Infrared defect image expansion method and device, computer equipment, readable storage medium and program product

Abstract

The invention relates to an infrared defect image expansion method and device, computer equipment, a readable storage medium and a program product. The method comprises the following steps: carrying out dimension reduction processing on an infrared defect image sample by utilizing a preset compressed sensing network model to generate low-dimensional data; optimizing a preset generative adversarial network model according to the low-dimensional data, and obtaining an optimized generator; generating low-dimensional synthetic data according to the noise data by using the generator; and performing dimension raising processing on the low-dimensional synthetic data by using the compressed sensing network model to generate an infrared defect expansion image. The generative adversarial network model is a double adversarial mechanism, the generator and the discriminator confront each other, the generator and the fault detector confront each other, and the discriminator and the fault detector are respectively utilized to perform feedback adjustment on the generator, so that the quality of low-dimensional synthetic data generated by the generator can be improved, and the fault detection accuracy is improved. And sufficient training samples are provided for intelligent defect detection of power equipment.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to an infrared defect image expansion method, apparatus, computer equipment, computer-readable storage medium, and computer program product. Background Art

[0002] Intelligent detection of defects in power equipment is an important application field of non-destructive testing technology. It mainly uses non-destructive means such as X-rays, ultrasound, and infrared thermal imaging to conduct multi-parameter evaluation of the mechanical properties, structural integrity, and material characteristics of power equipment, so as to ensure the accurate identification and positioning of internal defects in key components such as insulators, transformers, and circuit breakers under the premise of normal operation of power equipment.

[0003] Currently, deep learning-based intelligent detection technology faces significant data bottlenecks in this field: On the one hand, obtaining high-quality annotated samples is difficult, and on-site collection conditions are limited. On the other hand, existing public datasets are limited in size and lack sample diversity, making it difficult to improve model generalization capabilities. This data scarcity severely restricts the performance breakthroughs of intelligent diagnostic algorithms and hinders the intelligentization of power equipment condition monitoring systems. Summary of the Invention

[0004] Based on this, it is necessary to provide an infrared defect image expansion method, device, computer equipment, computer-readable storage medium and computer program product to address the above technical problems.

[0005] In a first aspect, the present application provides a method for expanding an infrared defect image, the method comprising:

[0006] Use the preset compressed sensing network model to perform dimensionality reduction processing on infrared defect image samples to generate low-dimensional data;

[0007] Optimizing a preset generative adversarial network model according to the low-dimensional data and obtaining an optimized generator;

[0008] generating low-dimensional synthetic data based on the noise data using the generator;

[0009] The low-dimensional synthetic data is processed by using the compressed sensing network model to generate an infrared defect expansion image.

[0010] In one embodiment, optimizing a preset generative adversarial network model according to the low-dimensional data and obtaining an optimized generator includes:

[0011] Using the initial generator in the generative adversarial network model, first low-dimensional synthetic data and second low-dimensional synthetic data are generated respectively;

[0012] The discriminator is optimized according to the low-dimensional data and the first low-dimensional synthetic data, the fault detector is optimized according to the low-dimensional data, the second low-dimensional synthetic data and the known fault data, and the generator is optimized using the optimized discriminator, the optimized fault detector and the low-dimensional data until the generator and the discriminator meet a first balance condition and / or the generator and the fault detector meet a second balance condition, so as to obtain the generator of the low-dimensional data.

[0013] In one embodiment, optimizing the discriminator based on the low-dimensional data and the first low-dimensional synthetic data includes:

[0014] Inputting the low-dimensional data and the first low-dimensional synthetic data into the discriminator to obtain an output probability of the discriminator for the first low-dimensional synthetic data;

[0015] Inputting the output probability of the discriminator for the first low-dimensional synthetic data into the loss function of the discriminator to obtain a loss value of the discriminator;

[0016] The parameters of the discriminator are adjusted according to the loss value of the discriminator.

[0017] In one embodiment, optimizing the fault detector based on the low-dimensional data, the second low-dimensional synthetic data, and known fault data includes:

[0018] Inputting the low-dimensional data, the second low-dimensional synthetic data, and the fault data into a fault detector, and obtaining an output probability of the fault detector for the low-dimensional data, an output probability for the second low-dimensional synthetic data, and an output probability for the fault data;

[0019] Inputting the output probability of the fault detector for the low-dimensional data, the output probability of the second low-dimensional synthetic data, and the output probability of the fault data into the loss function of the fault detector to obtain a loss value of the fault detector;

[0020] Parameters of the fault detector are adjusted according to the loss value of the fault detector.

