Image processing method, terminal device and computer readable storage medium

Abstract

The present application is applicable to the field of image processing technology, and provides an image processing method, a terminal device and a computer-readable storage medium, including: inputting a low-light image into a trained continuous update network to extract local features and global features to obtain feature images of different scales, inputting the feature images of different scales into a trained color enhancement network for processing to obtain a target image, extracting feature images of different scales through the continuous update network, extracting global information and local information, and ensuring that the restored normal light image does not have the phenomenon of missing details, and then enhancing the image color and texture of the restored image through the color enhancement network to ensure the image quality, thereby solving the problem that the current method for brightening images taken under low-light conditions has poor image processing effect.

Description

Technical Field

[0001] The present application belongs to the field of image processing technology, and in particular, relates to an image processing method, a terminal device, and a computer-readable storage medium. Background Art

[0002] In many fields such as industrial production, video surveillance, intelligent transportation, remote sensing control, etc., various processing such as image recognition and image information extraction are involved. However, images taken under low-light conditions often have problems such as low brightness, low contrast, and many noise points due to the dim light, resulting in poor subsequent image processing effects and the inability to identify effective image information.

[0003] In order to improve the reliability and robustness of image recognition, it is necessary to brighten images taken under low-light conditions. At present, the methods for brightening images taken under low-light conditions are usually based on retinal theory or generative adversarial networks to achieve brightening. Among them, the method based on retinal theory is to use two networks, one of which decomposes the low-light image into illumination and reflection, and the other network acts as an enhancer to improve the illumination map of the low-light image. However, due to the difficulty in decomposing illumination and reflection, the processing effect is not good; the method based on generative adversarial networks is to find the mapping between low-light and normal light domains, regard low-light enhancement as a domain transfer learning task, use the generator to estimate the normal light image from the low-light image, and use the discriminator to estimate the visual quality. However, the generator and discriminator need to be trained at the same time and the expectations are opposite, so it is difficult to get the expected output, which also leads to the problem of poor processing effect.

[0004] In summary, the current method for brightening images captured under low-light conditions has the problem of poor image processing effect. Summary of the invention

[0005] In view of this, the embodiments of the present application provide an image processing method, a terminal device, and a computer-readable storage medium to solve the problem that the current method for brightening images captured under low-light conditions has poor image processing effects.

[0006] In a first aspect, an embodiment of the present application provides an image processing method, comprising:

[0007] The low-light image is input into the trained continuous update network to extract local and global features, and feature images of different scales are obtained;

[0008] The feature images of different scales are input into a trained color enhancement network for processing to obtain a target image.

[0009] Optionally, the image processing method further includes:

[0010] Constructing a continuous update network and a color enhancement network;

[0011] Get a sample dataset;

[0012] The continuous update network and the color enhancement network are trained based on the sample data set to obtain the trained continuous update network and the trained color enhancement network.

[0013] Optionally, the continuously updated network includes N U-shaped network structures and feature fusion structures, where N is a positive integer greater than or equal to 2.

[0014] Optionally, the U-shaped network structure includes a downsampling structure and an upsampling structure, the downsampling structure includes four feature blocks of different scales, the upsampling structure includes four feature blocks of different scales, and different feature blocks are connected using skip connections.

[0015] Optionally, the feature fusion structure includes a convolutional layer, an average pooling layer and a fully connected layer.

[0016] Optionally, the color enhancement network includes a residual acquisition module, a convolution module and a weight determination module.

[0017] Optionally, the characteristic images of different scales include a first-scale characteristic image, a second-scale characteristic image, and a third-scale characteristic image.

[0018] In a second aspect, an embodiment of the present application provides a terminal device, including:

[0019] The characteristic images of different scales include a first-scale characteristic image, a second-scale characteristic image, and a third-scale characteristic image.

[0020] In a third aspect, an embodiment of the present application provides a terminal device, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method described in the first aspect or any optional method of the first aspect is implemented.

