WO2023184918A1

WO2023184918A1 - Image anomaly detection method, apparatus and system, and readable storage medium

Info

Publication number: WO2023184918A1
Application number: PCT/CN2022/123403
Authority: WO
Inventors: 赵冰
Original assignee: 苏州浪潮智能科技有限公司
Priority date: 2022-03-31
Filing date: 2022-09-30
Publication date: 2023-10-05
Also published as: CN114627102A; CN114627102B

Abstract

Disclosed in the present application are an image anomaly detection method, apparatus and system, and a non-volatile readable storage medium, which are applied to the technical field of images. With regard to the problem of the accuracy of image anomaly detection being low, the method is provided, which comprises: by using a pre-established image anomaly detection model, detecting an image to be detected, so as to obtain an anomaly score; and said image being an anomalous image when the anomaly score meets a preset condition, wherein the image anomaly detection model is obtained by means of training on the basis of a pre-established sample set, and simulated anomalous samples in the sample set are obtained on the basis of an original anomalous sample and an original normal sample. Therefore, the present application can improve the accuracy of image anomaly detection during use.

Description

An image anomaly detection method, device, system and readable storage medium

This application requires the priority of the Chinese patent application submitted to the China Patent Office on March 31, 2022, with the application number 202210331065.9, and the application title is "an image anomaly detection method, device, system and readable storage medium", all of which The contents are incorporated into this application by reference.

Technical field

The present application relates to the field of image technology, and in particular to an image anomaly detection method, device, system and non-volatile readable storage medium.

Background technique

Anomaly detection in the image field refers to identifying images that may contain abnormalities relative to normal images. Anomaly detection tasks generally assume that normal images have a large amount of data, while abnormal data have a small amount of data. It is impossible to collect enough abnormal data to carry out supervised learning to distinguish the two. Therefore, anomaly detection is often considered to comply with the single-class learning hypothesis, that is, the detection of abnormal samples is achieved only by learning from normal samples. Anomaly detection is widely used in practice. For example, the analysis of surveillance videos in highway scenarios is a very potential application scenario. The amount of surveillance video data in highway scenarios is very large, and there are various potential abnormal behaviors. , however, due to the small probability of occurrence of anomalies and the large differences between various abnormal behaviors and the different occurrence probabilities, it is impossible to collect reasonable data sets for supervised learning. Solve the problem of highway surveillance video through unsupervised, especially self-supervised learning. The application prospects of anomaly detection tasks in scenarios are huge.

Existing anomaly detection methods based on self-supervised learning usually obtain simulated abnormal images by performing data enhancement on normal images. However, this method can only further enhance the similarity between samples and normal samples, and cannot achieve a good result. Simulating abnormal samples affects the recognition accuracy of the detection model trained based on these images.

Contents of the invention

The purpose of the embodiments of the present application is to provide an image anomaly detection method, device, system and non-volatile readable storage medium, which can improve the accuracy of image anomaly detection during use.

In order to solve the above technical problems, embodiments of the present application provide an image anomaly detection method, including:

Use a pre-established image anomaly detection model to detect the image to be detected and obtain an anomaly score;

When the abnormality score meets the preset conditions, the image to be detected is an abnormal image; where:

The image anomaly detection model is trained based on a pre-established sample set, and the simulated abnormal samples in the sample set are obtained based on original abnormal samples and original normal samples.

Optionally, the pre-established image anomaly detection model is used to detect the image to be detected, and the anomaly score obtained includes:

Create a sample set;

The image anomaly detection model is obtained by training based on the sample set.

Optionally, the created sample set includes:

Get a preset number of videos;

Convert the video into an image by extracting video frames;

Classify the image according to the frame rate to obtain original normal samples and original abnormal samples;

Construct simulated abnormal samples based on the original normal samples and the original abnormal samples;

The sample set is established based on the original normal samples, the original abnormal samples and the simulated abnormal samples.

Optionally, the simulated abnormal samples in the sample set are obtained based on the original abnormal samples and the original normal samples, including:

Select the target original abnormal sample from each original abnormal sample, and select the target original normal sample from each original normal sample;

Obtaining a first replica image corresponding to the target area in the target original abnormal sample, and a second replica image corresponding to the target area in the target original normal sample;

Paste the first copied image and the second copied image onto the target original normal sample to obtain a simulated abnormal sample.