[0021] In one embodiment, optimizing the generator using the optimized discriminator, the optimized fault detector, and the low-dimensional data comprises:

[0022] Generating third low-dimensional synthetic data using the preset generator;

[0023] inputting the third low-dimensional synthetic data into the optimized discriminator and the optimized fault detector, respectively, to obtain an output probability and an uncertainty score of the optimized discriminator for the third low-dimensional synthetic data, and an output probability of the optimized fault detector for the third low-dimensional synthetic data;

[0024] Inputting the output probability and uncertainty score of the optimized discriminator for the third low-dimensional synthetic data and the output probability of the optimized fault detector for the third low-dimensional synthetic data into the loss function of the generator to obtain a loss value of the generator;

[0025] Adjust the parameters of the generator according to the loss value of the generator.

[0026] In one embodiment, the loss function of the generator can be determined by the following formula:

[0027]

[0028] in, is the loss function of the generator, is the third low-dimensional synthetic data, is the output probability of the optimized discriminator for the third low-dimensional synthetic data, is the output probability of the optimized fault detector for the third low-dimensional synthetic data, is the uncertainty score of the optimized discriminator for the third low-dimensional synthetic data, is the weight of the fault detector, is the number of third low-dimensional synthetic data.

[0029] In a second aspect, the present application further provides an infrared defect image expansion device, the device comprising:

[0030] A compression module is used to perform dimensionality reduction processing on infrared defect image samples using a preset compressed sensing network model to generate low-dimensional data;

[0031] A generative adversarial network module is configured to optimize a preset generative adversarial network model based on the low-dimensional data and obtain an optimized generator; and to generate low-dimensional synthetic data based on the noise data using the generator;

[0032] The reconstruction module is used to use the compressed sensing network model to perform dimensionality-increasing processing on the low-dimensional synthetic data to generate an infrared defect expansion image.

[0033] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method provided in any of the above embodiments when executing the computer program.

[0034] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method provided in any of the above embodiments when the computer program is executed by a processor.

[0035] In a fifth aspect, the present application also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the method provided in any of the above embodiments.

[0036] In the above-mentioned infrared defect image expansion method, device, computer equipment, computer-readable storage medium and computer program product, by using a preset compressed sensing network model to reduce the dimension of the infrared defect image samples and generate low-dimensional data, the training difficulty of the generative adversarial network model is reduced and the training time is reduced; further, the preset generative adversarial network model is optimized according to the low-dimensional data, and the optimized generator is obtained. The generator is used to generate low-dimensional synthetic data based on the noise data, and finally the compressed sensing network model is used to increase the dimension of the low-dimensional synthetic data to generate infrared defect expansion images, enriching the infrared defect image set and providing sufficient training samples for intelligent detection of power equipment defects. In addition, since the generative adversarial network model of the embodiment of the present application is a dual adversarial mechanism: the generator and the discriminator are in conflict with each other, and the generator and the fault detector are in conflict with each other, the discriminator and the fault detector are used to feedback and adjust the generator, respectively, which can improve the quality of the low-dimensional synthetic data generated by the generator, making the low-dimensional synthetic data generated by the generator closer to the real data and with clearer details. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments of the present application or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 1 is a flow chart of an infrared defect image expansion method according to an embodiment;

[0039] Figure 2 A schematic diagram of the output of each step in the infrared defect image expansion method provided by an embodiment;

[0040] Figure 3 A schematic diagram of the structure of a generative adversarial network model in an infrared defect image expansion method provided by one embodiment;

[0041] Figure 4is a schematic diagram of a process for optimizing a discriminator based on low-dimensional data and first low-dimensional synthetic data in one embodiment;

[0042] Figure 5 A schematic diagram of a process for optimizing a fault detector based on low-dimensional data, second low-dimensional synthetic data, and known fault data in one embodiment;

[0043] Figure 6 A schematic diagram of a process for utilizing an optimized discriminator, an optimized fault detector, and a low-dimensional data optimized generator in one embodiment;

[0044] Figure 7 is a structural block diagram of an infrared defect image expansion device in one embodiment;

[0045] Figure 8 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0047] In one embodiment, Figure 1 As shown, the present application provides an infrared defect image expansion method, including steps 102 to 108.