[0021] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect or any optional method of the first aspect is implemented.

[0022] In a fifth aspect, an embodiment of the present application provides a computer program product. When the computer program product runs on a terminal device, the terminal device executes the method described in the first aspect or any optional method of the first aspect.

[0023] The image processing method, terminal device, computer-readable storage medium and computer program product provided by the embodiments of the present application have the following beneficial effects:

[0024] By continuously updating the network to extract feature images of different scales, global information and local information are extracted, which can ensure that the restored normal lighting image will not have any missing details. The color enhancement network is then used to enhance the image color and texture of the restored image to ensure image quality, solving the problem of poor image processing effect in the current method of brightening images taken under low light conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0026] Figure 1 is a schematic flow chart of an image processing method provided in an embodiment of the present application;

[0027] Figure 2 It is a structural diagram of a U-shaped network structure of a continuously updated network in an embodiment of the present application;

[0028] Figure 3 It is a flowchart of the feature fusion process provided by the embodiment of the present application;

[0029] Figure 4 It is a schematic diagram of the process of processing feature images of different scales by the color enhancement network provided in the embodiment of the present application;

[0030] Figure 5 is a schematic diagram of the structure of a terminal device provided in an embodiment of the present application;

[0031] Figure 6 is a structural schematic diagram of a terminal device provided by another embodiment of the present application;

[0032] Figure 7 It is a structural schematic diagram of a computer-readable storage medium provided in an embodiment of the present application. DETAILED DESCRIPTION

[0033] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0034] It should be understood that the term "and / or" used in the specification of this application and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes these combinations. In addition, in the description of the specification of this application and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0035] It should also be understood that references to "one embodiment" or "some embodiments" etc. described in the specification of the present application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Thus, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in the specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0036] A low-light environment refers to a specific environment where natural lighting conditions are poor and affect people's vision. An environment with an illumination of less than 50 lumens is usually called a low-light environment. Images captured in a low-light environment usually have problems such as low brightness, low contrast, and many noise points. An image processing method provided in an embodiment of the present application extracts feature images of different scales by continuously updating a network, extracts global information and local information, and can ensure that the restored normal lighting image does not have the phenomenon of missing details, and then uses a color enhancement network to enhance the image color and texture of the restored image to ensure image quality, thereby solving the problem that the current method of brightening images captured under low-light conditions has poor image processing effect.

[0037] The following is a detailed description of the image processing method provided in the embodiment of the present application:

[0038] See also Figure 1 , Figure 1The following is a schematic flow chart of an image processing method provided in an embodiment of the present application. The execution subject of the image processing method provided in an embodiment of the present application is a terminal device, which can be a mobile terminal such as a smart phone, a tablet computer or a wearable device, or a computer, a cloud server, an auxiliary computer, etc. in various application scenarios.

[0039] like Figure 1 The image processing method shown may include S11 to S12, which are described in detail as follows:

[0040] S11: Input the low-light image into the trained continuous update network to extract local features and global features to obtain feature images of different scales.

[0041] In an embodiment of the present application, local features and global features in a low-light image are extracted by continuously updating a network, and then the local features and the global features are fused to obtain feature images of different scales.

[0042] In an embodiment of the present application, the above-mentioned continuous update network may include N U-shaped network structures, and image features of different scales are extracted through the U-shaped network. The above-mentioned continuous update network also includes a feature fusion structure, and the image features of different scales extracted by the U-shaped network are fused through the feature fusion structure to finally obtain feature images of different scales.

[0043] It should be noted that N is a positive integer greater than or equal to 2.

[0044] In one embodiment of the present application, the above-mentioned continuous update network may include four U-shaped network structures, and the structure of each U-shaped network structure can be seen in Figure 2 , Figure 2 A structural schematic diagram of a U-shaped network structure of a continuously updated network in an embodiment of the present application is shown (illustrated by two U-shaped networks).

