WO2021114633A1

WO2021114633A1 - Image confidence determination method, apparatus, electronic device, and storage medium

Info

Publication number: WO2021114633A1
Application number: PCT/CN2020/099490
Authority: WO
Inventors: 李风仪; 南洋; 王佳平; 谢春梅; 侯晓帅
Original assignee: 平安科技（深圳）有限公司
Priority date: 2020-05-20
Filing date: 2020-06-30
Publication date: 2021-06-17
Also published as: CN111695604A

Abstract

Provided is an image confidence determination method, relating to image processing, and used in the field of smart medical care, said method comprising: performing image classification processing on a sample image by means of network layers in an image classification model, each network layer comprising a classification calculation layer and an overfitting prevention layer (S101); on the basis of the output result of the classification calculation layer, determining a predicted probability that the sample image belongs to each image category (S102); obtaining a feature image corresponding to the sample image outputted by the overfitting prevention layer, identifying the feature image, and, on the basis of the identification result, determining a first feature parameter and a second feature parameter corresponding to the feature image (S103); on the basis of the first feature parameter and the second feature parameter, determining a calculated probability that the sample image belongs to each image category (S104); determining the degree of similarity between the predicted probability and the calculated probability under each image category, and on the basis of the degree of similarity under each image category, determining the credibility of the sample image (S105); thus is facilitated the efficient determination of the credibility of a sample image. In addition, the method also relates to blockchain technology, and the predicted probability and calculated probability can be stored in the blockchain.

Description

Method and device for determining image credibility, electronic equipment and storage medium

This application requires the priority of a Chinese patent application filed with the Chinese Patent Office on May 20, 2020, the application number is 202010431166.4, and the application title is "a method, device, electronic equipment, and storage medium for determining image credibility". The entire content is incorporated into this application by reference.

Technical field

This application relates to the field of smart medical care, and in particular to a method, device, electronic equipment, and storage medium for determining image credibility.

Background technique

At present, the inventor realizes that in the image classification problem, the focus is mostly on the prediction accuracy of the image classification model. In order to improve the prediction accuracy of the image classification model, a large number of sample images are usually used to train the image classification model. The training of the image classification model is for the purpose of fitting the target, and the prediction result is infinitely close to the training target. The prerequisite of this learning and training is that the sample image is true. If you ignore the credibility of the sample image and blindly chase the infinite closeness of the result, and use this to evaluate the prediction accuracy of the image classification model, it will seriously affect the actual output of the image classification model. The accuracy and credibility of the image classification results. Therefore, in the process of training the image classification model through the sample image, how to determine the credibility of the sample image has become an urgent problem to be solved.

Summary of the invention

The embodiments of the present application provide a method, a device, an electronic device, and a storage medium for determining the credibility of an image, which are conducive to efficiently determining the credibility of a sample image.

On the one hand, an embodiment of the present application provides a method for determining the credibility of an image, and the method includes:

Image classification processing is performed on the sample image through each network layer in the image classification model, and each network layer includes a classification calculation layer and an anti-overfitting layer;

Determining the predicted probability that the sample image belongs to each image category based on the output result of the classification calculation layer;

Acquiring a feature image corresponding to the sample image output by the over-fitting prevention layer;

Identifying the characteristic image, and determining the first characteristic parameter and the second characteristic parameter corresponding to the characteristic image based on the recognition result;

Determining, based on the first feature parameter and the second feature parameter, the calculated probability that the sample image belongs to each image category;

The similarity between the predicted probability and the calculated probability under the respective image categories is determined respectively, and the credibility of the sample image is determined based on the similarity under the respective image categories.

On the other hand, an embodiment of the present application provides a device for determining image credibility, including:

The processing module is used to perform image classification processing on the sample image through each network layer in the image classification model, and each network layer includes a classification calculation layer and an anti-overfitting layer;

The processing module is further configured to determine the predicted probability that the sample image belongs to each image category based on the output result of the classification calculation layer;

An acquisition module, configured to acquire a feature image corresponding to the sample image output by the over-fitting prevention layer;

The processing module is further configured to identify the characteristic image, and determine the first characteristic parameter and the second characteristic parameter corresponding to the characteristic image based on the recognition result;

The processing module is further configured to determine the calculated probability that the sample image belongs to each image category based on the first characteristic parameter and the second characteristic parameter;

The processing module is further configured to determine the similarity between the predicted probability and the calculated probability under each image category, and determine the sample image based on the similarity under each image category Credibility.