[0048] Step 102: Use a preset compressed sensing network model to perform dimensionality reduction processing on the infrared defect image sample to generate low-dimensional data.

[0049] According to the theory of compressed sensing (CS), if an N-dimensional signal is compressible, then it is in the sparse basis It can be sparsely represented as:

[0050]

[0051] Where: For signal In the transform domain The sparse representation under , and only non-zero elements By measuring the matrix Perform a low-dimensional projection:

[0052]

[0053] Where: for dimensional observation vector, so the low-dimensional projection process can be expressed as follows by combining equations (1) and (2):

[0054]

[0055] Where, is the perception matrix, and the perception matrix The following restricted isometry property (RIP) must be satisfied:

[0056]

[0057] In the formula Isometric constant, considering that it is not easy to verify whether the perception matrix satisfies the constrained isometric property, an equivalent irrelevance criterion is proposed, as shown in the following formula:

[0058]

[0059] In the formula Represents the measurement matrix With sparse basis The correlation, is a sparse basis Columns, are the rows of the observation matrix, The larger the value of , the closer the observation matrix and the sparse basis The higher the correlation, The smaller the value of , the lower the correlation, and a high correlation means that more observations are needed. Therefore, when selecting an observation matrix, one should try to choose an observation matrix that is uncorrelated with the sparse basis.

[0060] Therefore, if Figure 2 As shown in Figure 2, the compressed sensing network model can project the infrared defect sample image from high dimension to low dimension to obtain low-dimensional data. Among them, the measurement matrix in the compressed sensing network model is , sparse basis and perception matrix All can be set in advance. Measurement matrix It can be a Bernoulli random matrix, a partial Hadamard measurement matrix, or a Gaussian random matrix, etc.

[0061] Step 104: Optimize the preset generative adversarial network model based on the low-dimensional data and obtain the optimized generator.

[0062] like Figure 2 and Figure 3As shown, the preset generative adversarial network (GAN) model includes a generator, a discriminator, and a fault detector. The generator's input is noise data, from which it randomly generates low-dimensional synthetic samples and outputs them to the discriminator and fault detector. The discriminator is used to distinguish between real samples and synthetic samples generated by the generator. Through the competition between the discriminator and the generator, the quality of the synthetic samples generated by the generator is improved, and the difference between the synthetic samples and real samples (i.e., low-dimensional data) is reduced. Since the discriminator does not focus on sample quality, a fault detector is introduced into the GAN model. The fault detector is used to identify faulty samples in the synthetic samples, i.e., repeated, blurred, or distorted samples. Through the competition between the fault detector and the generator, the details of the synthetic samples generated by the generator are clearer. It can be understood that the GAN model in this embodiment of the application uses a dual adversarial mechanism: the generator and the discriminator compete with each other, and the generator competes with the fault detector. The generator needs to generate realistic samples to make it difficult for the discriminator to distinguish between real samples and synthetic samples, and the generator needs to generate high-quality samples to prevent the fault detector from identifying the synthetic samples as faulty data.

[0063] Step 106: Use a generator to generate low-dimensional synthetic data based on the noise data.

[0064] Step 108: Use a compressed sensing network model to perform dimensionality-upgrading processing on the low-dimensional synthetic data to generate an infrared defect expansion image.

[0065] After using the generator to generate low-dimensional synthetic data, it is necessary to use the reconstruction algorithm in the compressed sensing network model to project the low-dimensional synthetic data into a high dimension to generate an infrared defect expansion image.

