarity value between the sample semantic space result and a second sample description corresponding to each preset second sample category through a Word2vec model in the second sample detection model;
and acquiring the second sample category corresponding to the maximum similarity value in all the similarity values through the second sample detection model, and determining the acquired second sample category as the second abnormal result.
7. A sample class identification device, comprising:
the acquisition module is used for acquiring a sample to be identified; the sample to be identified is a sample image which needs to be identified;
The identification module is used for inputting the sample to be identified into a first sample detection model to extract sample characteristics and identify semantic characteristics so as to obtain an identification result of the sample to be processed; the sample identification result to be processed comprises a first sample detection result, a sample feature space result and a sample semantic space result;
the matching module is used for inputting the sample characteristic space result and the sample semantic space result into a second sample detection model when the first sample detection result is not matched with all the first sample categories;
the anomaly module is used for carrying out clustering treatment on the sample feature space results through the second sample detection model to obtain a first anomaly result, and carrying out similarity matching on the sample semantic space results to obtain a second anomaly result;
the output module is used for determining the sample category of the sample to be identified according to the first abnormal result and the second abnormal result and outputting the sample category;
the identification module is also used for:
acquiring a training sample set; the training sample set comprises a plurality of training samples, one of the training samples being associated with one of a first sample class and one of a first sample description;
Inputting the training sample into a multi-branch convolutional neural network model containing initial parameters;
extracting sample characteristics of the training samples through the multi-branch convolutional neural network model to obtain an image characteristic space result;
carrying out semantic feature recognition on the image feature space result through an embedded space model in the multi-branch convolutional neural network model to obtain a semantic feature space result; the embedded space model is obtained by constructing an association relation between the image feature vector and the semantic feature vector;
identifying the image feature space result through a K-means clustering algorithm to obtain a training class result, and determining the training class result, the image feature space result and the semantic feature space result as sample training results;
obtaining a first loss value according to the image feature space result and the first sample category corresponding to the training sample; obtaining a second loss value according to the semantic feature space result and the first sample description corresponding to the training sample;
obtaining a total loss value according to the first loss value and the second loss value;
And when the total loss value does not reach a preset convergence condition, iteratively updating initial parameters of the multi-branch convolutional neural network model, triggering the multi-branch convolutional neural network model to extract sample characteristics of the training samples, and obtaining an image characteristic space result, and recording the multi-branch convolutional neural network model after convergence as the first sample detection model after training is completed until the total loss value reaches the preset convergence condition.
8. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the sample class identification method according to any of claims 1 to 6 when executing the computer program.
9. A computer readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, implements the sample class identification method according to any one of claims 1 to 6.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010738650.1A CN111860674B (en) | 2020-07-28 | 2020-07-28 | Sample category identification method, sample category identification device, computer equipment and storage medium |
| PCT/CN2020/135337 WO2021121127A1 (en) | 2020-07-28 | 2020-12-10 | Sample type identification method, apparatus, computer device, and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010738650.1A CN111860674B (en) | 2020-07-28 | 2020-07-28 | Sample category identification method, sample category identification device, computer equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111860674A CN111860674A (en) | 2020-10-30 |
| CN111860674B true CN111860674B (en) | 2023-09-19 |
Family
ID=72948380
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010738650.1A Active CN111860674B (en) | 2020-07-28 | 2020-07-28 | Sample category identification method, sample category identification device, computer equipment and storage medium |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN111860674B (en) |
| WO (1) | WO2021121127A1 (en) |
Families Citing this family (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111860674B (en) * | 2020-07-28 | 2023-09-19 | 平安科技（深圳）有限公司 | Sample category identification method, sample category identification device, computer equipment and storage medium |
| CN112364912B (en) * | 2020-11-09 | 2023-10-13 | 腾讯科技（深圳）有限公司 | Information classification method, device, equipment and storage medium |
| CN112418320B (en) * | 2020-11-24 | 2024-01-19 | 杭州未名信科科技有限公司 | Enterprise association relation identification method, device and storage medium |
| CN112434746B (en) * | 2020-11-27 | 2023-10-27 | 平安科技（深圳）有限公司 | Pre-labeling method based on hierarchical migration learning and related equipment thereof |
| CN113255701B (en) * | 2021-06-24 | 2021-10-22 | 军事科学院***工程研究院网络信息研究所 | Small sample learning method and system based on absolute-relative learning framework |
| CN113378833B (en) * | 2021-06-25 | 2023-09-01 | 北京百度网讯科技有限公司 | Image recognition model training method, image recognition device and electronic equipment |
| CN113627470B (en) * | 2021-07-01 | 2023-09-05 | 汕头大学 | Zero-order learning-based unknown event classification method for optical fiber early warning system |
