A mine low-illumination image progressive restoration method, device, medium and product based on task self-learning reversible operator

Abstract

The application discloses a mine low-illumination image progressive restoration method and device based on a task self-learning reversible operator, a medium and a product, relates to the technical field of image processing, and the method is characterized in that: a reversible operator based on task self-learning is established, a mine low-illumination image is decomposed into a structure part and a detail part through the designed reversible operator, a structure information restoration module and a detail information restoration module are used to respectively restore the brightness information and the high-frequency characteristic information of the mine low-illumination image of the structure part and the detail part, a progressive restoration framework is adopted to gradually restore the mine low-illumination image, and finally, a loss function is established to supervise the training of the model, so that the reversible operator parameters and the restoration model parameters are optimized to the most suitable condition for processing the mine low-illumination image restoration task, and the restoration quality of the mine low-illumination image can be fully improved.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a method, device, medium and product for progressive restoration of low-light images in mines based on task self-learning reversible operators. Background Technology

[0002] Mining is generally an underground operation, and compared to surface operations, underground mining requires extra attention to safety. Therefore, numerous video surveillance devices are typically deployed inside mines to monitor activities in real time and provide intelligent alerts using detection algorithms. However, because underground lighting cannot consistently match the brightness of outdoor sunlight, and some areas are even affected by dust and other pollutants, the images captured by these video surveillance devices often suffer from low illumination, insufficient detail, and blurriness. To address this issue, researchers have conducted extensive research on the restoration of low-light images in mines, achieving relatively good results. Current common techniques typically convert low-light images into hue, saturation, and luminance components using HSV, then recover luminance information through nonlinear stretching of the luminance component. However, this approach overlooks the issues of insufficient detail and blurriness encountered during the degradation process. Furthermore, nonlinear stretching of the luminance component usually requires setting different parameters for different illumination levels to achieve satisfactory brightness restoration. Summary of the Invention

[0003] The purpose of this application is to provide a progressive restoration method, device, medium, and product for low-light images in mines based on task self-learning reversible operators, which can significantly improve the restoration quality of low-light images in mines.

[0004] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a progressive restoration method for low-light images in mines based on task-self-learning invertible operators, including: Acquire low-light images of underground mines; The low-light image is input into the forward transformation part of the reversible operator model to obtain the structural part and the detail part of the low-light image respectively; the reversible operator model performs forward and inverse transformations through convolution operations, and ensures the reversibility of the transformation by using the orthogonality of the convolution kernels for the forward and inverse transformations; The structure is restored to obtain the brightness information of the low-light image; The detailed parts are restored to obtain the high-frequency feature information of the low-light image; The brightness information and the high-frequency feature information are respectively subjected to the inverse transformation of the reversible operator model and then summed to obtain the image domain information; The image domain information is superimposed on the low-light image to obtain the restored low-light image of the mine.

[0005] Optionally, the low-light image is input into the forward transformation part of the reversible operator model to obtain the structural part and the detail part of the low-light image, specifically including: Using formula The low-light image is forward transformed to obtain the structural portion of the low-light image; wherein... This is the structural part obtained from the forward transformation operation; The specific parameters used for this convolution kernel; This is a convolution operation with a stride of 2; Using formula The low-light image is forward transformed to obtain the detailed portion of the low-light image; wherein, This refers to the detailed portion obtained from the forward transformation operation; The specific parameters used for the three convolutional kernels, This specifically refers to convolution operations with a stride of 2.

[0006] Optionally, the calculation formula for the inverse transformation of the brightness information using the reversible operator model includes: ; ; ; ; ; in, This is the result of the inverse transformation of the brightness information using a reversible operator model; Here it specifically refers to the transposed convolution with a stride of 2; They respectively take brightness information in the channel dimension , , The channel yields a matrix vector. The specific parameters used for these four convolutional kernels, r 1. r 2. r 3 and r All four are intermediate variables.

[0007] Optionally, the formula for ensuring the invertibility of the transformation by guaranteeing the orthogonality of the convolution kernel between the forward and inverse transformations is as follows: ; in, The specific parameters used for the convolution operation in the forward transform. The specific parameters used for the convolution operation in the inverse transformation.