[0066] In the embodiment of the present application, by using a preset compressed sensing network model to perform dimensionality reduction processing on infrared defect image samples to generate low-dimensional data, the training difficulty of the generative adversarial network model is reduced and the training time is reduced; further, the preset generative adversarial network model is optimized according to the low-dimensional data, and the optimized generator is obtained. The generator is used to generate low-dimensional synthetic data based on the noise data, and finally the compressed sensing network model is used to perform dimensionality increase processing on the low-dimensional synthetic data to generate infrared defect expansion images, enriching the infrared defect image set and providing sufficient training samples for intelligent detection of power equipment defects. In addition, since the generative adversarial network model of the embodiment of the present application is a dual adversarial mechanism: the generator and the discriminator are in conflict with each other, and the generator and the fault detector are in conflict with each other, the discriminator and the fault detector are used to perform feedback adjustment on the generator respectively, which can improve the quality of the low-dimensional synthetic data generated by the generator, making the low-dimensional synthetic data generated by the generator closer to the real data and with clearer details.

[0067] In one embodiment, a preset generative adversarial network model is optimized according to low-dimensional data, and an optimized generator is obtained, including: using the initial generator in the generative adversarial network model to generate first low-dimensional synthetic data and second low-dimensional synthetic data, respectively, optimizing the discriminator according to the low-dimensional data and the first low-dimensional synthetic data, optimizing the fault detector according to the low-dimensional data, the second low-dimensional synthetic data and known fault data, and optimizing the generator using the optimized discriminator, the optimized fault detector and the low-dimensional data, until the generator and the discriminator meet the first balance condition, and / or the generator and the fault detector meet the second balance condition, to obtain the generator of low-dimensional data.

[0068] When optimizing a generative adversarial network model, an alternating optimization strategy is employed: first, the generator and fault detector are kept constant while optimizing the discriminator; then, the generator and discriminator are kept constant while optimizing the fault detector; and finally, the discriminator and fault detector are kept constant while optimizing the generator. This process is repeated until the generator and discriminator meet a first equilibrium condition and / or the generator and fault detector meet a second equilibrium condition. The first equilibrium condition refers to a balance between the generator and discriminator, where the generator's loss function converges with the discriminator's loss function. The second equilibrium condition refers to a balance between the generator and fault detector, where the generator's loss function converges with the fault detector's loss function.

[0069] The first low-dimensional synthetic data and the second low-dimensional synthetic data are not generated at the same time. The first low-dimensional synthetic data is generated by the generator before the discriminator is optimized, and the second low-dimensional synthetic data is generated by the generator after the discriminator is optimized and before the fault detector is optimized.

[0070] Specifically, if Figure 4 As shown, optimizing the discriminator according to the low-dimensional data and the first low-dimensional synthetic data includes steps 402 to 406.

[0071] Step 402: input the low-dimensional data and the first low-dimensional synthetic data into a discriminator to obtain the output probability of the discriminator for the first low-dimensional synthetic data.

[0072] In step 404 , the output probability of the discriminator for the first low-dimensional synthetic data is input into the loss function of the discriminator to obtain the loss value of the discriminator.

[0073] The standard loss function for the Generative Adversarial Network model is:

[0074]

[0075] in, represents the standard loss function of the generative adversarial network model, represents the number of the first low-dimensional synthetic data, represents the expected output value of the discriminator when the first low-dimensional synthetic data i is input, and The expected value of is 1. is the first low-dimensional synthetic data, is the output probability of the discriminator for the first low-dimensional synthetic data.

[0076] By simplifying the standard loss function, we can get the loss function of the discriminator :

[0077]

[0078] Inputting the output probability of the discriminator for the first low-dimensional synthetic data into the above formula can obtain the loss value of the discriminator.

[0079] Step 406: Adjust the parameters of the discriminator according to the loss value of the discriminator.

[0080] According to the loss value of the discriminator, the back propagation algorithm can be used to start from the output layer of the discriminator and reversely calculate the loss value of each layer of the discriminator to obtain the gradient change of the loss function of the discriminator, and then according to the loss function of the discriminator, the gradient change of the loss function of the discriminator can be obtained. The maximization requirement adjusts the parameters of the discriminator.

[0081] Then, if Figure 5 As shown, the fault detector is optimized according to the low-dimensional data, the second low-dimensional synthetic data and the known fault data, including steps 502 to 506.

[0082] Step 502: input the low-dimensional data, the second low-dimensional synthetic data and the fault data into the fault detector, and obtain the output probability of the fault detector for the low-dimensional data, the output probability for the second low-dimensional synthetic data and the output probability for the fault data.