In another aspect, an embodiment of the present application provides an electronic device, including a processor, a storage device, and a communication interface. The processor, the storage device, and the communication interface are connected to each other, wherein the storage device is used for storing and supporting terminal execution. The computer program of the above method, the computer program includes program instructions, and the processor is configured to call the program instructions to perform the following steps: perform image classification processing on the sample image through each network layer in the image classification model, so The various network layers include a classification calculation layer and an anti-overfitting layer; the predicted probability of the sample image belonging to each image category is determined based on the output result of the classification calculation layer; the sample output by the anti-overfitting layer is obtained The characteristic image corresponding to the image; the characteristic image is identified, and the first characteristic parameter and the second characteristic parameter corresponding to the characteristic image are determined based on the recognition result; the first characteristic parameter and the second characteristic parameter are determined based on the first characteristic parameter and the second characteristic parameter The sample image belongs to the calculated probability of each image category; respectively determine the similarity between the predicted probability and the calculated probability under each image category, and determine based on the similarity under each image category The credibility of the sample image.

In another aspect, an embodiment of the present application provides a computer-readable storage medium that stores a computer program, and the computer program includes program instructions that, when executed by a processor, cause all The processor executes the method for determining the credibility of the image.

In this embodiment of the application, the calculated probability of a sample image belonging to each image category can be determined through the image classification model, and the similarity between the predicted probability and the calculated probability under each image category can be determined respectively, and then the sample can be determined based on the similarity under each image category The credibility of the image helps to efficiently determine the credibility of the sample image.

Description of the drawings

FIG. 1 is a schematic flowchart of a method for determining image credibility according to an embodiment of the present application;

Fig. 2 is a schematic structural diagram of an image classification model according to an embodiment of the present application;

3 is a schematic flowchart of another method for determining image credibility according to an embodiment of the present application;

4 is a schematic structural diagram of an image credibility determination apparatus according to an embodiment of the present application;

FIG. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

Detailed ways

Please refer to Figure 1. This solution can be applied in the field of smart medical care to promote the construction of smart cities. It is a schematic flowchart of a method for determining image credibility in an embodiment of the present application, and the method in the embodiment of the present application may be executed by an electronic device. The method of the embodiment of the present application includes the following steps.

S101: Perform image classification processing on a sample image through each network layer in the image classification model, where each network layer includes a classification calculation layer and an anti-overfitting layer. Among them, the electronic devices involved in this application may be terminal devices such as smart phones, tablet computers, notebook computers, desktop computers, in-vehicle smart terminals, etc., and may also be servers or server groups, which are not limited in the embodiments of this application.

Wherein, the above-mentioned image classification model may be, for example, an Inception model. Illustratively, the image classification model may be as shown in FIG. 2, including an input layer (for input images), a preprocessing layer, a hidden layer, a dimensionality reduction layer, and an average. Pooling layer, anti-overfitting layer, and classification calculation layer. The model has three hidden layers, namely: Inception-A first hidden layer, Inception-B second hidden layer, and Inception-C third hidden layer Containing layer, two dimensionality reduction layers, namely: Reduction-A first dimensionality reduction layer and Reduction-B second dimensionality reduction layer; the preprocessing layer, which can be used to preprocess the input data of Inception-A. Processing can include multiple convolution and pooling of the data; the anti-overfitting layer can be used to prevent overfitting of the image classification model, effectively avoiding the image classification model from being able to classify the training images well, but in After deployment, for the actual images that need to be classified, the classification effect is poor; the classification calculation layer, its output result can be the probability that the image input through the input layer belongs to each category.

In one embodiment, before the image classification model is trained, a large number of sample images can be collected to form a sample image set. Subsequently, each sample image can be input to an image classification model, and each network layer in the image classification model can perform image classification processing on the sample image, and output an image classification result for each sample image.

In one embodiment, before training and optimizing the image classification model, a design goal may be determined, and the design goal may be a classification model for distinguishing multiple target categories. Among them, in different application scenarios, the target category may refer to different categories. For example, in a pet identification scenario, the target category may be a dog or cat, or it may be a specific cat and/or More detailed categories such as dog breeds, for example, can be subdivided categories such as shepherd dog and Shiba Inu. In the context of glomerular classification and recognition, the target category can be coarse-grained categories such as normal glomeruli and sclerotic glomeruli, or segmental sclerosis, crescent glomeruli, and sclerotic kidneys. More fine-grained categories such as small balls.

Exemplarily, suppose the above-mentioned target categories include cats and dogs. In this case, when training the image classification model, M (M is a positive integer, such as 10000) that have been selected from the sample image set can be selected in advance. The images determined as the dog category are used as the dog training images of the image classification model and P (P is a positive integer, such as 10000) images that have been determined as the cat category are used as the cat training images of the image classification model. In one embodiment, after a certain dog training image is input to the image classification model, the image classification model can extract the image feature data of the dog training image, and classify the dog training image according to the image feature data. If the output classification result indicates that the category of the dog training image is also a dog, it indicates that the classification network model is successful in classifying the dog training image. Further, after classifying the M training images that have been marked as dog categories, if the success rate is greater than the preset success rate threshold (such as 90%), it is determined that the image classification model can perform well on the images of the dog category. Classification recognition, otherwise, the parameters corresponding to each node in the image classification model can be adjusted, and the M dog training images can be classified again through the adjusted classification model. In the same way, the image classification model can be trained and optimized using P cat training images in the same way. If the final classification success rate of the dog training image and cat training image meets the preset success rate threshold, then The training of the image classification model is completed, and the trained image classification model is used as the classification model in the embodiment of the present application. In other embodiments, more different categories can be set, and the image classification model can be trained and optimized by obtaining a large number of training images of different categories, so that the finally obtained classification model can successfully classify various types of images. The rates are all higher than a certain success rate threshold.