[0008] Optionally, the training process of the reversible operator model includes: Acquire training low-light images; The training low-light image is input into the forward transform part of the invertible operator network to obtain the training structure part and the training detail part, respectively. The training structure part and the training detail part are restored respectively to obtain training brightness information and training high-frequency feature information; The training brightness information and the training high-frequency feature information are respectively input into the inverse transformation part of the reversible operator network to train the inverse transformation result; With formula Using the loss function, an invertible operator network is iteratively trained to obtain an invertible operator model; where, To ensure the reversibility of the forward and inverse transformations, the loss function is... It is the L2 norm. The specific parameters used for the convolution operation in the forward transform. The specific parameters used for the convolution operation in the inverse transformation.

[0009] Optionally, the structural portion is restored to obtain the brightness information of the low-light image, specifically including: Perform a convolution operation on the structural part to obtain an extended vector dimension feature vector; The extended vector dimension feature vector is normalized to obtain a normalized feature vector; Using formula A preliminary brightness restoration is performed to obtain a preliminary brightness restoration result; among which, This is a preliminary brightness restoration result. They are respectively , Convolution operation, To expand the feature vector dimension, These are normalized feature vectors; Using formula A second brightness restoration is performed to obtain the second brightness restoration result; among which, For the result of brightness restoration again, They are respectively , Convolution operations; A convolution operation is performed on the brightness restoration result to merge the information of the vector dimension, thereby obtaining the brightness information of the low-light image.

[0010] Optionally, the detailed parts are restored to obtain the high-frequency feature information of the low-light image, specifically including: Perform layer normalization on the detailed parts to obtain normalized feature vectors; Perform a convolution operation on the normalized feature vector to obtain a convolutional high-dimensional feature vector; Using formula Extract high-frequency detail features from the high-dimensional feature vector of the convolution; wherein, For high-frequency detail features, They are respectively , , , Convolution operations; Using formula The high-frequency detail features are restored to obtain the high-frequency restored feature vector; wherein, For high-frequency restored feature vectors, They are respectively , Convolution operations; Using formula Perform attention mechanism operations to obtain attention mechanism operation feature vectors; Using formula The high-frequency feature information of the low-light image is obtained; wherein, For high-frequency feature information of low-light images, For the details mentioned above, They are respectively , The convolution operation.

[0011] In a second aspect, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the progressive restoration method for low-light images of mines based on task self-learning reversible operators as described above.

[0012] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the progressive restoration method for low-light images of mines based on task self-learning reversible operators as described above.

[0013] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the progressive restoration method for low-light images of mines based on task self-learning reversible operators as described above.

[0014] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a method, device, medium, and product for progressive restoration of low-light images in mines based on task-based self-learning reversible operators. The method includes: acquiring a low-light image in a mine; inputting the low-light image into the forward transformation part of a reversible operator model to obtain the structural part and detail part of the low-light image; the reversible operator model performs forward and inverse transformations through convolution operations, ensuring the reversibility of the transformations by using the orthogonality of the convolution kernels; restoring the structural part to obtain the brightness information of the low-light image; restoring the detail part to obtain the high-frequency feature information of the low-light image; performing an inverse transformation on the brightness information and the high-frequency feature information of the reversible operator model and summing them to obtain image domain information; and superimposing the image domain information with the low-light image to obtain the restored low-light image of the mine. This application establishes a task-based self-learning reversible operator to decompose low-light images in mines into structural and detail parts. The structural information restoration module and the detail information restoration module are used to restore the brightness information and high-frequency feature information of the low-light images in mines, respectively, thereby significantly improving the restoration quality of low-light images in mines. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is an application environment diagram of a progressive restoration method for low-light images in mines based on task self-learning reversible operators, as described in one embodiment of this application.

[0017] Figure 2 This is a flowchart illustrating a progressive restoration method for low-light images in mines based on task-based self-learning reversible operators, provided as an embodiment of this application.

[0018] Figure 3 This is a schematic diagram of the network structure of the structural information restoration module provided in one embodiment of this application.