| CN113657545B (en) * | 2021-08-30 | 2024-05-21 | 平安医疗健康管理股份有限公司 | User service data processing method, device, equipment and storage medium |
| CN113723341B (en) * | 2021-09-08 | 2023-09-01 | 北京有竹居网络技术有限公司 | Video identification method and device, readable medium and electronic equipment |
| CN114298122A (en) * | 2021-10-22 | 2022-04-08 | 腾讯科技（深圳）有限公司 | Data classification method, device, equipment, storage medium and computer program product |
| CN114155420B (en) * | 2021-11-12 | 2024-03-26 | 青岛信芯微电子科技股份有限公司 | Scene recognition model training method, device, equipment and medium |
| US20230186623A1 (en) * | 2021-12-14 | 2023-06-15 | Ping An Technology (Shenzhen) Co., Ltd. | Systems and methods for crop disease diagnosis |
| CN114998613B (en) * | 2022-06-24 | 2024-04-26 | 安徽工业大学 | Multi-mark zero sample learning method based on deep mutual learning |
| CN115331216B (en) * | 2022-08-11 | 2023-05-23 | 北京林业大学 | Walnut fruit maturity cluster identification method, device and system and storage medium |
| CN115677346B (en) * | 2022-11-07 | 2023-09-12 | 北京赛乐米克材料科技有限公司 | Preparation method of colored zirconium precious stone ceramic nose pad |
| CN116403077B (en) * | 2023-06-07 | 2023-08-15 | 中国科学院自动化研究所 | Abnormality detection model training method, abnormality detection device and electronic equipment |
| CN116595978B (en) * | 2023-07-14 | 2023-11-14 | 腾讯科技（深圳）有限公司 | Object category identification method, device, storage medium and computer equipment |
| CN116994076B (en) * | 2023-09-28 | 2024-01-19 | 中国海洋大学 | Small sample image recognition method based on double-branch mutual learning feature generation |
| CN117313899B (en) * | 2023-11-23 | 2024-02-23 | 全芯智造技术有限公司 | Method, apparatus and medium for data processing |
| CN117992898A (en) * | 2024-04-07 | 2024-05-07 | 腾讯科技（深圳）有限公司 | Training method of anomaly detection model, object anomaly detection method and device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108154222A (en) * | 2016-12-02 | 2018-06-12 | 北京市商汤科技开发有限公司 | Deep neural network training method and system, electronic equipment |
| CN110059807A (en) * | 2019-04-26 | 2019-07-26 | 腾讯科技（深圳）有限公司 | Image processing method, device and storage medium |
| CN110472652A (en) * | 2019-06-30 | 2019-11-19 | 天津大学 | A small amount of sample classification method based on semanteme guidance |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109377494B (en) * | 2018-09-14 | 2022-06-28 | 创新先进技术有限公司 | Semantic segmentation method and device for image |
| CN109359725B (en) * | 2018-10-24 | 2021-03-02 | 北京周同科技有限公司 | Training method, device and equipment of convolutional neural network model and computer readable storage medium |
| EP3874417A1 (en) * | 2018-10-29 | 2021-09-08 | HRL Laboratories, LLC | Systems and methods for few-shot transfer learning |
| CN111860674B (en) * | 2020-07-28 | 2023-09-19 | 平安科技（深圳）有限公司 | Sample category identification method, sample category identification device, computer equipment and storage medium |
-
2020
- 2020-07-28 CN CN202010738650.1A patent/CN111860674B/en active Active
- 2020-12-10 WO PCT/CN2020/135337 patent/WO2021121127A1/en active Application Filing
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108154222A (en) * | 2016-12-02 | 2018-06-12 | 北京市商汤科技开发有限公司 | Deep neural network training method and system, electronic equipment |
| CN110059807A (en) * | 2019-04-26 | 2019-07-26 | 腾讯科技（深圳）有限公司 | Image processing method, device and storage medium |
| CN110472652A (en) * | 2019-06-30 | 2019-11-19 | 天津大学 | A small amount of sample classification method based on semanteme guidance |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2021121127A1 (en) | 2021-06-24 |
| CN111860674A (en) | 2020-10-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111860674B (en) | Sample category identification method, sample category identification device, computer equipment and storage medium | |
| CN111667011B (en) | Damage detection model training and vehicle damage detection method, device, equipment and medium | |
| CN109165840B (en) | Risk prediction processing method, risk prediction processing device, computer equipment and medium | |
| US20220253856A1 (en) | System and method for machine learning based detection of fraud | |
| CN111651992A (en) | Named entity labeling method and device, computer equipment and storage medium | |
| CN109063921B (en) | Optimization processing method and device for client risk early warning, computer equipment and medium | |
| CN111291264A (en) | Access object prediction method and device based on machine learning and computer equipment | |
| WO2021159815A1 (en) | Method and apparatus for training facial recognition model, and computer device | |
| CN113707296B (en) | Medical scheme data processing method, device, equipment and storage medium | |
| CN116363357A (en) | Semi-supervised semantic segmentation method and device based on MIM and contrast learning | |
| Qiu et al. | Offline signature authentication algorithm based on the fuzzy set | |
| CN114842478A (en) | Text area identification method, device, equipment and storage medium | |
| Lawal et al. | Face-based gender recognition analysis for Nigerians using CNN | |
| Safavipour et al. | Deep Hybrid Multimodal Biometric Recognition System Based on Features‐Level Deep Fusion of Five Biometric Traits | |
| CN112966676A (en) | Document key information extraction method based on zero sample learning | |
| US20230206676A1 (en) | Systems and Methods for Generating Document Numerical Representations | |
| CN113505716B (en) | Training method of vein recognition model, and recognition method and device of vein image | |
| Vinod et al. | Handwritten Signature Identification and Fraud Detection using Deep Learning and Computer Vision | |
| CN112115443A (en) | Terminal user authentication method and system | |
| Jun et al. | Two-view correspondence learning via complex information extraction | |
| CN114067350A (en) | Model updating method, device and computer program product | |
| CN112926610A (en) | Construction method of license plate image screening model and license plate image screening method | |
| CN111091354A (en) | Real estate information processing system based on big data | |
| CN116959064B (en) | Certificate verification method and device, computer equipment and storage medium | |
| CN113239128B (en) | Data pair classification method, device, equipment and storage medium based on implicit characteristics |
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 |