An infrared camera intelligent target recognition method, system and device

Through the combination of U-net convolutional neural network model and significance extraction operator, the problem of low signal-to-noise ratio and lack of shape and texture information is solved, and the accuracy of infrared image target recognition is improved, adapting to different scenarios and target sizes is achieved. .

CN118397428BActive Publication Date: 2024-12-03SHENZHEN TANGE INTELLIGENT TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410569962.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-09
Publication Date
2024-12-03
Estimated Expiration
2044-05-09

AI Technical Summary

Technical Problem

The signal-to-noise ratio of infrared images is low, which makes it difficult to recognize the target under complex backgrounds, and the lack of shape and texture information increases the difficulty of recognition. In the prior art, infrared image target recognition accuracy is not high.

Method used

U-net convolutional neural network model combined with the significance extraction operator, calculate the significance value of each pixel and normalize it, obtain the significance weight value, multiply it with the characteristic image pixel, and make a jump connection to enhance Features and use 1×1 convolution and soft max layers for target recognition at the end of the network.

Benefits of technology

The accuracy of infrared image target recognition can be improved, and it can better adapt to different scenes and target sizes, improving the accuracy of recognition.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118397428B_ABST
    Figure CN118397428B_ABST
Patent Text Reader

Abstract

The present application provides an infrared camera intelligent target recognition method, system and device, including: based on the pixel value of the infrared image to be recognized, calculating the saliency value corresponding to each pixel, normalizing the saliency value to obtain a saliency weight value; the infrared image to be recognized is input into the U-net convolutional neural network model, obtaining a first feature map output by the first downsampling structure of the U-net convolutional neural network model, a second feature map output by the second downsampling structure, and a third feature map output by the third downsampling structure, multiplying the pixels of the first feature map, the second feature map, and the third feature map with the saliency weight values ​​corresponding to their positions to obtain a first enhanced feature, a second enhanced feature, and a third enhanced feature, performing jump connection between the first enhanced feature, the second enhanced feature, and the third enhanced feature and the corresponding symmetric path upsampling layer, and using a 1×1 convolution and soft max layer at the end of the network to obtain a target recognition result. It can at least be used to solve the technical problem that the accuracy of target recognition of infrared images in the prior art is not high.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of target recognition, and in particular to an infrared camera intelligent target recognition method system and device. Background Art

[0002] Infrared imaging technology does not need to rely on visible light sources and can work normally at night or in low light conditions. Infrared rays have good penetration and can penetrate obstacles such as smoke and dust to obtain information about obscured objects, making it widely used. However, infrared images usually have a low signal-to-noise ratio, which means that the ratio between the target signal and the background noise in the image is low, making it difficult to clearly identify the target in the image. This situation is particularly prominent when the background information is complex and the edges are unclear. Since the grayscale distribution of infrared image targets is low and the detail difference is small, when the background is disturbed by a lot of noise and clutter, the edge information of the image target may be submerged, thereby increasing the difficulty of target detection and recognition. Secondly, infrared image target recognition is also affected by the small imaging area of ​​weak targets. Due to the lack of structural information such as shape and texture, the information that the system can use is mainly the local grayscale continuity and motion continuity of the target, which increases the difficulty of target recognition. The accuracy of target recognition by infrared images in the existing technology needs to be improved urgently. Summary of the invention

[0003] The present application provides an infrared camera intelligent target recognition method, system and device, which are at least used to solve the problem of insufficient accuracy when using infrared images for target recognition in the prior art.

[0004] In a first aspect, the present application provides an infrared camera intelligent target recognition method, comprising:

[0005] Step S1, constructing a U-net convolutional neural network model;

[0006] Step S2, photographing an infrared image to be identified;

[0007] Step S3, based on the pixel values ​​of the infrared image to be identified, calculating the saliency value corresponding to each pixel, and normalizing the saliency value to obtain a saliency weight value;