[0083] Step 504 : Input the output probability of the fault detector for the low-dimensional data, the output probability for the second low-dimensional synthetic data, and the output probability for the fault data into the loss function of the fault detector to obtain the loss value of the fault detector.

[0084] Loss function for fault detector It can be expressed as:

[0085]

[0086] in, represents the number of low-dimensional data, represents low-dimensional data, represents the output probability of the fault detector for low-dimensional data, Indicates the number of fault data, Indicates fault data, represents the output probability of the fault detector for fault data, represents the number of second low-dimensional synthetic data, represents the second low-dimensional synthetic data, represents the output probability of the fault detector for the second low-dimensional synthetic data.

[0087] Step 506: Adjust the parameters of the fault detector according to the loss value of the fault detector.

[0088] According to the loss value of the fault detector, the back propagation algorithm can be used to reversely calculate the loss value of each layer of the fault detector starting from the output layer of the fault detector to obtain the gradient change of the loss function of the fault detector, and then the loss function of the fault detector can be used to calculate the gradient change of the loss function of the fault detector. Minimize the need to tune the parameters of the fault detector.

[0089] Finally, if Figure 6 As shown, the generator is optimized using the optimized discriminator, the optimized fault detector and the low-dimensional data, including steps 602 to 608.

[0090] Step 602: Generate third low-dimensional synthetic data using a preset generator.

[0091] Step 604: Input the third low-dimensional synthetic data into the optimized discriminator and the optimized fault detector respectively, and obtain the output probability and uncertainty score of the optimized discriminator for the third low-dimensional synthetic data, as well as the output probability of the optimized fault detector for the third low-dimensional synthetic data.

[0092] The discriminator's uncertainty score for the third low-dimensional synthetic data can be the variance or entropy of the discriminator's output probability for the third low-dimensional synthetic data. Introducing the uncertainty score allows the generator to focus more on areas with higher uncertainty, thereby optimizing the generation quality in these areas. When the uncertainty score is close to 1, the generator should focus on the current low-dimensional synthetic data. When the uncertainty score is closer to 0, the optimization priority of the third low-dimensional synthetic data is lowered.

[0093] Step 606: Input the output probability and uncertainty score of the optimized discriminator for the third low-dimensional synthetic data and the output probability of the optimized fault detector for the third low-dimensional synthetic data into the loss function of the generator to obtain the loss value of the generator.

[0094] Generator loss function It can be determined by the following formula:

[0095]

[0096] in, is the third low-dimensional synthetic data, is the output probability of the optimized discriminator for the third low-dimensional synthetic data, is the output probability of the optimized fault detector for the third low-dimensional synthetic data, is the uncertainty score of the optimized discriminator for the third low-dimensional synthetic data, is the weight of the fault detector, is the number of third low-dimensional synthetic data.

[0097] Step 608: Adjust the parameters of the generator according to the loss value of the generator.

[0098] According to the loss value of the generator, the back propagation algorithm can be used to start from the output layer of the generator and reversely calculate the loss value of each layer of the generator to obtain the gradient change of the loss function of the generator, and then according to the loss function of the generator, the gradient change of the loss function of the generator can be obtained. Minimize the need to tune the parameters of the fault detector.

[0099] In this embodiment, the discriminator, fault detector and generator in the preset generative adversarial network model are cyclically optimized according to low-dimensional data, especially based on the dual adversarial mechanism, and the optimized discriminator and optimized fault detector are used to optimize the generator respectively, so that the low-dimensional synthetic data finally generated by the generator is closer to the real data and has clearer details, providing a large number of high-quality training samples for the intelligent detection of defects in power equipment.

[0100] It should be understood that, although the steps in the flowcharts of the above embodiments are shown in sequence as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be performed in other orders. Moreover, at least a portion of the steps in the flowcharts of the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily performed at the same time, but can be performed at different times. The execution order of these steps or stages is not necessarily to be performed in sequence, but can be performed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0101] Based on the same inventive concept, embodiments of the present application also provide an infrared defect image expansion device for implementing the aforementioned infrared defect image expansion method. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more embodiments of the infrared defect image expansion device provided below can be found in the aforementioned limitations of the infrared defect image expansion method and will not be further elaborated here.