[0019] Figure 4 This is a schematic diagram of the network structure of the detail information restoration module provided in one embodiment of this application.

[0020] Figure 5 This is a schematic diagram illustrating the training process of the training reversible operator model, the structural information restoration module, and the detail information restoration module provided in an embodiment of this application.

[0021] Figure 6 This is a diagram illustrating the result of restoring a low-light image of a mine, as provided in an embodiment of this application.

[0022] Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] This application establishes a task-based self-learning reversible operator to decompose low-light images in mines into structural and detailed parts. Then, a structural information restoration module and a detailed information restoration module are used to restore the brightness information and high-frequency feature information of the structural and detailed parts of the low-light images in mines, respectively. Furthermore, a progressive restoration framework is adopted to gradually restore the low-light images in mines. Finally, a loss function is established to supervise the training of the model so that the parameters of the reversible operator and the restoration model are optimally suited for handling the task of restoring low-light images in mines, thereby significantly improving the restoration quality of low-light images in mines. First, this application uses a task-based self-learning forward operator to decompose the low-light mine image into structural and detail components, transforming it from the image domain to the operator transform domain, laying the foundation for the next restoration step. Next, a structural information restoration module is used to restore the structural components to recover the brightness information of the low-light mine image. Then, a detail information restoration module is used to restore the detail components to recover the detail information of the low-light mine image. Next, a task-based self-learning inverse operator is used to transform the restored structural and detail components back to the image domain, obtaining a preliminary restored low-light mine image. The above steps are repeated twice to progressively restore the low-light mine image, obtaining a high-fidelity restoration result. Finally, a corresponding loss function is established to effectively supervise the model during training, enabling the task-based self-learning reversible operator and model to obtain the optimal parameters best suited for the low-light mine image restoration task, thus achieving a high-fidelity restored low-light mine image during the inference stage.

[0025] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] The progressive restoration method for low-light mine images based on task-self-learning reversible operators provided in this application can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another server.

[0027] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0028] In one exemplary embodiment, such as Figure 2 As shown, a progressive restoration method for low-light images in mines based on task-based self-learning reversible operators is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 201 to 208. Wherein: S1. Obtain low-light images of the mine.

[0029] S2. Input the low-light image into the forward transformation part of the reversible operator model to obtain the structural part and the detail part of the low-light image respectively; the reversible operator model performs forward and inverse transformations through convolution operations, and ensures the reversibility of the transformation by guaranteeing the orthogonality of the convolution kernels for the forward and inverse transformations.

[0030] This embodiment uses low-light images obtained from underground mines. The low-light image is obtained by inputting it into the forward transform part of the designed reversible operator model. The structure and details are shown below: ; in, These are the structural and detailed parts after the forward transformation, and their sizes are respectively... , ; The input is a low-light image of a mine, its size is ; This is the forward transformation based on the task self-learning invertible operator proposed in this embodiment.

[0031] The reversible operator model based on task self-learning proposed in this embodiment includes forward and inverse transformations. The forward transformation decomposes the low-light mine image into structural and detailed parts, while the inverse transformation restores the structural and detailed parts back to the image domain. Furthermore, this embodiment designs the task-based reversible operator to be expressible through convolution, allowing parameter updates during training to make the transformation more suitable for the task of restoring low-light mine images. Specifically, the forward and inverse transformations are implemented through convolution operations, and the orthogonality of the convolution kernels used in the forward and inverse transformations is ensured to guarantee the reversibility of the transformation. For the forward transformation… It can convert a size of Image Transform into a Size of structural parts and Size details Image Transformation structure part Through a It is implemented by using convolution kernels to perform convolution operations with a stride of 2, specifically as follows: ; in, The structure obtained by the forward transformation operation has a size of ; The specific parameters used for this convolution kernel. This specifically refers to convolution operations with a stride of 2.

[0032] Image Conversion details Through three It is implemented by using convolution kernels to perform convolution operations with a stride of 2, specifically as follows: ; in, This is the detailed part obtained by the forward transformation operation, and its size is... ; The specific parameters used for the three convolutional kernels. This specifically refers to convolution operations with a stride of 2.