[0008] Step S4, the infrared image to be identified is input into the U-net convolutional neural network model to obtain a first feature map output by a first downsampling structure, a second feature map output by a second downsampling structure, and a third feature map output by a third downsampling structure of the U-net convolutional neural network model, and the pixels of the first feature map, the second feature map, and the third feature map are multiplied by the significance weight values ​​corresponding to their positions to obtain a first enhanced feature, a second enhanced feature, and a third enhanced feature, and the first enhanced feature, the second enhanced feature, and the third enhanced feature are jump-connected to the corresponding symmetric path upsampling layer, and a 1×1 convolution and soft max layer are used at the end of the network to obtain a target recognition result;

[0009] The calculation of the saliency value corresponding to each pixel in step S3 is specifically as follows:

[0010] Step S31, constructing a filter template, wherein the filter template is a square and consists of N×N sub-blocks, and the size of each sub-block is 8×8 pixels;

[0011] Step S32, using 3×3 sub-blocks in the filtering template as processing units, the infrared image to be identified corresponding to the central sub-block of the processing unit is the target area to be processed, and the infrared image to be identified corresponding to the remaining sub-blocks is the background area;

[0012] Step S33, calculating the saliency value of the pixel corresponding to the target area based on the pixel values ​​of the target area and the background area;

[0013] Step S34, sliding the filter template to repeat steps S31 and S32 until the infrared image to be identified is processed, thereby obtaining a saliency value of each pixel.

[0014] In a second aspect, the present application provides an infrared camera intelligent target recognition system, comprising:

[0015] Infrared detector, used for capturing infrared images to be identified;

[0016] A first processing module, configured to calculate a saliency value corresponding to each pixel based on the pixel value of the infrared image to be identified, and normalize the saliency value to obtain a saliency weight value;

[0017] A second processing module is used to input the infrared image to be identified into a U-net convolutional neural network model, obtain a first feature map output by a first downsampling structure of the U-net convolutional neural network model, a second feature map output by a second downsampling structure, and a third feature map output by a third downsampling structure, multiply the pixels of the first feature map, the second feature map, and the third feature map by the saliency weight values ​​corresponding to their positions to obtain a first enhanced feature, a second enhanced feature, and a third enhanced feature, perform jump connections between the first enhanced feature, the second enhanced feature, and the third enhanced feature and the corresponding symmetric path upsampling layer, and use a 1×1 convolution and soft max layer at the end of the network to obtain a target recognition result;

[0018] The calculation of the saliency value corresponding to each pixel is specifically as follows:

[0019] Step S31, constructing a filter template, wherein the filter template is a square and consists of N×N sub-blocks, and the size of each sub-block is 8×8 pixels;

[0020] Step S32, using 3×3 sub-blocks in the filtering template as processing units, the infrared image to be identified corresponding to the central sub-block of the processing unit is the target area to be processed, and the infrared image to be identified corresponding to the remaining sub-blocks is the background area;

[0021] Step S33, calculating the saliency value of the pixel corresponding to the target area based on the pixel values ​​of the target area and the background area;

[0022] Step S34, sliding the filter template to repeat steps S31 and S32 until the infrared image to be identified is processed, thereby obtaining a saliency value of each pixel.

[0023] In a third aspect, some embodiments of the present application further provide a terminal device, comprising: one or more processors; and a memory storing computer program instructions, wherein the computer program instructions, when executed, cause the processor to perform the above method.

[0024] Compared with the prior art, in the solution provided in the embodiment of the present application, the infrared image to be identified is input into the U-net convolutional neural network model to obtain a first feature map output by the first downsampling structure of the U-net convolutional neural network model, a second feature map output by the second downsampling structure, and a third feature map output by the third downsampling structure. The pixels of the first feature map, the second feature map, and the third feature map are multiplied by the significance weight values ​​corresponding to their positions to obtain a first enhanced feature, a second enhanced feature, and a third enhanced feature. The first enhanced feature, the second enhanced feature, and the third enhanced feature are jump-connected to the corresponding symmetric path upsampling layer. A 1×1 convolution and soft max layer are used at the end of the network to obtain a target recognition result. The specific method of calculating the saliency value corresponding to each pixel is as follows: step S31, constructing a filter template, the filter template is a square composed of N×N sub-blocks, and the size of each sub-block is 8×8 pixels; step S32, using 3×3 sub-blocks in the filter template as processing units, the infrared image to be identified corresponding to the central sub-block of the processing unit is the target area to be processed, and the infrared image to be identified corresponding to the remaining sub-blocks is the background area; step S33, based on the pixel values ​​of the target area and the background area, the saliency value of the pixel corresponding to the target area is calculated; step S34, sliding the filter template to repeat steps S31 and S32 until the infrared image to be identified is processed, and then obtaining the saliency value of each pixel. The present application sets the saliency map extracted from the infrared image as the attention weight information of the down-sampled features of the infrared image extracted by the U-net network for supervision and then jump-connecting with the up-sampling layer of the symmetric path, which can well adapt to the target sizes of different scenes and improve the target recognition accuracy of infrared images. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 The overall flow chart of the intelligent target recognition method provided in the embodiment of the present application;