[0045] like Figure 2 As shown, the U-shaped network structure may include a downsampling structure 21 and an upsampling structure 22. The downsampling structure includes four feature blocks of different scales, and the four feature blocks of different scales (such as Figure 2 The scale ratio of A1, A2, A3, and A4 in the figure can be 8:4:2:1 from high to low, that is, the step size of the convolution of two adjacent feature blocks of the same scale is 1, and the step size of the convolution of two adjacent features with reduced scale is 2. The upsampling structure also includes four feature blocks of different scales. Figure 2 The scale ratio of B1, B2, B3, B4) in can be 1:2:4:8 from low to high, that is, the step size of the convolution of two adjacent feature blocks of the same scale is 1, and the step size of the increased deconvolution obtained by the scale of two adjacent feature blocks is 2.

[0046] Different feature blocks are connected using skip connections. Specifically, the different-scale features obtained by the downsampling structure are directly merged with the same-scale features obtained by the upsampling settlement, and the different-scale features obtained in the upsampling stage are fused with the larger-scale features through upsampling. Through skip connections, the upsampling stage can simultaneously obtain global features and local features.

[0047] For the combination of two U-shaped structures, the connection method can be to first double or triple the four features of different scales obtained by the two U-shaped networks, and then fuse the four features of the same scale obtained through a feature fusion structure (FF structure) to obtain a single feature. The obtained single feature will be retransmitted to the U-shaped network for continued operation. This connection method enables the features of the second U-shaped network to obtain more global features and local features.

[0048] In one embodiment of the present application, the above-mentioned feature fusion structure may include a convolutional layer, an average pooling layer and a fully connected layer.

[0049] For details, please refer to Figure 3 , Figure 3 A flow chart of the feature fusion process provided by an embodiment of the present application is shown. As shown in the image, in a specific application, the features extracted from the U-shaped structure are first input into the feature fusion module, and the convolution layer in the feature fusion module is processed to obtain an initial convolution, and then the initial convolution is passed through the average pooling layer to compress the features, and then a single eigenvalue is obtained, and then the correlation between different channels is determined by two fully connected layers to obtain a weight value, and then the obtained weight value is multiplied by the eigenvalue obtained after the initial convolution to obtain a first weighted feature, and the first weighted feature is added to the eigenvalue obtained after the initial convolution, and the above operation is repeated to obtain a second weighted eigenvalue, and finally, the second weighted feature is compressed to obtain the corresponding feature image after the fusion of the four features. Such operations are performed on features of different scales, and finally feature images of different scales can be obtained.

[0050] In an embodiment of the present application, the feature images of different scales include a first-scale feature image, a second-scale feature image, and a third-scale feature image.

[0051] S12: Inputting the feature images of different scales into the trained color enhancement network for processing to obtain a target image.

[0052] In an embodiment of the present application, the color enhancement network can determine the weight of each feature image based on the residuals of feature images of different scales, and then multiply each feature image by the corresponding weight to merge into one image, namely the target image.

[0053] In one embodiment of the present application, please refer to Figure 4 , Figure 4 Schematic diagram of the process of processing feature images of different scales by the color enhancement network provided in the embodiment of the present application. Figure 4 As shown, the color enhancement network 400 may include a residual acquisition module 401 , a convolution module 402 and a weight determination module 403 .

[0054] By inputting the first scale characteristic image Y1, the second scale characteristic image Y2, and the third scale characteristic image Y3 into the residual acquisition module 401, a first residual Y2-Y1, a second residual Y3-Y2, and a third residual Y3-Y1 can be obtained. Then the first residual Y2-Y1, the second residual Y3-Y2 and the third residual Y3-Y1 are input into the convolution unit for processing, and the first feature D1 and the second feature D2 are obtained. The first feature D1 and the second feature D2 are convolved to obtain the third feature D3 and the fourth feature D4. Then the second feature D2 and the fourth feature D4 are fused through the weight determination module, and the fifth feature D5 is convolved. The fifth feature D5 is then weight-decomposed through the weight determination module 403 to obtain the first weight G1, the second weight G2 and the third weight G3. The first weight G1 is multiplied by the first scale feature image Y1, the second weight G2 is multiplied by the second scale feature image Y2, and the third weight G3 is multiplied by the third scale feature image Y3. Then the multiplied images are fused to obtain the target image.