S102: Determine the predicted probability that the sample image belongs to each image category based on the output result of the classification calculation layer.

In one embodiment, referring to the image classification model shown in FIG. 2, the output result of the classification calculation layer may be the probability that the image input through the Input layer belongs to each category. In this case, the electronic device can obtain the output of the classification calculation layer As a result, the output result of the classification calculation layer is analyzed to determine the predicted probability that the sample image belongs to each image category. Exemplarily, each image category to which the sample image belongs is predetermined. Assuming that each image category to which the sample image belongs is normal glomerulus and sclerotic glomerulus, then the sample image determined based on the output result of the classification calculation layer belongs to each The predicted probability of the image category may be, for example, 95% of normal glomeruli and 5% of sclerotic glomeruli.

S103: Obtain a feature image corresponding to the sample image output by the over-fitting layer, identify the feature image, and determine the first feature parameter and the second feature parameter corresponding to the feature image based on the recognition result.

In one embodiment, the above recognition result includes the size of the characteristic image and the value of each characteristic point in the characteristic image, and the specific implementation of the electronic device to determine the first characteristic parameter and the second characteristic parameter corresponding to the characteristic image based on the recognition result may be : Perform a summation calculation on the value of each feature point in the feature image, and determine the first feature parameter corresponding to the feature image based on the summation calculation result and the size of the feature image. Further, the difference between the value of each feature point and the first feature parameter can be determined, and the sum calculation of each difference can be performed, and then the determination is made based on the sum calculation result of the difference and the size of the feature image The second feature parameter corresponding to the feature image. Wherein, the size of the aforementioned feature image may refer to the size of the feature image, for example, 1*448.

Exemplarily, assuming that the first feature parameter is represented by μ, the second feature parameter is represented by σ, the size of the feature image is 1*448, and the value of each feature point in the feature image is _{represented by p i} , then the formula 1- 1 and Equations 1-2 respectively calculate the first characteristic parameter μ and the second characteristic parameter σ.

S104: Determine the calculated probability that the sample image belongs to each image category based on the first feature parameter and the second feature parameter.

In an embodiment, the electronic device determines the calculated probability of the sample image belonging to each image category based on the first feature parameter and the second feature parameter may be: based on a preset probability algorithm to compare the first feature parameter and the second feature parameter Perform calculations to determine the initial probability of the sample image belonging to each image category, and further, normalize each initial probability to obtain the calculated probability of the sample image belonging to each image category.

Exemplarily, each image category to which the sample image belongs may be pre-configured, and may include the first category, the second category, and the J-th category. The initial probability of the sample image belonging to each image category is expressed as p ^k (k represents the image category , K∈{0,1,...J}), the above-mentioned preset probability algorithm can be Formula 1-3. In this case, the electronic device can calculate the first feature parameter μ and the second feature parameter σ based on Equations 1-3 to determine the initial probability of the sample image belonging to each image category.

Further, normalize each initial probability p ^k (k∈{0,1,...J}) to obtain the calculated probability that the sample image belongs to each image category

Among them, the specific method of normalizing each initial probability p ^k can be normalized based on equations 1-4 to obtain the calculated probability that the sample image belongs to each image category

S105: Determine the similarity between the predicted probability and the calculated probability under each image category respectively, and determine the credibility of the sample image based on the similarity under each image category.

In one embodiment, the electronic device can compare the calculated probability of each image category

And the predicted probability, and based on the comparison result to determine the calculated probability under each image category

Based on the similarity between the predicted probability and the similarity, the credibility of the sample image is determined. Among them, the calculated probability under each image category

The higher the similarity with the predicted probability, the higher the credibility of the sample image.

Exemplarily, the probability of each image category can be calculated

Calculate the average value of the similarity with the predicted probability, and determine the obtained average value as the reliability of the sample image.

In one embodiment, after the electronic device determines the credibility of the sample image, it can obtain the credibility of the multiple sample images used to train the image classification model, and determine the credibility of the multiple sample images based on the credibility of the multiple sample images. The credibility of the image classification model obtained after training with multiple sample images.

In one embodiment, the image classification model can be trained through M sample images. In the process of training the image classification model through each sample image, the above steps S101 to S105 can be performed to determine the value of each sample image. Reliability, and store the credibility of each sample image and each sample image in a designated storage area. Subsequently, after training the image classification model through the M sample images, the electronic device can obtain the credibility of each sample image used for training the image classification model from the designated storage area, and based on the credibility of each sample image , To determine the credibility of the image classification model obtained after training through each of the above-mentioned sample images. As a feasible implementation manner, the average value of the credibility of the foregoing M sample images may be determined as the credibility of the image classification model.