[0026] Figure 2 A schematic diagram of the overall process of obtaining a saliency value map provided in an embodiment of the present application.

[0027] Figure 3 A schematic diagram of a partial process for obtaining a saliency value map provided in an embodiment of the present application

[0028] Figure 4 A flow chart of an improved convolutional neural network training method provided in an embodiment of the present application;

[0029] Figure 5 A system architecture diagram provided for an embodiment of the present application;

[0030] Figure 6 A schematic diagram of a device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0032] Embodiment 1

[0033] An intelligent target recognition method for infrared cameras, such as Figure 1 As shown, including:

[0034] Step S1, constructing a U-net convolutional neural network model;

[0035] Step S2, photographing an infrared image to be identified;

[0036] Step S3, based on the pixel values ​​of the infrared image to be identified, calculating the saliency value corresponding to each pixel, and normalizing the saliency value to obtain a saliency weight value;

[0037] Step S4, the infrared image to be identified is input into the U-net convolutional neural network model to obtain a first feature map output by a first downsampling structure, a second feature map output by a second downsampling structure, and a third feature map output by a third downsampling structure of the U-net convolutional neural network model, and the pixels of the first feature map, the second feature map, and the third feature map are multiplied by the significance weight values ​​corresponding to their positions to obtain a first enhanced feature, a second enhanced feature, and a third enhanced feature, and the first enhanced feature, the second enhanced feature, and the third enhanced feature are jump-connected to the corresponding symmetric path upsampling layer, and a 1×1 convolution and soft max layer are used at the end of the network to obtain a target recognition result;

[0038] The specific calculation of the saliency value corresponding to each pixel in step S3 is (eg Figure 2 shown):

[0039] Step S31, constructing a filter template, wherein the filter template is a square and consists of N×N sub-blocks, and the size of each sub-block is 8×8 pixels;

[0040] Step S32, using 3×3 sub-blocks in the filtering template as processing units, the infrared image to be identified corresponding to the central sub-block of the processing unit is the target area to be processed, and the infrared image to be identified corresponding to the remaining sub-blocks is the background area;

[0041] Step S33, calculating the saliency value of the pixel corresponding to the target area based on the pixel values ​​of the target area and the background area;

[0042] Step S34, sliding the filter template to repeat steps S31 and S32 until the infrared image to be identified is processed, thereby obtaining a saliency value of each pixel.

[0043] In some embodiments of the present application,

[0044] like Figure 3 As shown, step S33 includes:

[0045] Step S331: Using the processing unit as a measurement window, sort all pixels of the sub-block corresponding to the processing unit by grayscale value, take the maximum and second largest grayscale values ​​to calculate the average, and obtain M c ;

[0046] Step S332: Calculate the mean grayscale value of the pixels in the sub-block corresponding to the background area to obtain M l ; Get three sub-blocks corresponding to one of the eight directions of the target area to be processed, and then get three sub-blocks M l The mean value is M k , where k represents the sequence number of the direction, 1≤k≤8; the 8 directions are: up, down, left, right, upper left, upper right, lower left, and lower right;

[0047] Step S333: M c Value and M k The difference is taken as the significant value of the kth direction, and then the significant values ​​of the 8 directions are obtained;

[0048] Step S334: take the target area to be processed as a pixel of the saliency map, then take the absolute values ​​of the saliency values ​​in the eight directions and sum them up as the saliency value of the pixel.