[0033] For inverse transformation It can convert sizes of structural parts and size are Details Convert to size Image Specifically, four are used respectively. The convolution kernels respectively target the structural parts and details The restored image is obtained by performing a transposed convolution with a stride of 2 and then summing the results. As shown below: ; ; ; ; ; in, express The restored image is [size missing] ; Here, it specifically refers to the transposed convolution with a stride of 2. To address the details separately Take in channel dimension , , The channel yields a matrix vector. The specific parameters used for these four convolutional kernels, r 1. r 2. r 3 and r All four are intermediate variables.

[0034] In this embodiment, the structural part and the detailed part are restored respectively, that is, the above five formulas are executed twice.

[0035] The above steps explained how to use convolution to implement forward and inverse transformations. Next, we will explain how to set the convolution parameters to ensure that the designed forward and inverse transformations are reversible. Specifically, the convolution parameters used to implement the forward and inverse transformations are designed with the following relationship to ensure their reversibility, as shown below: ; in, The specific parameters used to implement the convolution operation in the forward transform. The specific parameters used to implement the convolution operation in the inverse transform.

[0036] Based on the above description, forward and inverse transformations can be achieved through convolution operations, while maintaining the reversibility of both transformations. During training, initial values ​​satisfying the above conditions are assigned to the parameters before training begins, ensuring the reversibility of the forward and inverse transformations from the outset. Then, this constraint is added to the loss function to guarantee that reversibility remains as training progresses. The parameters are updated through backpropagation while maintaining reversibility, making the reversible transformation more suitable for handling low-light mine image restoration tasks. Here are some examples of initial value settings that satisfy the above conditions: ; .

[0037] The training process of the reversible operator model includes: Acquire training low-light images; The training low-light image is input into the forward transform part of the invertible operator network to obtain the training structure part and the training detail part, respectively. The training structure part and the training detail part are restored respectively to obtain training brightness information and training high-frequency feature information; The training brightness information and the training high-frequency feature information are respectively input into the inverse transformation part of the reversible operator network to train the inverse transformation result; With formula Using the loss function, an invertible operator network is iteratively trained to obtain an invertible operator model; where, To ensure the reversibility of the forward and inverse transformations, the loss function is... It is the L2 norm. The specific parameters used for the convolution operation in the forward transform. The specific parameters used for the convolution operation in the inverse transformation. This loss function ensures that the forward and inverse transformations are reversible during training.

[0038] S3. Restore the structural portion to obtain the brightness information of the low-light image.

[0039] S4. Restore the detailed parts to obtain the high-frequency feature information of the low-light image.

[0040] S5. The brightness information and the high-frequency feature information are respectively subjected to the inverse transformation of the reversible operator model and then summed to obtain the image domain information.

[0041] S6. The image domain information is superimposed on the low-light image to obtain the restored low-light image of the mine.

[0042] Specifically, in this embodiment, the structural and detailed parts obtained after forward transformation are input into the structural information restoration module and the detailed information restoration module, respectively, to restore the brightness information and high-frequency feature information of the low-light image of the mine, as shown below: ; ; in, These are the structural information restoration module and the detailed information restoration module, representing the restored structural portion (luminance information) and the restored detailed portion (high-frequency feature information), respectively. Their sizes are as follows: , ; The proposed structural information restoration module and detailed information restoration module have corresponding network structures as follows: Figure 3 and Figure 4 As shown.

[0043] Then, the restored structure and details are subjected to an inverse transformation using a reversible operator, transformed to the image domain, and then the input low-light image is added. The low-light image of the mine after the first reconstruction is shown below: ; in, This is a low-light image of the mine after its first reconstruction, and its size is [size missing]. ; This is the inverse transformation based on the task self-learning invertible operator proposed in this embodiment.

[0044] Repeat the above steps to obtain the first restored low-light image of the mine. As input, further refine the brightness and detail information, as shown below: ; ; ; ; in, These are low-light images of the mine after the first reconstruction. The structural and detailed parts obtained by performing a forward transformation of the invertible operator have the following sizes: ; For structural information restoration module Detailed information restoration module The restored structure and details are respectively... ; This is a low-light image of the mine after the second reconstruction, its size is [size missing]. .

