Artillery barrel target image defogging method and system based on deep learning

Abstract

The application discloses a deep learning-based artillery barrel target image defogging method and system, which comprises the following steps: inputting an original artillery barrel target image into a trained defogging enhancement network model for processing, wherein the defogging enhancement network model comprises an improved U-Net module, an AOD-Net module and an atmospheric scattering model layer connected in sequence; the improved U-Net module is used to extract multi-scale features of the original artillery barrel target image, and output the multi-scale feature maps to the AOD-Net module; the AOD-Net module is used to extract features under different receptive fields of the multi-scale feature maps through different convolution layers, realize multi-scale feature fusion, and output input values of the atmospheric scattering model layer; and the atmospheric scattering model layer is used to obtain the defogged artillery barrel target image based on the input values. The model is constructed by connecting the U-Net and the AOD-Net in series, a new model optimizer is proposed, the algorithm of gradient descent is optimized, and a self-adaptive weight loss function is introduced, so that efficient defogging processing under complex fog and haze conditions is realized.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method and system for dehazing artillery barrel target images based on deep learning. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] The attitude of the artillery barrel is a key factor affecting the accuracy of artillery firing. With the continuous development of visual measurement technology, the visual measurement method of barrel pointing based on target images has a very broad application prospect due to its non-contact nature, high measurement accuracy, and fast response speed. However, the outdoor measurement of artillery barrel attitude is often affected by various complex weather conditions, especially the interference of fog and haze. Under adverse weather conditions such as dense fog and haze, the quality of the acquired barrel target images will be severely degraded, posing a great challenge to subsequent image analysis, target extraction, and attitude measurement. Therefore, how to efficiently and accurately remove fog and haze from target images and improve image clarity has become an urgent problem to be solved in the visual measurement of artillery attitude.

[0004] Image dehazing is essentially an image restoration problem, that is, recovering the original clear image from a disturbed image. Traditional physical model methods (such as dark channel prior methods) require a relatively detailed understanding of the imaging environment and physical conditions. While these methods can remove haze to some extent, they often perform poorly in complex scenes. Deep learning, on the other hand, can learn effective dehazing features from hazy images through large-scale data training, thus eliminating its dependence on physical prior information. However, existing deep learning methods still suffer from insufficient detail recovery and target-background image imbalance when handling image restoration under complex hazy conditions. Summary of the Invention

[0005] To overcome the shortcomings of the existing technologies, this invention provides a method and system for dehazing artillery barrel target images based on deep learning. By concatenating U-Net and AOD-Net and proposing a new model optimizer, the gradient descent algorithm is optimized. At the same time, an adaptive weight loss function is introduced to achieve efficient dehazing processing under complex fog and haze conditions.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0007] In a first aspect, the present invention provides a method for dehazing artillery barrel target images based on deep learning, comprising:

[0008] Acquire the original image of the artillery barrel target;

[0009] The original gun barrel target image is input into a pre-trained defogging enhancement network model for processing to obtain a defogging gun barrel target image.

[0010] The defogging enhancement network model includes an improved U-Net module, an AOD-Net module, and an atmospheric scattering model layer connected in sequence. The improved U-Net module extracts multi-scale features from the original artillery barrel target image and outputs a multi-scale feature map to the AOD-Net module. The AOD-Net module extracts features from different receptive fields of the multi-scale feature map through different convolutional layers to achieve multi-scale feature fusion and outputs the input value of the atmospheric scattering model layer. The atmospheric scattering model layer obtains the defogging artillery barrel target image based on the input value.

[0011] A further technical solution involves a lightweight design of the U-Net network structure in the improved U-Net module, which reduces its sampling process.