Optionally, pasting the first copied image and the second copied image onto the target original normal sample to obtain a simulated abnormal sample includes:

Paste the first copied image and the second copied image to a preset area on the target original normal sample; the preset area is an area centered on the center point of the target original normal sample;

The first replicated image and the second replicated image are respectively scaled to obtain simulated abnormal samples.

Optionally, the shape of the preset area is any one of the following shape areas:

circular area and rectangular area.

Optionally, scaling the first copied image and the second copied image respectively includes:

The copied image and the second copied image are respectively randomly deformed and scaled according to different aspect ratios.

Optionally, selecting the target original abnormal sample from each original abnormal sample, and selecting the target original normal sample from each original normal sample includes:

The target original abnormal sample is randomly selected from each original abnormal sample, and the target original normal sample is randomly selected from each original normal sample.

Optionally, the image anomaly detection model is trained based on a pre-established sample set and includes:

Use a pre-established sample set to train the feature extraction network to obtain the optimal weight parameters of the feature extraction network;

Using each original normal sample in the sample set and the feature extraction network based on the optimal weight parameters to train the classifier to obtain the optimal parameters of the classifier;

An image anomaly detection model is constructed based on the feature extraction network of the optimal weight parameters and the classifier of the optimal parameters.

Optionally, the classifier is a KDE classifier.

Optionally, the feature extraction network is trained using a pre-established sample set, and the optimal weight parameters of the feature extraction network include:

For each sample in the sample set, perform data enhancement processing on the sample to obtain a first enhanced sample and a second enhanced sample;

The first enhanced sample is processed through the first feature extraction network to obtain the first encoding vector corresponding to the first enhanced sample, and the second enhanced sample is processed through the second feature extraction network to obtain the first encoding vector corresponding to the first enhanced sample. The second encoding vector corresponding to the second enhanced sample; the first feature extraction network and the second feature extraction network are the same;

Using a first projection network to process the first encoding vector to obtain a first projection vector, using a second projection network to process the second encoding vector to obtain a second projection vector; the first projection network and the The second projection network is the same;

Calculate the contrast loss of the first projection vector and the second projection vector;

Based on the contrast loss corresponding to each sample, the total contrast loss is calculated;

Determine whether the total comparative loss meets the ending condition;

In response to the total contrast loss satisfying the end condition, the method ends, and the optimal weight parameters of the first feature extraction network and the second feature extraction network are obtained.

Optionally, the contrast loss is calculated through the following formula (1):

In the above formula (1),

represents the distribution to which x and x ⁺ belong, x represents the sample, x ⁺ represents the data enhancement of sample x,

Represents the data enhancement of negative samples, p represents distribution, f(x) ^T represents the transposition after extracting features for x, f(x ⁺ ) represents extracting features for x ⁺ ,

expresses right

Extract features, i represents the sample label, and N represents the total sample size.

Optionally, after determining whether the total contrast loss meets the end condition, the method further includes:

In response to the total contrast loss not meeting the end condition, the weight parameters of the first feature extraction network and the second feature extraction network are updated, and the next round of training is entered.

Optionally, when the anomaly score meets a preset condition, the image to be detected is an abnormal image, including:

When the abnormality score is greater than a preset threshold, the image to be detected is determined to be an abnormal image.

Optionally, after the pre-established image anomaly detection model is used to detect the image to be detected and the anomaly score is obtained, it also includes:

When the abnormality score is less than or equal to the threshold, the image to be detected is determined to be a normal image.

An embodiment of the present application also provides an image anomaly detection device, including:

The detection module is used to detect the image to be detected using a pre-established image anomaly detection model and obtain an anomaly score;

An analysis module configured to determine that the image to be detected is an abnormal image when the abnormality score satisfies a preset condition; wherein:

Embodiments of the present application also provide an image anomaly detection system, including:

Memory, used to store computer programs;

A processor, configured to implement the steps of the above-mentioned image anomaly detection method when executing the computer program.

Embodiments of the present application also provide a non-volatile readable storage medium. A computer program is stored on the non-volatile readable storage medium. When the computer program is executed by a processor, the above-mentioned image abnormality is realized. Steps of the detection method.

An embodiment of the present application also provides a computing processing device, including:

A memory having computer readable code stored therein;

One or more processors, when the computer readable code is executed by the one or more processors, the computing processing device performs the steps of the image anomaly detection method described above.