[0102] In an exemplary embodiment, Figure 7 As shown, an infrared defect image expansion device is provided, including: a compression module 702, a generative adversarial module 704 and a reconstruction module 706, wherein:

[0103] The compression module 702 is used to perform dimensionality reduction processing on the infrared defect image samples using a preset compressed sensing network model to generate low-dimensional data;

[0104] The generative adversarial module 704 is used to optimize the preset generative adversarial network model based on the low-dimensional data and obtain the optimized generator; using the generator, generate low-dimensional synthetic data based on the noise data;

[0105] The reconstruction module 706 is used to perform dimensionality-increasing processing on the low-dimensional synthetic data using a compressed sensing network model to generate an infrared defect expansion image.

[0106] In one embodiment, the generative adversarial module is further configured to generate first low-dimensional synthetic data and second low-dimensional synthetic data using an initial generator in the generative adversarial network model;

[0107] The discriminator is optimized according to the low-dimensional data and the first low-dimensional synthetic data, the fault detector is optimized according to the low-dimensional data, the second low-dimensional synthetic data and the known fault data, and the generator is optimized using the optimized discriminator, the optimized fault detector and the low-dimensional data until the generator and the discriminator meet the first balance condition and / or the generator and the fault detector meet the second balance condition to obtain a generator of low-dimensional data.

[0108] In one embodiment, the generative adversarial module is further configured to input the low-dimensional data and the first low-dimensional synthetic data into the discriminator to obtain an output probability of the discriminator for the first low-dimensional synthetic data;

[0109] Inputting the output probability of the discriminator for the first low-dimensional synthetic data into the loss function of the discriminator to obtain the loss value of the discriminator;

[0110] Adjust the discriminator parameters according to the discriminator loss value.

[0111] In one embodiment, the generative adversarial module is further configured to input the low-dimensional data, the second low-dimensional synthetic data, and the fault data into the fault detector, and obtain an output probability of the fault detector for the low-dimensional data, an output probability for the second low-dimensional synthetic data, and an output probability for the fault data;

[0112] Inputting the output probability of the fault detector for the low-dimensional data, the output probability for the second low-dimensional synthetic data, and the output probability for the fault data into the loss function of the fault detector to obtain a loss value of the fault detector;

[0113] According to the loss value of the fault detector, the parameters of the fault detector are adjusted.

[0114] In one embodiment, the generative adversarial module is further configured to generate third low-dimensional synthetic data using a preset generator;

[0115] Inputting the third low-dimensional synthetic data into the optimized discriminator and the optimized fault detector, respectively, to obtain an output probability and an uncertainty score of the optimized discriminator for the third low-dimensional synthetic data, and an output probability of the optimized fault detector for the third low-dimensional synthetic data;

[0116] Inputting the output probability and uncertainty score of the optimized discriminator for the third low-dimensional synthetic data and the output probability of the optimized fault detector for the third low-dimensional synthetic data into the loss function of the generator to obtain the loss value of the generator;

[0117] According to the loss value of the generator, adjust the parameters of the generator.

[0118] In one embodiment, the generative adversarial module is further configured to determine the generator's loss function using the following formula:

[0119]

[0120] in, is the loss function of the generator, is the third low-dimensional synthetic data, is the output probability of the optimized discriminator for the third low-dimensional synthetic data, is the output probability of the optimized fault detector for the third low-dimensional synthetic data, is the uncertainty score of the optimized discriminator for the third low-dimensional synthetic data, is the weight of the fault detector, is the number of third low-dimensional synthetic data.

[0121] Each module in the infrared defect image expansion device can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in hardware form, or can be stored in a computer device memory in software form, so that the processor can call and execute the corresponding operations of each module.

[0122] In an exemplary embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 8As shown. The computer device includes a processor, memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals via wired or wireless means, and the wireless means can be implemented via Wi-Fi, mobile cellular networks, near-field communication (NFC), or other technologies. When executed by the processor, the computer program implements a method for expanding infrared defect images. The display unit of the computer device is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device casing, or an external keyboard, touchpad or mouse.

[0123] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0124] In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor implements the infrared defect image expansion method provided in any of the above embodiments when executing the computer program.

[0125] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the infrared defect image expansion method provided by any of the above embodiments is implemented.