In an embodiment, after the electronic device determines the credibility of the image classification model, the credibility of the image classification model may also be stored in the aforementioned designated storage area. Later, the user can obtain and view the credibility of each sample image and image classification model from the designated storage area. In this way, the interpretability of the prediction results of the image classification model can be increased, and the subsequent use of the image classification model Productization provides a strong basis, not a black box network.

In this embodiment of the application, the sample image can be classified by each network layer in the image classification model, and the predicted probability of the sample image belonging to each image category can be determined based on the output result of the classification calculation layer, and the output of the anti-overfitting layer can be obtained. The feature image corresponding to the sample image. Further, the feature image can be identified, the first feature parameter and the second feature parameter corresponding to the feature image can be determined based on the recognition result, and the calculated probability of the sample image belonging to each image category can be determined based on the first feature parameter and the second feature parameter, and respectively determine The similarity between the predicted probability and the calculated probability under each image category, and then determine the credibility of the sample image based on the similarity under each image category, which is conducive to efficiently determine the credibility of the sample image and prevent the sample image itself The problem leads to the problem of unreliable or inaccurate image classification results output by image classification models.

It should be emphasized that, in order to further ensure the privacy and security of the predicted probability and calculated probability, the predicted probability and calculated probability may also be stored in a node of a blockchain.

Please refer to FIG. 3 again, which is a schematic flowchart of another method for determining image credibility according to an embodiment of the present application. The method of the embodiment of the present application includes the following steps.

S301: Perform image classification processing on the sample image through each network layer in the image classification model, where each network layer includes a classification calculation layer and an overfitting prevention layer.

S302: Determine the predicted probability of the sample image belonging to each image category based on the output result of the classification calculation layer.

S303: Obtain a feature image corresponding to the sample image output by the over-fitting layer, identify the feature image, and determine the first feature parameter and the second feature parameter corresponding to the feature image based on the recognition result

S304: Determine the calculated probability that the sample image belongs to each image category based on the first feature parameter and the second feature parameter.

S305: Determine the similarity between the predicted probability and the calculated probability under each image category respectively, and determine the credibility of the sample image based on the similarity under each image category. For the specific implementation manners of step S301 to step S305, please refer to the related description of step S101 to step S105 in the foregoing embodiment, which will not be repeated here.

S306: Compare the credibility of the sample image with the first credibility threshold, and if the credibility of the sample image is greater than the first credibility threshold, add a credible sample label to the sample image.

In one embodiment, when sample images need to be subsequently acquired to train other image classification models, sample images carrying credible sample labels can be obtained from the sample image set, and based on the pair of sample images carrying credible sample labels Training and optimization of other image classification models is beneficial to improve the credibility and accuracy of the output results of the other image classification models.

In one embodiment, after the electronic device compares the credibility of the sample image with the first credibility threshold, if the credibility of the sample image obtained by the comparison is less than or equal to the first credibility threshold (for example, 0.5), Then the credibility of the sample image can be compared with the second credibility threshold. If the credibility of the sample image is greater than the second credibility threshold (for example, 0.3), the sample image to be reviewed is added to the sample image. , And output a review prompt message, which is used to prompt the user to review the sample image. Further, after viewing the review prompt information, the user can correct the image classification mark of the sample image, for example, correct the image classification mark from crescent glomeruli to sclerotic glomeruli.

Or, in another embodiment, if the credibility of the sample image obtained by the electronic device comparison is less than or equal to the second credibility threshold, the sample image may be deleted from the sample image set. Later, the sample image will not be used to train any image classification model.

In an embodiment, each of the above-mentioned network layers further includes a first classification layer and a second classification layer. The first classification layer is used to determine the image category of the first granularity to which the sample image belongs, and the second classification layer is used to determine the sample The image category of the second granularity to which the image belongs. The first granularity is coarser than the second granularity. After the electronic device performs image classification processing on the sample image through each network layer in the image classification model, it can also be based on the sample output from the first classification layer. The classification result of the image and the classification result of the sample image output by the second classification layer determine the target image category to which the sample image belongs. In this way, it is beneficial to improve the recognition accuracy of the image classification model.

Exemplarily, in order to improve the recognition accuracy of the image classification model corresponding to Figure 2, Inception can be improved by adding two network layers, namely Gather1 (i.e., the first classification layer) and Gather2 (i.e., the second classification layer). Gather1 is added after stem in Figure 2, and Gather2 is added after Inception-C in Figure 2. Among them, Gather1 is used to classify the more easily distinguishable categories, and Gather2 is used to classify the more difficult categories. Taking the identification of glomerular types as an example, Gather1 can be used to distinguish between normal glomeruli and sclerotic glomeruli and other easily distinguishable categories, and Gather2 can be used to distinguish segmental sclerosis, crescent glomeruli and sclerotic glomeruli. Etc. which are more difficult to distinguish. Further, the electronic device can merge the classification result of the sample image by Gather1 and the classification result of the sample image by Gather2, and output the final image classification and recognition result, thereby improving the recognition accuracy of the image classification model.