Embodiments of the present application also provide a computer program product, which includes computer readable code. When the computer readable code is run on a computing processing device, it causes the computing processing device to execute the above-mentioned image anomaly detection method. step.

Embodiments of the present application provide an image anomaly detection method, device, system and non-volatile readable storage medium. The method includes: using a pre-established image anomaly detection model to detect the image to be detected, and obtaining an anomaly score; When the anomaly score meets the preset conditions, the image to be detected is an abnormal image; where: the image anomaly detection model is trained based on a pre-established sample set, and the simulated abnormal samples in the sample set are obtained based on the original abnormal samples and the original normal samples. of.

It can be seen that in this application, simulated abnormal samples are obtained based on original abnormal samples and original normal samples, and the image anomaly detection model is trained based on the sample set including simulated abnormal samples, original normal samples and original abnormal samples, which improves the accuracy of the image anomaly detection model. , and then use the image anomaly detection model to detect the image to be detected, obtain the anomaly score, and when the anomaly score meets the preset conditions, determine that the image to be detected is an abnormal image, thereby realizing image anomaly detection; during the use of this application It can improve the accuracy of image anomaly detection.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the prior art and the drawings required to be used in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some of the drawings of the present application. Embodiments, for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without exerting creative efforts.

Figure 1 is a schematic flow chart of an image anomaly detection method provided by an embodiment of the present application;

Figure 2 is a schematic diagram of a simulated abnormal sample acquisition process provided by an embodiment of the present application;

Figure 3 is a schematic diagram of an abnormal score acquisition process provided by an embodiment of the present application;

Figure 4 is a schematic diagram of a comparative loss acquisition process provided by an embodiment of the present application;

Figure 5 is a schematic structural diagram of an image anomaly detection device provided by an embodiment of the present application;

Figure 6 schematically illustrates a block diagram of a computing processing device for performing a method according to the present application; and

Figure 7 schematically shows a storage unit for holding or carrying program code for implementing the method according to the present application.

Detailed ways

Embodiments of the present application provide an image anomaly detection method, device, system and computer-readable storage medium, which can improve image anomaly detection accuracy during use.

In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments These are part of the embodiments of this application, but not all of them. Based on the embodiments in this application, other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application.

Please refer to FIG. 1 , which is a schematic flow chart of an image anomaly detection method provided by an embodiment of the present application. The method includes:

S110: Use the pre-established image anomaly detection model to detect the image to be detected and obtain an anomaly score;

It should be noted that in the embodiment of the present application, a sample set can be established first, and then the image anomaly detection model can be trained based on the sample set. The amount of surveillance video data in the highway scene is very large, the probability of abnormal occurrence is small, and the differences between various abnormal behaviors are large and the probability of occurrence is different, which makes it impossible to collect reasonable data sets. Therefore, simulated abnormal samples can be constructed to Increase the number of abnormal samples in the sample set.

Specifically, the simulated abnormal samples in the sample set in the embodiment of the present application are obtained based on the original abnormal samples and the original normal samples. In practical applications, for highway monitoring scenarios, massive monitoring videos can be collected in advance and extracted by video frames. Convert the surveillance video into an image, and label the image as normal/abnormal according to the frame rate, that is, label each image with a category. Specifically, you can add a category label, and the category is normal or abnormal, so as to obtain each original normal sample and each original normal samples, and then construct simulated abnormal samples based on the original abnormal samples and original normal samples, thereby obtaining a sample set composed of original normal samples, original abnormal samples and simulated abnormal samples, and train the image anomaly detection model based on this sample set.

In practical applications, the image to be detected is collected, and the image anomaly detection model is used to detect anomalies in the image to be detected, and the anomaly score is obtained.

S120: When the abnormality score meets the preset conditions, the image to be detected is an abnormal image.

Specifically, in the embodiment of the present application, it can be further determined whether the preset conditions are met based on the abnormality. If the preset conditions are met, the image to be detected is determined to be an abnormal image. For example, a threshold can be set in advance. When the anomaly score is less than or equal to the threshold, the image to be detected can be determined to be a normal image. When the anomaly score is greater than the threshold, the image to be detected can be determined to be an abnormal image.