[0055] In another embodiment of the present application, the above image processing method may further include the following steps:

[0056] Constructing a continuous update network and a color enhancement network;

[0057] Get a sample dataset;

[0058] The continuous update network and the color enhancement network are trained based on the sample data set to obtain the trained continuous update network and the trained color enhancement network.

[0059] In the embodiment of the present application, a network structure can be constructed as follows: Figure 2 The continuously updated network shown and Figure 4 The color enhancement network shown is then trained using sample data to obtain a trained continuous update network and a color enhancement network.

[0060] In a specific application, the sample data may include historical low-light images and corresponding normal-light images.

[0061] In practical applications, no less than 1,000 sets of sample data can be selected to obtain a sample data set. The sample data set is divided into a training set, a validation set, and a test set. In order to meet the training requirements, 50% of the sample data can be used as a training set, and the rest as a validation set and a test set.

[0062] After obtaining the sample data, the continuously updated network is trained using the training set data, and the validation set is used to quickly adjust the parameters. The continuously updated network is then tested using the test set to obtain a trained continuously updated network.

[0063] When training the continuous update network, the historical low-light images in the sample images can be input into the pre-built continuous update network for processing to obtain feature images of different scales, and then the network parameters in the continuous update network can be adjusted based on the loss function of the normal light image and the obtained feature images of different scales. Specifically, the structural similarity loss and the total variation (TV) loss of the third-scale feature image and the corresponding normal light image can be used as the loss function for adjustment. When the loss function converges, the continuously updated network with adjusted network parameters is verified and tested based on the sample data in the verification set and the test set. If the verification and test pass, it means that the training of the continuous update network is completed, and the continuously updated network obtained after the training is completed can be used in S11.

[0064] It should be noted that since images taken under low-light conditions usually have obvious structural distortion problems, in order to quantitatively and qualitatively improve the quality of the image, the structural similarity (SSIM) loss is used. The normal-light image restored from the low-light image may have unstable lighting and noise, so the total variation loss is used as the smoothness prior to minimize the gradient of the entire image.

[0065] When training the color enhancement network, feature images of different scales output by the trained continuous update network can be input into a pre-constructed color enhancement network for processing to obtain a target image, and then the network parameters in the color enhancement network can be adjusted based on the loss function of the normal illumination image and the obtained target image. Specifically, the perceptual loss (perception loss) and the total variation (TV) loss of the target image and the corresponding normal illumination image can be used as the loss function to adjust the network parameters. When the loss function converges, the color enhancement network with adjusted network parameters is verified and tested based on the sample data in the verification set and the test set. Passing the verification and test means that the color enhancement network training is completed, and the color enhancement network obtained after the training can be used in S12.

[0066] It should be noted that the above-mentioned perceptual loss includes feature reconstruction loss and style reconstruction loss. The above-mentioned feature reconstruction loss can represent the Euclidean distance between features to measure the similarity of features. The style reconstruction loss is mainly to obtain better image color and texture. The style reconstruction loss is the square Frobenius norm of the difference between the output image (output target image) and the Gram matrix of the normal illumination image. The image that minimizes the style reconstruction loss retains the style features of the normal illumination image, but does not retain its spatial structure.

[0067] Perceptual loss is a loss function that combines feature reconstruction loss and style reconstruction loss. Similarly, in order to minimize the gradient of the entire image, the loss function of the color enhancement network is constructed based on total variation loss.

[0068] From the above, it can be seen that the image processing method provided in the embodiment of the present application extracts feature images of different scales by continuously updating the network, extracts global information and local information, and can ensure that the restored normal illumination image does not have the phenomenon of missing details, and then enhances the image color and texture of the restored image through the color enhancement network to ensure the image quality, thereby solving the problem that the current method of brightening images taken under low light conditions has poor image processing effect.