[0126] In one embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the infrared defect image expansion method provided in any of the above embodiments is implemented.

[0127] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), quantum computing-based data processing logic devices, artificial intelligence (AI) processors, and the like.

[0128] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0129] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A method for expanding an infrared defect image, characterized in that: The method comprises: Use the preset compressed sensing network model to perform dimensionality reduction processing on infrared defect image samples to generate low-dimensional data; Optimizing a preset generative adversarial network model according to the low-dimensional data and obtaining an optimized generator; generating low-dimensional synthetic data based on the noise data using the generator; The low-dimensional synthetic data is processed by using the compressed sensing network model to generate an infrared defect expansion image.

2. The method according to claim 1, characterized in that Optimizing a preset generative adversarial network model according to the low-dimensional data and obtaining an optimized generator includes: Using the initial generator in the generative adversarial network model, first low-dimensional synthetic data and second low-dimensional synthetic data are generated respectively; The discriminator is optimized according to the low-dimensional data and the first low-dimensional synthetic data, the fault detector is optimized according to the low-dimensional data, the second low-dimensional synthetic data and the known fault data, and the generator is optimized using the optimized discriminator, the optimized fault detector and the low-dimensional data until the generator and the discriminator meet a first balance condition and / or the generator and the fault detector meet a second balance condition, so as to obtain the generator of the low-dimensional data.

3. The method according to claim 2, characterized in that Optimizing the discriminator according to the low-dimensional data and the first low-dimensional synthetic data includes: Inputting the low-dimensional data and the first low-dimensional synthetic data into the discriminator to obtain an output probability of the discriminator for the first low-dimensional synthetic data; Inputting the output probability of the discriminator for the first low-dimensional synthetic data into the loss function of the discriminator to obtain a loss value of the discriminator; The parameters of the discriminator are adjusted according to the loss value of the discriminator.

4. The method according to claim 2, characterized in that Optimizing the fault detector according to the low-dimensional data, the second low-dimensional synthetic data, and known fault data comprises: Inputting the low-dimensional data, the second low-dimensional synthetic data, and the fault data into a fault detector, and obtaining an output probability of the fault detector for the low-dimensional data, an output probability for the second low-dimensional synthetic data, and an output probability for the fault data; Inputting the output probability of the fault detector for the low-dimensional data, the output probability of the second low-dimensional synthetic data, and the output probability of the fault data into the loss function of the fault detector to obtain a loss value of the fault detector; Parameters of the fault detector are adjusted according to the loss value of the fault detector.

5. The method according to claim 2, characterized in that Optimizing the generator by using the optimized discriminator, the optimized fault detector, and the low-dimensional data includes: Generating third low-dimensional synthetic data using the preset generator; inputting the third low-dimensional synthetic data into the optimized discriminator and the optimized fault detector, respectively, to obtain an output probability and an uncertainty score of the optimized discriminator for the third low-dimensional synthetic data, and an output probability of the optimized fault detector for the third low-dimensional synthetic data; Inputting the output probability and uncertainty score of the optimized discriminator for the third low-dimensional synthetic data and the output probability of the optimized fault detector for the third low-dimensional synthetic data into the loss function of the generator to obtain a loss value of the generator; Adjust the parameters of the generator according to the loss value of the generator.

6. The method according to claim 5, characterized in that The loss function of the generator can be determined by the following formula: in, is the loss function of the generator, is the third low-dimensional synthetic data, is the output probability of the optimized discriminator for the third low-dimensional synthetic data, is the output probability of the optimized fault detector for the third low-dimensional synthetic data, is the uncertainty score of the optimized discriminator for the third low-dimensional synthetic data, is the weight of the fault detector, is the number of third low-dimensional synthetic data.

7. An infrared defect image expansion device, characterized in that: The device comprises: A compression module is used to perform dimensionality reduction processing on infrared defect image samples using a preset compressed sensing network model to generate low-dimensional data; A generative adversarial network module is configured to optimize a preset generative adversarial network model based on the low-dimensional data and obtain an optimized generator; and to generate low-dimensional synthetic data based on the noise data using the generator; The reconstruction module is used to use the compressed sensing network model to perform dimensionality-increasing processing on the low-dimensional synthetic data to generate an infrared defect expansion image.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images