[0049] In some embodiments of the present application, the U-net convolutional neural network model is composed of an encoder on the left and a decoder on the right. The encoder includes three downsampling layers. The infrared image to be identified is processed by the first downsampling layer to obtain the first feature map, and the first feature map is input into the second downsampling layer to obtain the second feature map. The second feature map is input into the third downsampling layer to obtain the third feature map, wherein the first downsampling layer, the second downsampling layer, and the third downsampling layer, each layer performs two consecutive 3×3 convolutions, BN layers, and Relu activation functions on the feature map input from the previous layer, and then uses 2×2 maximum pooling for two times downsampling. In the decoder, three upsampling layers are included corresponding to the encoder, and each layer uses deconvolution to perform two times upsampling on the feature map input from the upper layer to restore the compressed features, and the corresponding features of the upsampling layer of the encoder symmetric path are merged through jump connections, and the merged feature map is sent to the next layer after two 3×3 convolutions, BN layers, and Relu activation functions. A 1×1 convolution and soft max layer are used at the end of the network to output the recognition result.

[0050] In some embodiments of the present application, Figure 4 The U-net convolutional neural network model training adopts the following steps:

[0051] S101, collect infrared images of different targets, annotate target locations and categories, and create a data set;

[0052] S102, dividing the data set into a training set and a test set;

[0053] S103, input the training set into the U-net convolutional neural network model training iteration until the set training times are reached and the training is stopped, and then the test set is used for verification. If the accuracy meets the requirements, a trained U-net convolutional neural network model is obtained.

[0054] In some embodiments of the present application, after step S2 and before step S3, the method further includes performing liveness detection on the infrared image to be identified, wherein the liveness detection includes: a facial biopsy model and a palm biopsy model;

[0055] Step S210, calculating a difference image between the infrared image to be identified and the previous frame image based on a frame difference method;

[0056] Step S220, inputting the difference image into the facial biopsy model and the palm biopsy model;

[0057] Step S230: If the facial living body model and the palm biopsy model determine that there is a living body, target recognition is performed.

[0058] Embodiment 2

[0059] An infrared camera intelligent target recognition system, such as Figure 5 As shown, including:

[0060] Infrared detector 501, used for capturing infrared images to be identified;

[0061] A first processing module 504 is used to calculate a saliency value corresponding to each pixel based on the pixel value of the infrared image to be identified, and normalize the saliency value to obtain a saliency weight value;

[0062] The second processing module 505 is used to input the infrared image to be identified into the U-net convolutional neural network model, obtain the first feature map output by the first downsampling structure of the U-net convolutional neural network model, the second feature map output by the second downsampling structure, and the third feature map output by the third downsampling structure, multiply the pixels of the first feature map, the second feature map, and the third feature map by the saliency weight values ​​corresponding to their positions to obtain the first enhanced feature, the second enhanced feature, and the third enhanced feature, perform jump connection between the first enhanced feature, the second enhanced feature, and the third enhanced feature and the corresponding symmetric path upsampling layer, and use a 1×1 convolution and soft max layer at the end of the network to obtain the target recognition result;

[0063] The calculation of the saliency value corresponding to each pixel is specifically as follows:

[0064] Step S31, constructing a filter template, wherein the filter template is a square and consists of N×N sub-blocks, and the size of each sub-block is 8×8 pixels;

[0065] Step S32, using 3×3 sub-blocks in the filtering template as processing units, the infrared image to be identified corresponding to the central sub-block of the processing unit is the target area to be processed, and the infrared image to be identified corresponding to the remaining sub-blocks is the background area;

[0066] Step S33, calculating the saliency value of the pixel corresponding to the target area based on the pixel values ​​of the target area and the background area;

[0067] Step S34, sliding the filter template to repeat steps S31 and S32 until the infrared image to be identified is processed, thereby obtaining a saliency value of each pixel.

[0068] The infrared camera intelligent target recognition system also includes:

[0069] A liveness detection module 502, used for performing liveness detection on the infrared image to be identified;

[0070] The determination module 503 is used to determine whether there is a living body, and trigger the first processing module 504 and the second processing module 505 when it is determined that there is a living body.

[0071] Embodiment 3

[0072] like Figure 6 As shown, it is a schematic diagram of a terminal device provided in an embodiment of the present application, and the terminal device 6 may include: a processor 601, a memory 602, and a computer program 603 stored in the memory 602 and executable on the processor 601, such as a target recognition loader. When the processor 601 executes the computer program 603, the steps in the above-mentioned target recognition method embodiments are implemented.