Among them, as a feasible implementation, when performing image recognition through the improved image classification model, the output data of Gather1 can be extracted. If the classification result of Gather1 is judged to be more accurate based on the output data, there is no need to call Gather2 for image recognition. More granular analysis. In this way, the computational overhead of the image classification model can be reduced, and the recognition efficiency of the image classification model can be improved.

In the embodiment of this application, the sample image can be classified by each network layer in the image classification model, and the predicted probability of the sample image belonging to each image category is determined based on the output result of the classification calculation layer, and the output of the anti-overfitting layer is obtained. The feature image corresponding to the sample image. Further, the feature image can be identified, the first feature parameter and the second feature parameter corresponding to the feature image can be determined based on the recognition result, and the calculated probability of the sample image belonging to each image category can be determined based on the first feature parameter and the second feature parameter, and respectively determine The similarity between the predicted probability and the calculated probability under each image category, and then determine the credibility of the sample image based on the similarity under each image category, and compare the credibility of the sample image with the first credibility threshold, If the credibility of the sample image obtained by comparison is greater than the first credibility threshold, add credible sample labels to the sample image so that other image classification models can be trained and optimized directly based on the sample image with the credible sample label. .

The embodiment of the present application also provides a computer storage medium, the computer storage medium stores program instructions, and when the program instructions are executed, they are used to implement the corresponding methods described in the foregoing embodiments. Optionally, the computer-readable storage medium may be non-volatile or volatile.

Please refer to FIG. 4 again, which is a schematic structural diagram of an image credibility determination apparatus according to an embodiment of the present application.

In an implementation manner of the device in the embodiment of the present application, the device includes the following structure.

The processing module 40 is configured to perform image classification processing on the sample image through each network layer in the image classification model, and each network layer includes a classification calculation layer and an anti-overfitting layer;

The processing module 40 is further configured to determine the predicted probability that the sample image belongs to each image category based on the output result of the classification calculation layer;

The obtaining module 41 is configured to obtain a feature image corresponding to the sample image output by the over-fitting prevention layer;

The processing module 40 is further configured to identify the characteristic image, and determine the first characteristic parameter and the second characteristic parameter corresponding to the characteristic image based on the recognition result;

The processing module 40 is further configured to determine the calculated probability that the sample image belongs to each image category based on the first characteristic parameter and the second characteristic parameter;

The processing module 40 is further configured to determine the similarity between the predicted probability and the calculated probability under each image category, and determine the sample based on the similarity under each image category The credibility of the image.

In one embodiment, the recognition result includes the size of the feature image and the value of each feature point in the feature image, and the processing module 40 is specifically configured to perform a calculation of each feature point in the feature image. The value is summed and calculated, and the first characteristic parameter corresponding to the characteristic image is determined based on the result of the summation calculation and the size of the characteristic image; the difference between the value of each characteristic point and the first characteristic parameter is determined , Performing a summation calculation for each of the difference values, and determining a second characteristic parameter corresponding to the characteristic image based on the summation calculation result for the difference values and the size of the characteristic image.

In one embodiment, the processing module 40 is further specifically configured to calculate the first characteristic parameter and the second characteristic parameter based on a preset probability algorithm, and determine that the sample image belongs to each image category. Initial probability; normalizing each of the initial probabilities to obtain the calculated probability that the sample image belongs to each image category.

In one embodiment, the acquiring module 41 is further configured to acquire the credibility of a plurality of sample images used for training the image classification model; the processing module 40 is also configured to acquire credibility based on the plurality of The credibility of the sample image determines the credibility of the image classification model obtained after training on the multiple sample images.

In an embodiment, the processing module 40 is further configured to compare the credibility of the sample image with a first credibility threshold; if the comparison shows that the credibility of the sample image is greater than the first credibility For the credibility threshold, a credible sample label is added to the sample image.

In one embodiment, the processing module 40 is further configured to compare the credibility of the sample image with the first credibility threshold if the credibility of the sample image is less than or equal to the first credibility threshold. The two credibility thresholds are compared; if the credibility of the sample image is greater than the second credibility threshold, the sample image to be reviewed is added to the sample image, and the review prompt information is output, and the review The prompt information is used to prompt the user to review the sample image; if the credibility of the sample image obtained by comparison is less than or equal to the second credibility threshold, the sample image is deleted from the sample image set.

In an embodiment, each of the network layers further includes a first classification layer and a second classification layer, the first classification layer is used to determine the image category of the first granularity to which the sample image belongs, and the second classification The layer is used to determine the image category of the second granularity to which the sample image belongs, and the first granularity is coarser than the second granularity. The processing module 40 is also used to determine the image category based on the output of the first classification layer. The classification result of the sample image and the classification result of the sample image output by the second classification layer determine the target image category to which the sample image belongs.