[0045] Repeat the above steps one last time to perform the final restoration, resulting in the second restored low-light image of the mine. As input, the low-light image restoration result of the mine is obtained from the method output, as shown below: ; ; ; ; in, These are low-light images of the mine after the second reconstruction. The structural and detailed parts obtained by performing a forward transformation of the invertible operator have the following sizes: ; For structural information restoration module Detailed information restoration module The restored structure and details are respectively... ; This is the result of the mine low-light image restoration output by this method, and its size is [size missing]. .

[0046] The structural information restoration module in this embodiment is as follows: Its network structure is as follows Figure 3 As shown. The structural information restoration module mentioned in the above steps. The network structures used are all the same, therefore, only one of them will be discussed. A detailed introduction will be provided. For the forward transformation of low-light images using task-based self-learning invertible operators... Transformed into structural parts and details In this step, the structural part To restore low-light images of the mine Brightness information.

[0047] First, the structural part After inputting a convolutional layer, layer normalization is applied to expand the vector dimension and perform normalization operations, as shown below: ; ; in, These are the feature vectors for expanding the vector dimension and for normalization, respectively, both of which are of size . ; For the torch.nn.LayerNorm function, for The convolution operation has 3 input channels and 32 output channels.

[0048] Then, for the feature vector Perform two consecutive convolution operations, then input a non-linear activation function to generate non-linear weights, and then expand the feature vector of the vector dimension. A preliminary brightness restoration is performed, as shown below: ; in, This is the feature vector after our initial brightness restoration operation, and its size is... ; They are respectively for , The convolution operation has 32 input and output channels, and sigmod is the torch.nn.Sigmoid function.

[0049] Next, the above steps are repeated to generate nonlinear weights for the feature vector after the initial brightness restoration. The detailed brightness information is restored as follows: ; in, It is the feature vector after fine reconstruction of brightness information, and its size is ; They are respectively , The convolution operations have 32 input and output channels; For the torch.nn.LayerNorm function, This refers to the torch.nn.Sigmoid function.

[0050] Finally, the feature vector after fine reconstruction of degree information By combining information from the vector dimension through a convolution operation, the restored structure of the brightness information output by this module is generated, as shown below: ; in, This is the structural part after restoring the brightness information output by this module. for The convolution operation has 32 input channels and 3 output channels.

[0051] The detailed information restoration module in this embodiment is as follows: Its network structure is as follows Figure 4 As shown. The detailed information restoration module mentioned in the above steps. The network structures used are all the same, so we will only discuss one of them. A detailed introduction will be provided. For the forward transformation of low-light images using task-based self-learning invertible operators... Transformed into structural parts and details In this step, the details are discussed. Processing is performed to restore low-light images of the mine. High-frequency feature information. First, the details... After inputting a convolutional layer, layer normalization is applied to expand the vector dimension and perform normalization operations, as shown below: ; in, The feature vector is normalized after expanding the vector dimension, and its size is... ; For the torch.nn.LayerNorm function, for The convolution operation has 9 input channels and 32 output channels.

[0052] Then, the high-dimensional feature vectors processed above... After a convolution operation, convolution operations are performed on different receptive fields. Then, the summation is applied and the results are input into an activation function to extract high-frequency detail features from different spatial receptive fields, as shown below: ; ; in, These are, respectively, the high-dimensional feature vectors for further processing and the feature vectors extracted from different spatial receptive fields; both of them are of size . ; They are respectively , , , The convolution operations have 32 input and 32 output channels. This refers to the torch.nn.Relu function.

[0053] Next, the feature vectors of the different spatial receptive fields are extracted. The input is fed into two consecutive convolutional layers, and then the attention mechanism is applied, as shown below: ; ; in, These are the feature vectors for further high-frequency feature restoration and the feature vectors for attention mechanism operation, respectively; both are of size . ; They are respectively , The convolution operations have 32 input and 32 output channels. This refers to the torch.nn.Sigmoid function.