[0073] Beneficial effects:

[0074] This application uses a designed saliency extraction operator to extract the saliency map information of the infrared image and fuse it with the down-sampling output features of the U-net convolutional neural network. The fused features are used as jump connections with the upsampling layer on the symmetric path. The saliency map information is used as the attention weight information, which can better adapt to targets of different sizes and improve the accuracy of infrared image target recognition.

[0075] It should be noted that the present application can be implemented in software and / or a combination of software and hardware, for example, can be implemented using an application specific integrated circuit (ASIC), a general purpose computer or any other similar hardware device. In certain embodiments, the software program of the present application can be executed by a processor to implement the above steps or functions. Similarly, the software program of the present application (including relevant data structures) can be stored in a computer-readable recording medium, for example, a RAM memory, a magnetic or optical drive or a floppy disk and similar devices. In addition, some steps or functions of the present application can be implemented using hardware.

Claims

1. An infrared camera intelligent target recognition method, characterized in that: include: Step S1, constructing a U-net convolutional neural network model; Step S2, photographing an infrared image to be identified; Step S3, based on the pixel values ​​of the infrared image to be identified, calculating the saliency value corresponding to each pixel, and normalizing the saliency value to obtain a saliency weight value; Step S4, the infrared image to be identified is input into the U-net convolutional neural network model to obtain a first feature map output by a first downsampling structure, a second feature map output by a second downsampling structure, and a third feature map output by a third downsampling structure of the U-net convolutional neural network model, and the pixels of the first feature map, the second feature map, and the third feature map are multiplied by the significance weight values ​​corresponding to their positions to obtain a first enhanced feature, a second enhanced feature, and a third enhanced feature, and the first enhanced feature, the second enhanced feature, and the third enhanced feature are jump-connected to the corresponding symmetric path upsampling layer, and a 1×1 convolution and soft max layer are used at the end of the network to obtain a target recognition result; The calculation of the saliency value corresponding to each pixel in step S3 is specifically as follows: Step S31, constructing a filter template, wherein the filter template is a square and consists of N×N sub-blocks, and the size of each sub-block is 8×8 pixels; Step S32, using 3×3 sub-blocks in the filtering template as processing units, the infrared image to be identified corresponding to the central sub-block of the processing unit is the target area to be processed, and the infrared image to be identified corresponding to the remaining sub-blocks is the background area; Step S33, calculating the saliency value of the pixel corresponding to the target area based on the pixel values ​​of the target area and the background area; Step S34, sliding the filter template to repeat steps S31 and S32 until the infrared image to be identified is processed, thereby obtaining a saliency value of each pixel; Wherein, the step S33 comprises: Step S331: Using the processing unit as a measurement window, sort all pixels of the sub-block corresponding to the processing unit by grayscale value, take the maximum and second largest grayscale values ​​to calculate the average, and obtain M c ; Step S332: Calculate the mean grayscale value of the pixels in the sub-block corresponding to the background area to obtain M l ; Get three sub-blocks corresponding to one of the eight directions of the target area to be processed, and then get three sub-blocks M l The mean value is M k , where k represents the sequence number of the direction, 1≤k≤8; the 8 directions are: up, down, left, right, upper left, upper right, lower left, and lower right; Step S333: M c Value and M k The difference is taken as the significant value of the kth direction, and then the significant values ​​of the 8 directions are obtained; Step S334: take the target area to be processed as a pixel of the saliency map, then take the absolute values ​​of the saliency values ​​in the eight directions and sum them up as the saliency value of the pixel.