Furthermore, the simulated abnormal samples in the above sample set are obtained based on the original abnormal samples and the original normal samples, and may specifically include:

It can be understood that in the embodiment of the present application, the target original abnormal sample can be randomly selected from each original abnormal sample, and the target original normal sample can also be randomly selected from each original normal sample, such as the target original abnormal sample in Figure 2 For A and the target original normal sample is B, (randomly) select a target area (specifically, it can be a rectangular area) from the target original abnormal sample, copy the target area, obtain the first copied image a, and then select the target original normal sample from (Randomly) select a target area on the sample and copy the target area to obtain the second copy image b. Paste both the first copy image a and the second copy image b onto the target original normal sample to obtain the simulation Abnormal samples.

Furthermore, the first copied image and the second copied image are pasted onto the target original normal sample to obtain the process of simulating the abnormal sample, which may include:

It should be noted that in the embodiment of the present application, a preset area (border c in Figure 2) can be selected in advance from the target original normal sample. Specifically, the preset area can be centered on the center point of the target original normal sample. area, the shape of the preset area can be a circular area or a rectangular area, and the first copied image and the second copied image are pasted to the preset area on the target original normal sample, and then the two pasted The images are scaled accordingly. Specifically, random deformation and scaling of different length-to-width ratios can be performed. For example, deformation and scaling can be randomly performed according to ratios of 1:2, 1:3, or 1:4. After the scaling is completed, the deformed and scaled images a are obtained. ' and b', thereby obtaining the final simulated anomaly sample.

Furthermore, the above image anomaly detection model is trained based on a pre-established sample set, which may include:

Use the pre-established sample set to train the feature extraction network and obtain the optimal weight parameters of the feature extraction network;

The classifier is trained using each original normal sample in the sample set and the feature extraction network based on the optimal weight parameters to obtain the optimal parameters of the classifier;

An image anomaly detection model is constructed based on the feature extraction network with optimal weight parameters and the classifier with optimal parameters.

Specifically, in practical applications, the above-mentioned pre-established sample sets can be used to train the feature extraction network to obtain the optimal weight parameters of the feature extraction network, and then the feature extraction network is fixed based on the optimal weight network parameters. Please refer to the figure. 3. The feature extraction network is followed by a classifier, and then each original normal sample in the sample set is sequentially input to the feature extraction network with fixed weight parameters for feature extraction. The output of the feature extraction network is used as the input of the classifier to obtain the anomaly score. When the anomaly score does not meet the preset requirements, the parameters of the classifier are updated and the next round of training is entered. Specifically, the number of training times can be preset. When the number of training times is reached, the training is stopped and the optimal parameters of the classifier are obtained. The feature extraction network with optimal weight parameters and the classifier with optimal parameters construct an image anomaly detection model.

Among them, the classifier in the embodiment of the present application can specifically be a KDE (kernel density estimation, kernel density estimation) classifier, where KDE is used to estimate unknown density functions in probability theory, and the output result of the KDE classifier can be used as the current The degree of difference between the sample and the kernel density. The smaller the difference, the samples belong to the same distribution. The larger the difference, the samples belong to different distributions. That is to say, the image anomaly detection model in the embodiment of the present application finally outputs an anomaly score, and further determines whether the image to be detected is an abnormal image through the anomaly score and the preset threshold.

Among them, KDE is calculated as follows:

in,

Represents the overall probability density function, n represents the total number of normal samples, j represents the jth normal sample, x represents a specific distribution sample, x _j represents the jth real sample point, K(.) represents the kernel function, and h represents the bandwidth. Among them, the Gaussian distribution probability density function can be selected as the kernel function. The Gaussian distribution probability density function is as follows:

Among them, σ represents the standard deviation, e represents the natural logarithm, and μ represents the expectation.

Furthermore, the above-mentioned process of using a pre-established sample set to train the feature extraction network and obtain the optimal weight parameters of the feature extraction network may include:

For each sample in the sample set, perform data enhancement processing on the sample to obtain the first enhanced sample and the second enhanced sample;

The first enhanced sample is processed through the first feature extraction network to obtain the first encoding vector corresponding to the first enhanced sample, and the second enhanced sample is processed through the second feature extraction network to obtain the third encoding vector corresponding to the second enhanced sample. Two encoding vectors; the first feature extraction network and the second feature extraction network are the same;

The first projection network is used to process the first encoding vector to obtain the first projection vector, and the second projection network is used to process the second encoding vector to obtain the second projection vector; the first projection network and the second projection network are the same;

Calculate the contrast loss between the first projection vector and the second projection vector;

Determine whether the total comparative loss meets the ending conditions;

When the total contrast loss meets the end condition, it ends, and the optimal weight parameters of the first feature extraction network and the second feature extraction network are obtained;

When the total comparison loss does not meet the end condition, the weight parameters of the first feature extraction network and the second feature extraction network are updated, and the next round of training is entered.