Please refer to FIG. 5 again, which is a schematic structural diagram of an electronic device in an embodiment of the present application. The electronic device in an embodiment of the present application includes a power supply module and other structures, and includes a processor 501, a storage device 502, and a communication interface 503. The processor 501, the storage device 502, and the communication interface 503 can exchange data, and the processor 501 implements the corresponding image credibility determination function.

The storage device 502 may include a volatile memory (volatile memory), such as random-access memory (RAM); the storage device 502 may also include a non-volatile memory (non-volatile memory), such as fast Flash memory (flash memory), solid-state drive (SSD), etc.; the storage device 502 may also include a combination of the foregoing types of memories.

The processor 501 may be a central processing unit (CPU) 501. In an embodiment, the processor 501 may also be a graphics processor 501 (Graphics Processing Unit, GPU). The processor 501 may also be a combination of a CPU and a GPU. The electronic device may include multiple CPUs and GPUs as needed to determine the credibility of the corresponding image. In one embodiment, the storage device 502 is used to store program instructions. The processor 501 can call the program instructions to implement various methods mentioned above in the embodiments of the present application.

In a first possible implementation manner, the processor 501 of the electronic device calls the program instructions stored in the storage device 502 for image classification of the sample image through each network layer in the image classification model Processing, the various network layers include a classification calculation layer and an anti-overfitting layer; determine the predicted probability that the sample image belongs to each image category based on the output result of the classification calculation layer; obtain the output of the anti-overfitting layer The characteristic image corresponding to the sample image; identifying the characteristic image, and determining the first characteristic parameter and the second characteristic parameter corresponding to the characteristic image based on the recognition result; based on the first characteristic parameter and the second characteristic parameter Determine the calculated probability that the sample image belongs to each image category; determine the similarity between the predicted probability and the calculated probability under each image category, and determine the similarity based on the calculated probability under each image category The similarity determines the credibility of the sample image.

In one embodiment, the recognition result includes the size of the feature image and the value of each feature point in the feature image, and the processor 501 is specifically configured to perform a calculation of each feature point in the feature image. The value is summed and calculated, and the first characteristic parameter corresponding to the characteristic image is determined based on the result of the summation calculation and the size of the characteristic image; the difference between the value of each characteristic point and the first characteristic parameter is determined , Performing a summation calculation for each of the difference values, and determining a second characteristic parameter corresponding to the characteristic image based on the summation calculation result for the difference values and the size of the characteristic image.

In one embodiment, the processor 501 is further specifically configured to calculate the first characteristic parameter and the second characteristic parameter based on a preset probability algorithm, and determine that the sample image belongs to the image category. Initial probability; normalizing each of the initial probabilities to obtain the calculated probability that the sample image belongs to each image category.

In one embodiment, the processor 501 is further configured to obtain the credibility of a plurality of sample images used for training the image classification model, and based on the credibility of the plurality of sample images, determine the pass The credibility of the image classification model obtained after the multiple sample images are trained.

In one embodiment, the processor 501 is further configured to compare the credibility of the sample image with a first credibility threshold; if the comparison shows that the credibility of the sample image is greater than the first credibility A degree threshold, a credible sample label is added to the sample image.

In one embodiment, the processor 501 is further configured to compare the credibility of the sample image with the second credibility if the credibility of the sample image is less than or equal to the first credibility threshold. The reliability threshold is compared; if the credibility of the sample image is greater than the second credibility threshold, the sample image to be reviewed is tagged with the sample image, and the review prompt information is output, the review prompt information It is used to prompt the user to review the sample image; if the credibility of the sample image obtained by comparison is less than or equal to the second credibility threshold, the sample image is deleted from the sample image set.

In an embodiment, each of the network layers further includes a first classification layer and a second classification layer, the first classification layer is used to determine the image category of the first granularity to which the sample image belongs, and the second classification The layer is used to determine the image category of the second granularity to which the sample image belongs, and the first granularity is coarser than the second granularity. The classification result of the sample image and the classification result of the sample image output by the second classification layer determine the target image category to which the sample image belongs.

Reference may be made to the description of related content in the embodiments corresponding to the foregoing drawings.

A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing relevant hardware through a computer program. The program can be stored in a computer readable storage medium, and the program can be stored in a computer readable storage medium. During execution, it may include the procedures of the above-mentioned method embodiments. Wherein, the storage medium may be a magnetic disk, an optical disc, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM), etc.

Further, the computer usable storage medium may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required by at least one function, etc.; the storage data area may store a block chain node Use the created data, etc.

The blockchain referred to in this application is a new application mode of computer technology such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm. Blockchain, essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information for verification. The validity of the information (anti-counterfeiting) and the generation of the next block. The blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.

The above-disclosed are only part of the embodiments of this application. Of course, it cannot be used to limit the scope of rights of this application. Those of ordinary skill in the art can understand all or part of the procedures for implementing the above-mentioned embodiments and make them in accordance with the claims of this application. The equivalent changes of is still within the scope of the invention.