[0069] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0070] Based on the image processing method provided in the above embodiment, the embodiment of the present invention further provides an embodiment of a terminal device for implementing the above method embodiment.

[0071] See also Figure 5 , Figure 5 Schematic diagram of the structure of a terminal device provided in an embodiment of the present application. In the embodiment of the present application, each unit included in the terminal device is used to execute Figures 1 to 4 For details, please refer to the steps in the corresponding embodiment. Figures 1 to 4 as well as Figures 1 to 4 For the convenience of explanation, only the parts related to this embodiment are shown. Figure 5 As shown, the terminal device 50 includes: a feature extraction module 51 and a color enhancement module 52. Among them:

[0072] The feature extraction module 51 is used to input the low-light image into the trained continuous update network to extract local features and global features, and obtain feature images of different scales.

[0073] The color enhancement module 52 is used to input the feature images of different scales into the trained color enhancement network for processing to obtain the target image.

[0074] Optionally, the terminal device further includes:

[0075] Building modules for building continuous update networks and color enhancement networks;

[0076] An acquisition module is used to obtain a sample data set;

[0077] A training module is used to train the continuous update network and the color enhancement network based on a sample data set to obtain the trained continuous update network and the trained color enhancement network.

[0078] Optionally, the continuously updated network includes N U-shaped network structures and feature fusion structures, where N is a positive integer greater than or equal to 2.

[0079] Optionally, the U-shaped network structure includes a downsampling structure and an upsampling structure, the downsampling structure includes four feature blocks of different scales, the upsampling structure includes four feature blocks of different scales, and different feature blocks are connected using skip connections.

[0080] Optionally, the feature fusion structure includes a convolutional layer, an average pooling layer and a fully connected layer.

[0081] Optionally, the color enhancement network includes a residual acquisition module, a convolution module and a weight determination module.

[0082] It should be noted that the information interaction, execution process and other contents between the above-mentioned modules are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0083] Based on this, the terminal device provided in the embodiment of the present application can also extract feature images of different scales through a continuous updating network, extract global information and local information, and ensure that the restored normal lighting image will not have the phenomenon of missing details, and then enhance the image color and texture of the restored image through a color enhancement network to ensure image quality, thereby solving the problem of poor image processing effect in the current method of brightening images taken under low light conditions.

[0084] Figure 6 FIG. 1 is a schematic diagram of a terminal device provided in another embodiment of the present application. Figure 6 As shown, the terminal device 6 provided in this embodiment includes: a processor 60, a memory 61, and a computer program 62 stored in the memory 61 and executable on the processor 60, such as an image processing program. When the processor 60 executes the computer program 62, the steps in the above-mentioned various image processing method embodiments are implemented, such as Figure 1 Alternatively, when the processor 60 executes the computer program 62, the functions of the modules / units in the above-mentioned terminal device embodiments are realized, for example, Figure 5 The functions of the units 51-52 are shown.

[0085] Exemplarily, the computer program 62 may be divided into one or more modules / units, which are stored in the memory 61 and executed by the processor 60 to complete the present application. The one or more modules / units may be a series of computer program instruction segments that can complete specific functions, and the instruction segments are used to describe the execution process of the computer program 62 in the terminal device 6. For example, the computer program 62 may be divided into a first acquisition unit and a first processing unit. For specific functions of each unit, please refer to Figure 5 The relevant descriptions in the corresponding embodiments are not repeated here.

[0086] The terminal device may include, but is not limited to, a processor 60 and a memory 61. Those skilled in the art will appreciate that Figure 6 It is only an example of the terminal device 6 and does not constitute a limitation on the terminal device 6. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the terminal device may also include input and output devices, network access devices, buses, etc.