[0012] A further technical solution is that the atmospheric scattering model layer is implemented by an atmospheric scattering model, which is represented as follows:

[0013] I(x)=J(x)·t(x)+A·(1-t(x))

[0014] J(x)=K(x)I(x)-K(x)+b

[0015]

[0016] Where b is a constant deviation with a default value of 1; I(x) represents the pixel value of the foggy image captured by the imaging sensor at the pixel; J(x) represents the pixel value of the real scene image without fog; A represents atmospheric light; and t(x) represents transmittance.

[0017] A further technical solution involves designing an optimizer to shorten the training time during the training of the dehazing enhancement network model.

[0018] A further technical solution is that the optimizer is represented as:

[0019]

[0020] Where l is the slope of the gradient descent curve, Ir is the step size, and t is the threshold.

[0021] In a further technical solution, the dehazing enhancement network model is designed with an adaptive weight loss function to constrain the output.

[0022] A further technical solution is that the loss function is expressed as:

[0023]

[0024] Where w(x) is the adaptive weight, and J(x) represents the pixel value of the real scene image without fog.

[0025] Secondly, the present invention provides a deep learning-based dehazing system for artillery barrel target images, comprising:

[0026] The image acquisition module is configured to acquire raw images of the artillery barrel target.

[0027] The image dehazing module is configured to input the original artillery barrel target image into a pre-trained dehazing enhancement network model for processing, and obtain a dehazed artillery barrel target image.

[0028] The defogging enhancement network model includes an improved U-Net module, an AOD-Net module, and an atmospheric scattering model layer connected in sequence. The improved U-Net module extracts multi-scale features from the original artillery barrel target image and outputs a multi-scale feature map to the AOD-Net module. The AOD-Net module extracts features from different receptive fields of the multi-scale feature map through different convolutional layers to achieve multi-scale feature fusion and outputs the input value of the atmospheric scattering model layer. The atmospheric scattering model layer obtains the defogging artillery barrel target image based on the input value.

[0029] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the deep learning-based dehazing method for artillery barrel target images as described in the first aspect.

[0030] Fourthly, the present invention 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 program to implement the steps in the deep learning-based dehazing method for artillery barrel target images as described in the first aspect.

[0031] The above one or more technical solutions have the following beneficial effects:

[0032] This invention constructs a dehazing enhancement network model by concatenating U-Net and AOD-Net. U-Net can extract rich multi-scale features, enabling the model to perform well when processing targets of different sizes. Through upsampling and downsampling operations, U-Net can not only capture global information but also preserve local details and edge information of the image, which is particularly important for dehazing tasks. By concatenating it with AOD-Net, global information and local detail information of foggy target images can be restored.

[0033] This invention considers both improving the dehazing effect and highlighting image details in artillery barrel target images, as well as addressing the efficiency of image processing. Two optimizations are made: First, the U-Net structure is lightweighted by reducing the sampling process from four to two, thus accelerating image processing and reducing information loss due to sampling. Second, a new model optimizer is proposed, optimizing the gradient descent algorithm to enable the model to fit the objective function (loss function) to the training data more quickly and effectively and converge to the optimal solution, significantly reducing model training time.

[0034] This invention designs an adaptive weight loss function for the defogging enhancement network model, dynamically adjusting the weights according to the importance of different regions in the target image of the artillery barrel, paying particular attention to the detail restoration of the target region, thereby improving the defogging effect.

[0035] The method provided by this invention can adapt to different haze conditions, especially when the haze concentration is unevenly distributed, and can flexibly adjust the dehazing process for each area to ensure the overall image quality. Attached Figure Description

[0036] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0037] Figure 1 This is a flowchart illustrating the image dehazing method according to an embodiment of the present invention;

[0038] Figure 2 This is a structural diagram of the dehazing enhancement network model with AOD-Net and U-Net cascaded in an embodiment of the present invention;

[0039] Figure 3 This is a comparison chart of the output results of the image dehazing method provided in the embodiments of the present invention. Detailed Implementation

[0040] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0041] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0042] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0043] Example 1

[0044] like Figure 1 As shown in the figure, this embodiment discloses a method for dehazing artillery barrel target images based on deep learning. The method includes the following steps:

[0045] S1: Acquire the original image of the artillery barrel target;

[0046] In this embodiment, images of artillery barrel targets under various weather conditions, including clear skies and fog, are collected in the field to obtain clear images of artillery barrels with fog. Simultaneously, foggy images of the targets (i.e., foggy artillery target images) are simulated and generated, together constructing a dataset. The controllability of the simulated foggy images allows for more accurate evaluation of the performance of the defogging enhancement network model.