[0054] Finally, the feature vectors of the high-frequency features will be fully recovered. After being fed into two consecutive convolutional layers, the residual mechanism is used to generate the detailed information restored from the output of this module, as shown below: ; in, These are the restored detailed information of the output module and the detailed information input into the input module, respectively; both are of size [size missing]. ; They are respectively , Convolution operation, Both the input and output channels are 32. The input channel of conv14 is 32 and the output channel is 9.

[0055] It is worth noting that this embodiment also provides a training process, including training a reversible operator model, a structural information restoration module, and a detail information restoration module. The training process is as follows: Figure 5 As shown.

[0056] A loss function for progressive restoration of low-light mine images is established to effectively supervise the model during training, enabling the task-based self-learning invertible operators and the model to obtain the optimal parameters best suited for the low-light mine image restoration task. This allows for the acquisition of high-fidelity low-light mine images during the inference phase. Specifically, to meet the above requirements, the established loss function mainly consists of two parts: one is a loss function that constrains the invertible operators to have invertibility, and the other is the restoration constraint for progressively restoring low-light mine images, as shown below: ; ; ; in, This embodiment establishes a loss function for the progressive restoration of low-light mine images based on task-self-learning invertible operators. These are constraints on the reversibility of reversible operators and constraints on the gradual restoration of low-light images of mines. For the true values ​​in the training data pairs, The images show the results of the first, second, and final restorations of the progressive restoration method proposed after inputting low-light downhole images from the training data pair. It is an L1 norm. It is an L2 norm; The specific parameters used to implement the convolution operation in the forward transform. The specific parameters used to implement the convolution operation in the inverse transform.

[0057] This embodiment proposes a progressive restoration method for low-light images in mines based on task-based self-learning reversible operators. By establishing a task-based self-learning reversible operator, the low-light image in the mine is decomposed into structural and detailed parts. The structural information restoration module and the detailed information restoration module restore the brightness information and high-frequency feature information of the low-light image in the mine, respectively. A progressive restoration framework is adopted to restore the low-light image in the mine step by step. Finally, a loss function is established to supervise the training of the model so that the parameters of the reversible operator and the restoration model are optimally suited to the task of restoring low-light images in mines, thereby significantly improving the restoration quality of low-light images in mines.

[0058] To verify the effectiveness of this embodiment, please refer to Figure 6 , Figure 6 The images show the results of the mine low-light image restoration provided in this embodiment. (a) is the original low-light image in the mine; (b) is the final restoration result of the mine low-light image progressive restoration method based on task self-learning reversible operators proposed in this embodiment.

[0059] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 7 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When executed by the processor, the computer program implements a progressive restoration method for low-light images of mines based on task-based self-learning reversible operators.

[0060] Those skilled in the art will understand that Figure 7The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0061] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0062] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0063] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0064] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0065] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile 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 many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0066] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0067] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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 specification.

[0068] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A progressive restoration method for low-light images in mines based on task-based self-learning invertible operators, characterized in that, include: Acquire low-light images of underground mines; The low-light image is input into the forward transformation part of the reversible operator model to obtain the structural part and the detail part of the low-light image respectively; the reversible operator model performs forward and inverse transformations through convolution operations, and ensures the reversibility of the transformation by using the orthogonality of the convolution kernels for the forward and inverse transformations; The structure is restored to obtain the brightness information of the low-light image; The detailed parts are restored to obtain the high-frequency feature information of the low-light image; The brightness information and the high-frequency feature information are respectively subjected to the inverse transformation of the reversible operator model and then summed to obtain the image domain information; The image domain information is superimposed on the low-light image to obtain the restored low-light image of the mine.

2. The progressive restoration method for low-light images in mines based on task-self-learning reversible operators according to claim 1, characterized in that, The low-light image is input into the forward transform part of the reversible operator model to obtain the structural and detail parts of the low-light image, specifically including: Using formula The low-light image is forward transformed to obtain the structural portion of the low-light image; wherein... This is the structural part obtained from the forward transformation operation; The specific parameters used for this convolution kernel; This is a convolution operation with a stride of 2; Using formula The low-light image is forward transformed to obtain the detailed portion of the low-light image; wherein, This refers to the detailed portion obtained from the forward transformation operation; The specific parameters used for the three convolutional kernels, This specifically refers to convolution operations with a stride of 2.