Claims

A method for determining the credibility of an image, wherein the method includes:

Image classification processing is performed on the sample image through each network layer in the image classification model, and each network layer includes a classification calculation layer and an anti-overfitting layer;

Determining the predicted probability that the sample image belongs to each image category based on the output result of the classification calculation layer;

Acquiring a feature image corresponding to the sample image output by the over-fitting prevention layer;

Identifying the characteristic image, and determining the first characteristic parameter and the second characteristic parameter corresponding to the characteristic image based on the recognition result;

Determining, based on the first feature parameter and the second feature parameter, the calculated probability that the sample image belongs to each image category;

The similarity between the predicted probability and the calculated probability under the respective image categories is determined respectively, and the credibility of the sample image is determined based on the similarity under the respective image categories.
The method according to claim 1, wherein the recognition result includes the size of the characteristic image and the value of each characteristic point in the characteristic image, and the first characteristic corresponding to the characteristic image is determined based on the recognition result Parameters and second characteristic parameters, including:

Performing a summation calculation on the value of each characteristic point in the characteristic image, and determining the first characteristic parameter corresponding to the characteristic image based on the sum calculation result and the size of the characteristic image;

Determine the difference between the value of each feature point and the first feature parameter;

A summation calculation is performed on each of the differences, and a second characteristic parameter corresponding to the characteristic image is determined based on the sum calculation result for the difference and the size of the characteristic image.
The method according to claim 2, wherein the determining the calculated probability of the sample image belonging to the respective image category based on the first characteristic parameter and the second characteristic parameter comprises:

Calculating the first feature parameter and the second feature parameter based on a preset probability algorithm, and determining the initial probability that the sample image belongs to each image category;

Perform normalization processing on each of the initial probabilities to obtain the calculated probability that the sample image belongs to each image category.
The method according to claim 1, wherein after said determining the credibility of the sample image, the method further comprises:

Acquiring the credibility of a plurality of sample images used for training the image classification model;

Based on the credibility of the plurality of sample images, the credibility of the image classification model obtained after training through the plurality of sample images is determined.
The method according to claim 1, wherein after the determining the credibility of the sample image based on the similarity in the respective image categories, the method further comprises:

Comparing the credibility of the sample image with the first credibility threshold;

If the credibility of the sample image obtained by comparison is greater than the first credibility threshold, a credible sample label is added to the sample image.
The method according to claim 5, wherein after the comparing the credibility of the sample image with a first credibility threshold, the method further comprises:

If the comparison shows that the credibility of the sample image is less than or equal to the first credibility threshold, then the credibility of the sample image is compared with the second credibility threshold;

If the comparison shows that the credibility of the sample image is greater than the second credibility threshold, add a sample tag to be reviewed to the sample image, and output a review prompt message, which is used to prompt the user to check The sample image is reviewed;

If the credibility of the sample image is less than or equal to the second credibility threshold, the sample image is deleted from the sample image set.
The method according to claim 1, wherein the respective network layers further comprise a first classification layer and a second classification layer, the first classification layer is used to determine the image category of the first granularity to which the sample image belongs, The second classification layer is used to determine the image category of the second granularity to which the sample image belongs, the first granularity is coarser than the second granularity, and the sample image is processed by each network layer in the image classification model. After the image classification processing, the method further includes:

Based on the classification result for the sample image output by the first classification layer and the classification result for the sample image output by the second classification layer, the target image category to which the sample image belongs is determined.
An image credibility determination device, which includes:

The processing module is configured to perform image classification processing on the sample image through each network layer in the image classification model, and each network layer includes a classification calculation layer and an anti-overfitting layer;

The processing module is further configured to determine the predicted probability that the sample image belongs to each image category based on the output result of the classification calculation layer;

An acquisition module, configured to acquire a feature image corresponding to the sample image output by the over-fitting prevention layer;

The processing module is further configured to identify the characteristic image, and determine the first characteristic parameter and the second characteristic parameter corresponding to the characteristic image based on the recognition result;

The processing module is further configured to determine the calculated probability that the sample image belongs to each image category based on the first characteristic parameter and the second characteristic parameter;