[0047] In this implementation example, the dataset contains 5400 images, each 648×648 pixels. The dataset is divided into training and testing sets for model training. The dataset is input into the dehazing enhancement network model for 90 training iterations. Then, a loss function is used to constrain the output of each training iteration, adjusting the parameters of the dehazing enhancement network model to obtain a trained dehazing enhancement network model for artillery barrel target images.

[0048] When acquiring images of artillery barrel targets with fog (i.e., original artillery barrel target images) under adverse weather conditions such as dense fog or haze, it is necessary to efficiently and accurately remove the fog and haze from the target images to improve image clarity.

[0049] S2: Input the original gun barrel target image into the pre-trained defogging enhancement network model for processing to obtain the defogging gun barrel target image;

[0050] The defogging enhancement network model includes an improved U-Net module, an AOD-Net module, and an atmospheric scattering model layer connected in sequence. The improved U-Net module extracts multi-scale features from the original artillery barrel target image and outputs a multi-scale feature map to the AOD-Net module. The AOD-Net module extracts features from different receptive fields of the multi-scale feature map through different convolutional layers to achieve multi-scale feature fusion and outputs the input value of the atmospheric scattering model layer. The atmospheric scattering model layer obtains the defogging artillery barrel target image based on the input value.

[0051] I. Model Construction

[0052] In this embodiment, as Figure 2 As shown, a deep learning network model with a U-Net and AOD-Net cascaded structure is constructed as a dehazing enhancement network model, which includes an improved U-Net module, an AOD-Net module, and an atmospheric scattering model layer connected in sequence.

[0053] The improved U-Net module consists of an encoder and a decoder connected sequentially, with a skip connection module between them. The original U-Net module had four sampling layers for both the encoder and decoder, while the improved U-Net module reduces both to two sampling layers. Specifically, the encoder design includes a first downsampling layer and a second downsampling layer, and the decoder design includes a first upsampling layer and a second upsampling layer, making the entire model more lightweight. The encoder consists of a first convolutional layer, a second convolutional layer, a first downsampling layer, a third convolutional layer, a fourth convolutional layer, a second downsampling layer, and a fifth convolutional layer connected sequentially, while the decoder consists of a sixth convolutional layer, a first upsampling layer, a seventh convolutional layer, an eighth convolutional layer, a second upsampling layer, a ninth convolutional layer, and a tenth convolutional layer connected sequentially.

[0054] The encoder is responsible for progressively downsampling the input image (i.e., the image of the target gun barrel with fog), extracting features, and reducing the resolution. Specifically, the first, second, third, fourth, fifth, and sixth convolutional layers each perform three layers of feature extraction operations, with each layer using two 3×3 convolutional kernels to extract multi-scale features from the input image. Each convolutional layer is followed by a ReLU activation layer, which helps introduce non-linearity and assists the network in learning more complex features. The first and second downsampling layers use 2x2 max pooling layers. Max pooling downsamples the feature map of each layer, preserving key information at different resolutions to reduce spatial resolution and extract multi-scale features.

[0055] The decoder is responsible for progressively upsampling to restore the spatial resolution of the feature maps, generating segmentation maps, which are images that skip connections to the encoder output. Specifically, this includes a first upsampling layer and a second upsampling layer. The upsampling layer restores the image resolution through bilinear interpolation and concatenates it with the feature maps from the downsampling path. Finally, the output feature maps of the U-Net module, i.e., multi-scale feature maps, are generated through the ninth and tenth convolutional layers. The seventh and eighth convolutional layers are applied to the upsampled feature maps for further feature extraction. Each convolutional layer is followed by a ReLU activation layer, which helps introduce non-linearity and assists the network in learning more complex features.