[0087] The processor 60 may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0088] The memory 61 may be an internal storage unit of the terminal device 6, such as a hard disk or memory of the terminal device 6. The memory 61 may also be an external storage device of the terminal device 6, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal device 6. Further, the memory 61 may also include both an internal storage unit and an external storage device of the terminal device 6. The memory 61 is used to store the computer program and other programs and data required by the terminal device. The memory 61 may also be used to temporarily store data that has been output or is to be output.

[0089] The present application also provides a computer-readable storage medium. Figure 7 , Figure 7 is a schematic diagram of the structure of a computer-readable storage medium provided in an embodiment of the present application, such as Figure 7 As shown, a computer program 71 is stored in a computer-readable storage medium 70, and the computer program 71 can implement the above-mentioned image processing method when executed by a processor.

[0090] An embodiment of the present application provides a computer program product. When the computer program product runs on a terminal device, the terminal device can implement the above-mentioned image processing method when executing the computer program product.

[0091] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the terminal device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0092] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0093] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0094] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. An image processing method, characterized in that: include: Inputting the low-light image into the trained continuous update network to extract local features and global features, and obtaining feature images of different scales; wherein the continuous update network includes N U-type network structures and feature fusion structures, wherein N is a positive integer greater than or equal to 2, extracting image features of different scales through the U-type network, and fusion of the image features of different scales extracted by the U-type network through the feature fusion structure, and finally obtaining feature images of different scales; The U-shaped network structure includes a downsampling structure and an upsampling structure, the downsampling structure includes four feature blocks of different scales, the scale ratio of the four feature blocks of different scales is 8:4:2:1 from high to low, that is, the step length of the convolution of two adjacent feature blocks of the same scale is 1, and the step length of the convolution of two adjacent features with reduced scale is 2; the upsampling structure also includes four feature blocks of different scales, the scale ratio of the four feature blocks of different scales is 1:2:4:8 from low to high, that is, the step length of the convolution of two adjacent feature blocks of the same scale is 1, and the step length of the increased deconvolution obtained by the scale of two adjacent feature blocks is 2; Different feature blocks are connected using skip connections. By directly merging the different scale features obtained by the downsampling structure with the same scale features obtained by the upsampling settlement, the different scale features obtained in the upsampling stage are fused with the larger scale features through upsampling. Through skip connections, the upsampling stage can obtain global features and local features at the same time. For the combination of two U-shaped structures, the connection method is to first amplify the four features of different scales obtained by the two U-shaped networks by one or two times, and then fuse the four features of the same scale obtained through the feature fusion structure to obtain a single feature. The obtained single feature will be retransmitted to the U-shaped network for further operation. This connection method enables the features of the second U-shaped network to obtain more global features and local features; The feature fusion structure includes a convolutional layer, an average pooling layer and a fully connected layer; First, the features extracted by the U-shaped structure are input into the feature fusion module, and the convolution layer in the feature fusion module is processed to obtain the initial convolution. Then, the initial convolution is passed through the average pooling layer to compress the features and obtain a single eigenvalue. Then, the correlation between different channels is determined through two fully connected layers to obtain the weight value. Then, the obtained weight value is multiplied by the eigenvalue obtained after the initial convolution to obtain the first weighted feature. The first weighted feature is added to the eigenvalue obtained after the initial convolution. The above operation is repeated to obtain the second weighted eigenvalue. Finally, the second weighted feature is compressed to obtain the corresponding feature image after the fusion of the four features. Such operation is performed on features of different scales to obtain feature images of different scales. The feature images of different scales are input into a trained color enhancement network for processing to obtain a target image.

2. The image processing method according to claim 1, characterized in that: Also includes: Constructing a continuous update network and a color enhancement network; Get a sample dataset; The continuous update network and the color enhancement network are trained based on the sample data set to obtain the trained continuous update network and the trained color enhancement network.

3. The image processing method according to claim 1, characterized in that: The color enhancement network includes a residual acquisition module, a convolution module and a weight determination module.

4. The image processing method according to any one of claims 1 to 3, characterized in that: The characteristic images of different scales include a first-scale characteristic image, a second-scale characteristic image, and a third-scale characteristic image.