It should be noted that, as shown in Figure 4, comparative SSL (Self-supervised learning, self-supervised learning) training can be performed on the samples in the sample set. SSL training is a special unsupervised learning method. By setting The agent task automatically generates labels for unlabeled images, thereby achieving feature learning for unlabeled data. Specifically, in the embodiment of the present application, data enhancement processing is performed on each sample X to obtain the first enhanced sample v and the second enhanced sample v'. The first enhanced sample v is input into the first feature extraction network f to obtain the first Encoding vector y, input the second enhanced sample v' into the second feature extraction network f' to obtain the second encoding vector y' (where f and f' are exactly the same), input the first encoding vector y into the first Projection processing is performed in the projection network g to obtain the first projection vector z, and the second encoding vector y' is input to the second projection network g' for projection processing to obtain the second projection vector z' (where g and g' are completely Same), and calculate the contrast loss of the first projection vector z and the second projection vector z'. A contrast loss is obtained for each sample. The total contrast loss can be obtained based on each contrast loss. Specifically, it can be calculated by averaging and other methods. Obtain the total contrast loss, and then determine whether the total contrast loss meets the end condition. When it is satisfied, the training ends, and the current weight parameters are used as the optimal weight parameters of the feature extraction network; if the end condition is not met, the first feature extraction network The weight parameters of f and the second feature extraction network f' are updated, and the next round of training is entered until the training is completed and the optimal weight parameters are determined. Among them, the end condition can be that the number of training times reaches the preset number of training times, or the total contrast loss is less than the preset value.

It should also be noted that in the embodiment of the present application, it is assumed that the data enhancement is a, then v=a(x), y=f(v), z=g(y). Among them, the contrast loss function in the embodiment of this application is as follows:

in,

expresses right

Based on the above embodiments, embodiments of the present application also provide an image anomaly detection device. Please refer to Figure 5. The device includes:

The detection module 21 is used to detect the image to be detected using a pre-established image anomaly detection model and obtain an anomaly score;

The analysis module 22 is used to determine that the image to be detected is an abnormal image when the abnormality score meets the preset conditions; where:

The image anomaly detection model is trained based on a pre-established sample set, and the simulated abnormal samples in the sample set are obtained based on the original abnormal samples and the original normal samples.

It should be noted that the image anomaly detection device provided in the embodiments of the present application has the same beneficial effects as the image anomaly detection method involved in the above embodiments, and is more effective for the image anomaly detection involved in the embodiments of the present application. Please refer to the above embodiments for specific introduction of the method, which will not be described again in this application.

Based on the above embodiments, embodiments of the present application also provide an image anomaly detection system, which includes:

Memory, used to store computer programs;

A processor is used to implement the steps of the above image anomaly detection method when executing a computer program.

For example, the processor in the embodiment of the present application can be used to detect the image to be detected using a pre-established image anomaly detection model to obtain an abnormality score; when the abnormality score meets the preset conditions, the image to be detected is abnormal. Image; where: the image anomaly detection model is trained based on a pre-established sample set, and the simulated abnormal samples in the sample set are obtained based on the original abnormal samples and the original normal samples.

On the basis of the above embodiments, embodiments of the present application also provide a non-volatile readable storage medium. The non-volatile readable storage medium stores a computer program. When the computer program is executed by the processor, the above-mentioned implementation is implemented. Steps of image anomaly detection method.

The non-volatile readable storage medium can include: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disk, etc. The medium on which program code is stored.

Various component embodiments of the present application may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. Those skilled in the art will understand that a microprocessor or a digital signal processor (DSP) may be used in practice to implement some or all functions of some or all components in the computing processing device according to embodiments of the present application. The present application may also be implemented as an apparatus or device program (eg, computer program and computer program product) for performing part or all of the methods described herein. Such a program implementing the present application may be stored on a computer-readable medium, or may be in the form of one or more signals. Such signals may be downloaded from an Internet website, or provided on a carrier signal, or in any other form.