2. The method according to claim 1, characterized in that: The U-net convolutional neural network model is composed of an encoder on the left and a decoder on the right. The encoder includes three downsampling layers. The infrared image to be identified is processed by the first downsampling layer to obtain the first feature map, the first feature map is input into the second downsampling layer to obtain the second feature map, and the second feature map is input into the third downsampling layer to obtain the third feature map. In the first downsampling layer, the second downsampling layer, and the third downsampling layer, each layer performs two consecutive 3×3 convolutions, BN layers, and Relu activation functions on the feature map input from the previous layer, and then uses 2×2 maximum pooling for two times downsampling. In the decoder, three upsampling layers are included corresponding to the encoder, each layer uses deconvolution to perform two times upsampling on the feature map input from the previous layer to restore the compressed features, and the corresponding features of the upsampling layer of the encoder symmetric path are merged through jump connections, and the merged feature map is sent to the next layer after two 3×3 convolutions, BN layers, and Relu activation functions. A 1×1 convolution and soft max layer are used at the end of the network to output the recognition result.

3. The method according to claim 1, characterized in that After step S2 and before step S3, the method further includes performing liveness detection on the infrared image to be identified, wherein the liveness detection includes: a facial biopsy model and a palm biopsy model; Step S210, calculating a difference image between the infrared image to be identified and the previous frame image based on a frame difference method; Step S220, inputting the difference image into the facial biopsy model and the palm biopsy model; Step S230: Target recognition is performed only if the facial biopsy model and the palm biopsy model determine that a living body exists.

4. The method according to claim 1, characterized in that: The U-net convolutional neural network model training adopts the following steps: S101, collect infrared images of different targets, annotate target locations and categories, and create a data set; S102, dividing the data set into a training set and a test set; S103, input the training set into the U-net convolutional neural network model training iteration until the set training times are reached and the training is stopped, and then the test set is used for verification. If the accuracy meets the requirements, a trained U-net convolutional neural network model is obtained.

5. An infrared camera intelligent target recognition system, characterized in that: include: Infrared detector, used for capturing infrared images to be identified; A first processing module, configured to calculate a saliency value corresponding to each pixel based on the pixel value of the infrared image to be identified, and normalize the saliency value to obtain a saliency weight value; A second processing module is used to input the infrared image to be identified into a U-net convolutional neural network model, obtain a first feature map output by a first downsampling structure of the U-net convolutional neural network model, a second feature map output by a second downsampling structure, and a third feature map output by a third downsampling structure, multiply the pixels of the first feature map, the second feature map, and the third feature map by the saliency weight values ​​corresponding to their positions to obtain a first enhanced feature, a second enhanced feature, and a third enhanced feature, perform jump connections between the first enhanced feature, the second enhanced feature, and the third enhanced feature and the corresponding symmetric path upsampling layer, and use a 1×1 convolution and soft max layer at the end of the network to obtain a target recognition result; The calculation of the saliency value corresponding to each pixel is specifically as follows: Step S31, constructing a filter template, wherein the filter template is a square and consists of N×N sub-blocks, and the size of each sub-block is 8×8 pixels; Step S32, using 3×3 sub-blocks in the filtering template as processing units, the infrared image to be identified corresponding to the central sub-block of the processing unit is the target area to be processed, and the infrared image to be identified corresponding to the remaining sub-blocks is the background area; Step S33, calculating the saliency value of the pixel corresponding to the target area based on the pixel values ​​of the target area and the background area; Step S34, sliding the filter template to repeat steps S31 and S32 until the infrared image to be identified is processed, thereby obtaining a saliency value of each pixel; Wherein, the step S33 comprises: Step S331: Using the processing unit as a measurement window, sort all pixels of the sub-block corresponding to the processing unit by grayscale value, take the maximum and second largest grayscale values ​​to calculate the average, and obtain M c ; Step S332: Calculate the mean grayscale value of the pixels in the sub-block corresponding to the background area to obtain M l ; Get three sub-blocks corresponding to one of the eight directions of the target area to be processed, and then get three sub-blocks M l The mean value is M k , where k represents the sequence number of the direction, 1≤k≤8; the 8 directions are: up, down, left, right, upper left, upper right, lower left, and lower right; Step S333: M c Value and M k The difference is taken as the significant value of the kth direction, and then the significant values ​​of the 8 directions are obtained; Step S334: take the target area to be processed as a pixel of the saliency map, then take the absolute values ​​of the saliency values ​​in the eight directions and sum them up as the saliency value of the pixel.

6. A terminal device, characterized in that: The device comprises: one or more processors; and a memory storing computer program instructions, wherein when the computer program instructions are executed, the processors perform the method according to any one of claims 1 to 4.