5. A terminal device, characterized in that: include: A feature extraction module is used to input the low-light image into the trained continuous update network to extract local features and global features, and obtain feature images of different scales; wherein the continuous update network includes N U-type network structures and feature fusion structures, wherein N is a positive integer greater than or equal to 2, and image features of different scales are extracted through the U-type network, and the image features of different scales extracted by the U-type network are fused through the feature fusion structure, and finally feature images of different scales are obtained; The U-shaped network structure includes a downsampling structure and an upsampling structure, the downsampling structure includes four feature blocks of different scales, the scale ratio of the four feature blocks of different scales is 8:4:2:1 from high to low, that is, the step length of the convolution of two adjacent feature blocks of the same scale is 1, and the step length of the convolution of two adjacent features with reduced scale is 2; the upsampling structure also includes four feature blocks of different scales, the scale ratio of the four feature blocks of different scales is 1:2:4:8 from low to high, that is, the step length of the convolution of two adjacent feature blocks of the same scale is 1, and the step length of the increased deconvolution obtained by the scale of two adjacent feature blocks is 2; Different feature blocks are connected using skip connections. By directly merging the different scale features obtained by the downsampling structure with the same scale features obtained by the upsampling settlement, the different scale features obtained in the upsampling stage are fused with the larger scale features through upsampling. Through skip connections, the upsampling stage can obtain global features and local features at the same time. For the combination of two U-shaped structures, the connection method is to first amplify the four features of different scales obtained by the two U-shaped networks by one or two times, and then fuse the four features of the same scale obtained through the feature fusion structure to obtain a single feature. The obtained single feature will be retransmitted to the U-shaped network for further operation. This connection method enables the features of the second U-shaped network to obtain more global features and local features; The feature fusion structure includes a convolutional layer, an average pooling layer and a fully connected layer; First, the features extracted by the U-shaped structure are input into the feature fusion module, and the convolution layer in the feature fusion module is processed to obtain the initial convolution. Then, the initial convolution is passed through the average pooling layer to compress the features and obtain a single eigenvalue. Then, the correlation between different channels is determined through two fully connected layers to obtain the weight value. Then, the obtained weight value is multiplied by the eigenvalue obtained after the initial convolution to obtain the first weighted feature. The first weighted feature is added to the eigenvalue obtained after the initial convolution. The above operation is repeated to obtain the second weighted eigenvalue. Finally, the second weighted feature is compressed to obtain the corresponding feature image after the fusion of the four features. Such operation is performed on features of different scales to obtain feature images of different scales. The color enhancement module is used to input the feature images of different scales into the trained color enhancement network for processing to obtain a target image.

6. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer-readable instructions, the following steps are implemented: Inputting the low-light image into the trained continuous update network to extract local features and global features, and obtaining feature images of different scales; wherein the continuous update network includes N U-type network structures and feature fusion structures, wherein N is a positive integer greater than or equal to 2, extracting image features of different scales through the U-type network, and fusion of the image features of different scales extracted by the U-type network through the feature fusion structure, and finally obtaining feature images of different scales; The U-shaped network structure includes a downsampling structure and an upsampling structure, the downsampling structure includes four feature blocks of different scales, the scale ratio of the four feature blocks of different scales is 8:4:2:1 from high to low, that is, the step length of the convolution of two adjacent feature blocks of the same scale is 1, and the step length of the convolution of two adjacent features with reduced scale is 2; the upsampling structure also includes four feature blocks of different scales, the scale ratio of the four feature blocks of different scales is 1:2:4:8 from low to high, that is, the step length of the convolution of two adjacent feature blocks of the same scale is 1, and the step length of the increased deconvolution obtained by the scale of two adjacent feature blocks is 2; Different feature blocks are connected using skip connections. By directly merging the different scale features obtained by the downsampling structure with the same scale features obtained by the upsampling settlement, the different scale features obtained in the upsampling stage are fused with the larger scale features through upsampling. Through skip connections, the upsampling stage can obtain global features and local features at the same time. For the combination of two U-shaped structures, the connection method is to first amplify the four features of different scales obtained by the two U-shaped networks by one or two times, and then fuse the four features of the same scale obtained through the feature fusion structure to obtain a single feature. The obtained single feature will be retransmitted to the U-shaped network for further operation. This connection method enables the features of the second U-shaped network to obtain more global features and local features; The feature fusion structure includes a convolutional layer, an average pooling layer and a fully connected layer; First, the features extracted by the U-shaped structure are input into the feature fusion module, and the convolution layer in the feature fusion module is processed to obtain the initial convolution. Then, the initial convolution is passed through the average pooling layer to compress the features and obtain a single eigenvalue. Then, the correlation between different channels is determined through two fully connected layers to obtain the weight value. Then, the obtained weight value is multiplied by the eigenvalue obtained after the initial convolution to obtain the first weighted feature. The first weighted feature is added to the eigenvalue obtained after the initial convolution. The above operation is repeated to obtain the second weighted eigenvalue. Finally, the second weighted feature is compressed to obtain the corresponding feature image after the fusion of the four features. Such operation is performed on features of different scales to obtain feature images of different scales. The feature images of different scales are input into a trained color enhancement network for processing to obtain a target image.

7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the following steps are implemented: Inputting the low-light image into the trained continuous update network to extract local features and global features, and obtaining feature images of different scales; wherein the continuous update network includes N U-type network structures and feature fusion structures, wherein N is a positive integer greater than or equal to 2, extracting image features of different scales through the U-type network, and fusion of the image features of different scales extracted by the U-type network through the feature fusion structure, and finally obtaining feature images of different scales; The U-shaped network structure includes a downsampling structure and an upsampling structure, the downsampling structure includes four feature blocks of different scales, the scale ratio of the four feature blocks of different scales is 8:4:2:1 from high to low, that is, the step length of the convolution of two adjacent feature blocks of the same scale is 1, and the step length of the convolution of two adjacent features with reduced scale is 2; the upsampling structure also includes four feature blocks of different scales, the scale ratio of the four feature blocks of different scales is 1:2:4:8 from low to high, that is, the step length of the convolution of two adjacent feature blocks of the same scale is 1, and the step length of the increased deconvolution obtained by the scale of two adjacent feature blocks is 2; Different feature blocks are connected using skip connections. By directly merging the different scale features obtained by the downsampling structure with the same scale features obtained by the upsampling settlement, the different scale features obtained in the upsampling stage are fused with the larger scale features through upsampling. Through skip connections, the upsampling stage can obtain global features and local features at the same time. For the combination of two U-shaped structures, the connection method is to first amplify the four features of different scales obtained by the two U-shaped networks by one or two times, and then fuse the four features of the same scale obtained through the feature fusion structure to obtain a single feature. The obtained single feature will be retransmitted to the U-shaped network for further operation. This connection method enables the features of the second U-shaped network to obtain more global features and local features; The feature fusion structure includes a convolutional layer, an average pooling layer and a fully connected layer; First, the features extracted by the U-shaped structure are input into the feature fusion module, and the convolution layer in the feature fusion module is processed to obtain the initial convolution. Then, the initial convolution is passed through the average pooling layer to compress the features and obtain a single eigenvalue. Then, the correlation between different channels is determined through two fully connected layers to obtain the weight value. Then, the obtained weight value is multiplied by the eigenvalue obtained after the initial convolution to obtain the first weighted feature. The first weighted feature is added to the eigenvalue obtained after the initial convolution. The above operation is repeated to obtain the second weighted eigenvalue. Finally, the second weighted feature is compressed to obtain the corresponding feature image after the fusion of the four features. Such operation is performed on features of different scales to obtain feature images of different scales. The feature images of different scales are input into a trained color enhancement network for processing to obtain a target image.

Images