For example, Figure 6 shows a computing processing device that may implement methods according to the present application. The computing processing device includes a processor 610 and a computer program product in the form of a memory 620 or a non-volatile readable storage medium. Memory 620 may be electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read Only Memory), EPROM, hard disk, or ROM. The memory 620 has a storage space 630 for program code 631 for executing any method steps in the above-described methods. For example, the storage space 630 for program codes may include individual program codes 631 respectively used to implement various steps in the above method. These program codes can be read from or written into one or more computer program products. These computer program products include program code carriers such as hard disks, compact disks (CDs), memory cards or floppy disks. Such a computer program product is typically a portable or fixed storage unit as described with reference to FIG. 7 . The storage unit may have storage segments, storage spaces, etc. arranged similarly to the memory 620 in the computing processing device of FIG. 6 . The program code may, for example, be compressed in a suitable form. Typically, the storage unit includes computer readable code 631', ie code that can be read by, for example, a processor such as 610, which code, when executed by a computing processing device, causes the computing processing device to perform the methods described above. various steps.

Each embodiment in this specification is described in a progressive manner. Each embodiment focuses on its differences from other embodiments. The same and similar parts between the various embodiments can be referred to each other. As for the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple. For relevant details, please refer to the description in the method section.

It should also be noted that in this specification, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that these entities or operations There is no such actual relationship or sequence between operations. Furthermore, the terms "comprises," "comprises," or any other variations thereof are intended to cover a non-exclusive inclusion such that a process, method, article, or apparatus that includes a list of elements includes not only those elements, but also those not expressly listed other elements, or elements inherent to the process, method, article or equipment. Without further limitation, an element qualified by the statement "comprises a..." does not exclude the presence of additional identical elements in the process, method, article, or device that includes the element.

Those skilled in the art may further realize that the units and algorithm steps of each example described in connection with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of both. In order to clearly illustrate the possible functions of hardware and software, Interchangeability, in the above description, the composition and steps of each example have been generally described according to functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may implement the described functionality using different methods for each specific application, but such implementations should not be considered beyond the scope of this application.

The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly in hardware, in software modules executed by a processor, or in a combination of both. Software modules may be located in random access memory (RAM), memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or anywhere in the field of technology. any other known form of storage media.

The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be practiced in other embodiments without departing from the spirit or scope of the application. Therefore, the present application is not to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

An image anomaly detection method, characterized by including:

Use a pre-established image anomaly detection model to detect the image to be detected and obtain an anomaly score;

When the abnormality score meets the preset conditions, the image to be detected is an abnormal image; where:

The image anomaly detection model is trained based on a pre-established sample set, and the simulated abnormal samples in the sample set are obtained based on original abnormal samples and original normal samples.
The image anomaly detection method according to claim 1, wherein the step of using a pre-established image anomaly detection model to detect the image to be detected and obtaining the anomaly score includes:

Create a sample set;

The image anomaly detection model is obtained by training based on the sample set.
The method according to claim 2, characterized in that establishing the sample set includes:

Get a preset number of videos;

Convert the video into an image by extracting video frames;

Classify the image according to the frame rate to obtain original normal samples and original abnormal samples;

Construct simulated abnormal samples based on the original normal samples and the original abnormal samples;

The sample set is established based on the original normal samples, the original abnormal samples and the simulated abnormal samples.
The image anomaly detection method according to claim 1, characterized in that the simulated abnormal samples in the sample set are obtained based on original abnormal samples and original normal samples and include:

Select the target original abnormal sample from each original abnormal sample, and select the target original normal sample from each original normal sample;

Obtain a first replica image corresponding to the target area in the target original abnormal sample, and a second replica image corresponding to the target area in the target original normal sample;

Paste the first copied image and the second copied image onto the target original normal sample to obtain a simulated abnormal sample.
The image anomaly detection method according to claim 4, wherein pasting the first copied image and the second copied image onto the target original normal sample to obtain a simulated abnormal sample includes:

Paste the first copied image and the second copied image to a preset area on the target original normal sample; the preset area is an area centered on the center point of the target original normal sample;

The first replicated image and the second replicated image are respectively scaled to obtain simulated abnormal samples.
The image anomaly detection method according to claim 5, wherein the shape of the preset area is any one of the following shape areas:

Circular area and rectangular area.
The image anomaly detection method according to claim 5, wherein scaling the first copied image and the second copied image respectively includes:

The copied image and the second copied image are respectively randomly deformed and scaled according to different aspect ratios.
The image anomaly detection method according to claim 4, wherein selecting the target original abnormal sample from each original abnormal sample and selecting the target original normal sample from each original normal sample includes:

The target original abnormal sample is randomly selected from each original abnormal sample, and the target original normal sample is randomly selected from each original normal sample.
The image anomaly detection method according to any one of claims 1 to 8, characterized in that the image anomaly detection model is trained based on a pre-established sample set and includes:

Using a pre-established sample set to train the feature extraction network, obtain the optimal weight parameters of the feature extraction network;

Using each original normal sample in the sample set and the feature extraction network based on the optimal weight parameters to train the classifier to obtain the optimal parameters of the classifier;

An image anomaly detection model is constructed based on the feature extraction network of the optimal weight parameters and the classifier of the optimal parameters.
The image anomaly detection method according to claim 9, wherein the classifier is a KDE classifier.
The image anomaly detection method according to claim 9, characterized in that, using a pre-established sample set to train a feature extraction network, obtaining the optimal weight parameters of the feature extraction network includes:

For each sample in the sample set, perform data enhancement processing on the sample to obtain a first enhanced sample and a second enhanced sample;

The first enhanced sample is processed through the first feature extraction network to obtain the first encoding vector corresponding to the first enhanced sample, and the second enhanced sample is processed through the second feature extraction network to obtain the first encoding vector corresponding to the first enhanced sample. The second encoding vector corresponding to the second enhanced sample; the first feature extraction network and the second feature extraction network are the same;

Using a first projection network to process the first encoding vector to obtain a first projection vector, using a second projection network to process the second encoding vector to obtain a second projection vector; the first projection network and the The second projection network is the same;

Calculate the contrast loss of the first projection vector and the second projection vector;

Based on the contrast loss corresponding to each sample, the total contrast loss is calculated;

Determine whether the total comparative loss meets the ending condition;

In response to the total contrast loss satisfying the end condition, the method ends, and the optimal weight parameters of the first feature extraction network and the second feature extraction network are obtained.
The method according to claim 11, characterized in that the contrast loss is calculated by the following formula (1):

In the above formula (1),
represents the distribution to which x and x + belong, x represents the sample, x + represents the data enhancement of the sample x, xi - represents the data enhancement of the negative sample, p represents the distribution, f(x) T represents the transpose of x after extracting features, f(x + ) means extracting features for x + , f( xi - ) means extracting features for x i - , i represents the sample label, and N represents the total sample size.
The image anomaly detection method according to claim 11, characterized in that after determining whether the total contrast loss satisfies the end condition, it further includes:

In response to the total contrast loss not meeting the end condition, the weight parameters of the first feature extraction network and the second feature extraction network are updated, and the next round of training is entered.
The image anomaly detection method according to claim 1, characterized in that, when the anomaly score meets a preset condition, the image to be detected is an abnormal image, including:

When the abnormality score is greater than a preset threshold, the image to be detected is determined to be an abnormal image.
The image anomaly detection method according to claim 14, characterized in that after using a pre-established image anomaly detection model to detect the image to be detected and obtaining the anomaly score, it also includes:

When the abnormality score is less than or equal to the threshold, the image to be detected is determined to be a normal image.
An image anomaly detection device, characterized by including:

The detection module is used to detect the image to be detected using a pre-established image anomaly detection model and obtain an anomaly score;

An analysis module configured to determine that the image to be detected is an abnormal image when the abnormality score satisfies a preset condition; wherein:

The image anomaly detection model is trained based on a pre-established sample set, and the simulated abnormal samples in the sample set are obtained based on original abnormal samples and original normal samples.
An image anomaly detection system, characterized by including:

Memory, used to store computer programs;

A processor, configured to implement the steps of the image anomaly detection method according to any one of claims 1 to 15 when executing the computer program.
A non-volatile readable storage medium, characterized in that a computer program is stored on the non-volatile readable storage medium, and when the computer program is executed by a processor, it implements any one of claims 1 to 15 The steps of the image anomaly detection method.
A computing processing device, characterized by including:

A memory having computer readable code stored therein;

One or more processors, when the computer readable code is executed by the one or more processors, the computing processing device performs the steps of the image anomaly detection method described in any one of claims 1-15.
A computer program product, comprising computer readable code that, when run on a computing processing device, causes the computing processing device to perform a process according to any one of claims 1-15 Steps of image anomaly detection method.