[0056] The skip connection module passes high-resolution feature maps from the encoder to the decoder. During decoding, the feature maps from the encoder are directly concatenated with the feature maps of the corresponding layers in the decoder, thus preserving more high-resolution detail information. Through skip connections, the decoder can utilize the high-resolution features of the encoder, enhancing its ability to recover details and boundary information.

[0057] The AOD-Net module consists of the following layers connected in sequence: the eleventh convolutional layer, the twelfth convolutional layer, the first feature connection layer Concat1, the thirteenth convolutional layer, the second feature connection layer Concat2, the fourteenth convolutional layer, the third feature connection layer Concat3, and the fifteenth convolutional layer.

[0058] The AOD-Net module receives multi-scale feature maps from U-Net and first processes them further through convolutional layers. Different sized convolutional kernels are used to extract features from different receptive fields. These features are then concatenated to achieve multi-scale feature fusion, and finally, the input value of the atmospheric scattering model layer, i.e., the K value, is output.

[0059] The atmospheric scattering model layer is implemented using an atmospheric scattering model. Inputting the K value into the atmospheric scattering model formula yields a fog-free image of the artillery barrel target. The atmospheric scattering model is an existing model, and its specific structure will not be elaborated upon here.

[0060] II. Model Training

[0061] The detailed training process of the deep learning network model with the U-Net and AOD-Net cascade structure, i.e. the dehazing enhancement network model, is as follows:

[0062] Step a: Pass the input image x of the fogged artillery barrel target through the first convolutional layer of U-Net. The first convolutional layer uses a kernel size of 3x3, a stride of 1, and a channel size of 64: O1=F1(x);

[0063] Step b: Input the output feature O1 into the second convolutional layer of U-Net. The second convolutional layer has a kernel size of 3x3, a stride of 1, and a channel size of 64: O2 = F2(O1);

[0064] Step c: Input the output feature O2 into the first downsampling layer for downsampling operation in order to extract deeper features: O3 = Downsample1(O2);

[0065] Step d: Input the output feature O3 into the third convolutional layer of U-Net. The kernel size of the third convolutional layer is 3x3, the stride is 1, and the channel size is 128: O4 = F3(O3);

[0066] Step e: Input the output feature O4 into the fourth convolutional layer of U-Net. The kernel size of the fourth convolutional layer is 3x3, the stride is 1, and the channel size is 128: O5 = F4(O4);

[0067] Step f: Input the output feature O5 into the second downsampling layer and continue the downsampling operation: O6 = Downsample2(O5);

[0068] Step g: Input the output feature O6 into the fifth convolutional layer of U-Net. The kernel size of the fifth convolutional layer is 3x3, the stride is 1, and the channel size is 256: O7 = F5(O6);

[0069] Step h: Input the output feature O7 into the sixth convolutional layer of U-Net. The kernel size of the sixth convolutional layer is 3x3, the stride is 1, and the channel size is 256: O8 = F6(O7);

[0070] Step i: Input the output feature O8 into the first upsampling layer to restore the resolution of the image through upsampling: O9 = Upsample1(O8);

[0071] Step j: Concatenate the output features O9 and O5 channels (skip connection) and input them into the seventh convolutional layer for processing. The kernel size of the seventh convolutional layer is 3x3, the stride is 1, and the channel size is 128:O 10 =F7(Concat(O 9， O 5， ));

[0072] Step k: Output feature O 10 Input the eighth convolutional layer of U-Net. The kernel size of the eighth convolutional layer is 3x3, the stride is 1, and the channel size is 128:O 11 =F8(O 10 );