3. The progressive restoration method for low-light images in mines based on task-self-learning reversible operators according to claim 1, characterized in that, The calculation formula for the inverse transformation of the brightness information using the reversible operator model includes: ； ； ； ； ； in, This is the result of the inverse transformation of the brightness information using a reversible operator model; Here it specifically refers to the transposed convolution with a stride of 2; They respectively take brightness information in the channel dimension , , The channel yields a matrix vector. The specific parameters used for these four convolutional kernels, r 1. r 2. r 3 and r All four are intermediate variables.

4. The progressive restoration method for low-light mine images based on task-self-learning reversible operators according to claim 1, characterized in that, The formula for ensuring the invertibility of the transformation by using the orthogonality of the convolution kernel to guarantee the forward and inverse transformations is as follows: ； in, The specific parameters used for the convolution operation in the forward transform. The specific parameters used for the convolution operation in the inverse transformation.

5. The progressive restoration method for low-light images in mines based on task-self-learning reversible operators according to claim 1, characterized in that, The training process of the reversible operator model includes: Acquire training low-light images; The training low-light image is input into the forward transform part of the invertible operator network to obtain the training structure part and the training detail part, respectively. The training structure part and the training detail part are restored respectively to obtain training brightness information and training high-frequency feature information; The training brightness information and the training high-frequency feature information are respectively input into the inverse transformation part of the reversible operator network to train the inverse transformation result; With formula Using the loss function, an invertible operator network is iteratively trained to obtain an invertible operator model; where, To ensure the reversibility of the forward and inverse transformations, the loss function is... It is the L2 norm. The specific parameters used for the convolution operation in the forward transform. The specific parameters used for the convolution operation in the inverse transformation.

6. The progressive restoration method for low-light images in mines based on task-self-learning reversible operators according to claim 1, characterized in that, The structure is restored to obtain the brightness information of the low-light image, specifically including: Perform a convolution operation on the structural part to obtain an extended vector dimension feature vector; The extended vector dimension feature vector is normalized to obtain a normalized feature vector; Using formula A preliminary brightness restoration is performed to obtain a preliminary brightness restoration result; among which, This is a preliminary brightness restoration result. They are respectively , Convolution operation, To expand the feature vector dimension, These are normalized feature vectors; Using formula A second brightness restoration is performed to obtain the second brightness restoration result; among which, For the result of brightness restoration again, They are respectively , Convolution operations; A convolution operation is performed on the brightness restoration result to merge the information of the vector dimension, thereby obtaining the brightness information of the low-light image.

7. The progressive restoration method for low-light images in mines based on task-self-learning reversible operators according to claim 1, characterized in that, The detailed parts are restored to obtain the high-frequency feature information of the low-light image, specifically including: Perform layer normalization on the detailed parts to obtain normalized feature vectors; Perform a convolution operation on the normalized feature vector to obtain a convolutional high-dimensional feature vector; Using formula Extract high-frequency detail features from the high-dimensional feature vector of the convolution; wherein, For high-frequency detail features, They are respectively , , , Convolution operations; Using formula The high-frequency detail features are restored to obtain the high-frequency restored feature vector; wherein, For high-frequency restored feature vectors, They are respectively , Convolution operations; Using formula Perform attention mechanism operations to obtain attention mechanism operation feature vectors; Using formula The high-frequency feature information of the low-light image is obtained; wherein, For high-frequency feature information of low-light images, For the details mentioned above, They are respectively , The convolution operation.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement the progressive restoration method for low-light images of mines based on task self-learning reversible operators as described in any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the progressive restoration method for low-light images of mines based on task-self-learning reversible operators as described in any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the progressive restoration method for low-light images of mines based on task-self-learning reversible operators as described in any one of claims 1-7.

Images