The processing module is further configured to determine the similarity between the predicted probability and the calculated probability under each image category, and determine the sample image based on the similarity under each image category Credibility.
An electronic device, including a processor, a storage device, and a communication interface, the processor, the storage device, and the communication interface are connected to each other, wherein the storage device is used to store computer program instructions, and the processing The device is configured to call the program instructions for performing image classification processing on the sample image through each network layer in the image classification model. The network layers include a classification calculation layer and an overfitting prevention layer; based on the classification calculation layer The output result of determining the predicted probability that the sample image belongs to each image category; acquiring the characteristic image corresponding to the sample image output by the over-fitting layer; identifying the characteristic image, and determining the characteristic image based on the recognition result Corresponding first feature parameters and second feature parameters; determine the calculated probability that the sample image belongs to each image category based on the first feature parameter and the second feature parameter; determine the calculated probability of each image category The similarity between the predicted probability and the calculated probability, and the credibility of the sample image is determined based on the similarity in each image category.
The electronic device according to claim 9, wherein the recognition result includes the size of the characteristic image and the value of each characteristic point in the characteristic image, and the processor is specifically configured to compare the size of the characteristic image The value of each feature point is summed and calculated, and the first feature parameter corresponding to the feature image is determined based on the result of the summation calculation and the size of the feature image; the value of each feature point and the first feature parameter are determined A summation calculation is performed on each of the differences, and a second characteristic parameter corresponding to the characteristic image is determined based on the sum calculation result for the difference and the size of the characteristic image.
The electronic device according to claim 10, wherein the processor is further specifically configured to calculate the first characteristic parameter and the second characteristic parameter based on a preset probability algorithm, and determine that the sample image belongs to all The initial probability of each image category; normalization processing is performed on each of the initial probabilities to obtain the calculated probability that the sample image belongs to each image category.
The electronic device according to claim 9, wherein the processor is further configured to obtain the credibility of a plurality of sample images used for training the image classification model; based on the reliability of the plurality of sample images The reliability determines the reliability of the image classification model obtained after training through the multiple sample images.
The electronic device according to claim 9, wherein the processor is further configured to compare the credibility of the sample image with a first credibility threshold; if the comparison shows that the credibility of the sample image is greater than For the first credibility threshold, a credible sample label is added to the sample image.
The electronic device according to claim 13, wherein the processor is further configured to: if the credibility of the sample image obtained by comparison is less than or equal to the first credibility threshold, the The credibility is compared with the second credibility threshold; if the comparison shows that the credibility of the sample image is greater than the second credibility threshold, add a sample label to be reviewed to the sample image, and output a review prompt Information, the review prompt information is used to prompt the user to review the sample image; if the credibility of the sample image is less than or equal to the second credibility threshold, the sample image is removed from the sample image. Deleted from the image collection.
The electronic device according to claim 9, wherein the respective network layers further comprise a first classification layer and a second classification layer, and the first classification layer is used to determine the image category of the first granularity to which the sample image belongs The second classification layer is used to determine the image category of the second granularity to which the sample image belongs, and the processor is also used to determine the classification result and the result of the sample image based on the output of the first classification layer. The classification result for the sample image output by the second classification layer determines the target image category to which the sample image belongs.
A computer-readable storage medium, wherein computer program instructions are stored in the computer-readable storage medium, and when the computer program instructions are executed by a processor, the following steps are implemented:

Image classification processing is performed on the sample image through each network layer in the image classification model, and each network layer includes a classification calculation layer and an anti-overfitting layer;

Determining the predicted probability that the sample image belongs to each image category based on the output result of the classification calculation layer;

Acquiring a feature image corresponding to the sample image output by the over-fitting prevention layer;

Identifying the characteristic image, and determining the first characteristic parameter and the second characteristic parameter corresponding to the characteristic image based on the recognition result;

Determining, based on the first feature parameter and the second feature parameter, the calculated probability that the sample image belongs to each image category;

The similarity between the predicted probability and the calculated probability under the respective image categories is determined respectively, and the credibility of the sample image is determined based on the similarity under the respective image categories.
The computer-readable storage medium according to claim 16, wherein the recognition result includes the size of the characteristic image and the value of each characteristic point in the characteristic image, and the determination that the characteristic image corresponds to the characteristic image based on the recognition result For the first characteristic parameter and the second characteristic parameter, the following steps are implemented:

Performing a summation calculation on the value of each characteristic point in the characteristic image, and determining the first characteristic parameter corresponding to the characteristic image based on the sum calculation result and the size of the characteristic image;

Determine the difference between the value of each feature point and the first feature parameter;

A summation calculation is performed on each of the differences, and a second characteristic parameter corresponding to the characteristic image is determined based on the sum calculation result for the difference and the size of the characteristic image.
The computer-readable storage medium according to claim 17, wherein when the calculated probability of the sample image belonging to the respective image category is determined based on the first characteristic parameter and the second characteristic parameter, the following steps are implemented :

Calculating the first feature parameter and the second feature parameter based on a preset probability algorithm, and determining the initial probability that the sample image belongs to each image category;

Perform normalization processing on each of the initial probabilities to obtain the calculated probability that the sample image belongs to each image category.
The computer-readable storage medium according to claim 16, wherein after said determining the credibility of the sample image, the following steps are further implemented:

Acquiring the credibility of a plurality of sample images used for training the image classification model;

Based on the credibility of the plurality of sample images, the credibility of the image classification model obtained after training through the plurality of sample images is determined.
The computer-readable storage medium according to claim 16, wherein after the credibility of the sample image is determined based on the similarity in the respective image categories, the following steps are further implemented:

Comparing the credibility of the sample image with the first credibility threshold;

If the credibility of the sample image obtained by comparison is greater than the first credibility threshold, a credible sample label is added to the sample image.