[0073] Step 1: Output feature O11 It is concatenated with the O2 channel (skip connection) and then fed into the ninth convolutional layer for processing. The kernel size of the ninth convolutional layer is 3x3, the stride is 1, and the channel size is 64:O 12 =F9(Concat(O 11， O 2， ));

[0074] Step m: Output feature O 12 Inputting the tenth convolutional layer of U-Net generates the final output feature map of U-Net: U out =F 10 (O 12 );

[0075] Step n: Convert the output feature map U of U-Net into a single image. out The input is fed into the eleventh convolutional layer of the AOD-Net module. The eleventh convolutional layer uses a kernel size of 1x1, a stride of 1, and a channel size of 3 to extract feature information from different receptive fields: O 13 =F 11 (U out );

[0076] Step o, output feature O 13 Inputting the twelfth convolutional layer of AOD-Net with a kernel size of 3×3, a stride of 1, and a channel size of 3, further extracts mesoscale feature information: O 14 =F 12 (O 13 );

[0077] Step p: Output feature O 14 With O 13 Channel concatenation is performed to form Concat1, which is then input into the thirteenth convolutional layer. The kernel size of the thirteenth convolutional layer is 5×5, the stride is 1, and the channel size is 3:0. 15 =F 13 (Concat(O 14， O 13， ));

[0078] Step q: Output feature O 15 With O 14 Channel concatenation is performed to form Concat2, which is then input into the fourteenth convolutional layer. The kernel size of the fourteenth convolutional layer is 7×7, the stride is 1, and the channel size is 3, further fusing multi-scale feature information: O 16 =F 14 (Concat(O 15， O 14， ));

[0079] Step r: Output feature O16 The outputs of the eleventh, twelfth, and thirteenth convolutional layers are concatenated to form Concat3, which is then input into the fifteenth convolutional layer to generate the final output K value: K = F 15 (Concat(O 16， O 14， O 13， O 12 ));

[0080] Step s: Input the output K value into the atmospheric scattering model to obtain the dehazed image. The atmospheric scattering model formula is:

[0081] I(x)=J(x)·t(x)+A·(1-t(x))

[0082] J(x)=K(x)I(x)-K(x)+b

[0083]

[0084] Where b is a constant deviation with a default value of 1; I(x) represents the pixel value of the foggy image captured by the imaging sensor at the pixel; J(x) represents the pixel value of the real scene image without fog, i.e. the clear image that we want to recover; A represents atmospheric light, i.e. the intensity of ambient light reflected by scattering particles in the air in foggy conditions, which is a constant independent of the scene; t(x) represents transmittance, i.e. the proportion of scene light transmitted through the atmosphere, reflecting the proportion of scene radiated light remaining when it reaches the camera. Transmittance is related to the distance between the object and the camera and the fog concentration.

[0085] The atmospheric scattering model provides an intuitive understanding of the degradation effect of fog on images: objects closer to the camera are less affected by fog, with a transmittance close to 1, resulting in clearer details in the observed image; while objects farther away are more affected by fog, with a significantly reduced transmittance, causing distant objects in the image to become blurry or even difficult to identify.

[0086] III. Optimization Model

[0087] To enable the above model to fit the objective function (loss function) and converge to the optimal solution more quickly and effectively during training, this invention proposes an optimizer that determines the step size based on the real-time gradient descent curve, thereby significantly shortening the training time. The formula is as follows:

[0088]

[0089] Where l is the slope of the gradient descent curve, and Ir is the step size. The optimizer works as follows: Each step returns a gradient descent slope value. A threshold t is set based on the model (here, the model threshold parameter is t = 0.15). When the slope l is greater than or equal to the threshold t, the step size is p1; when the slope l is less than the threshold t, the step size is p2, and the principle of p1 ≤ p2 must always be followed (here, the model parameters are p1 = 0.002, p2 = 0.005). This method of determining the step size based on the real-time curve can significantly reduce training time.

[0090] After training the aforementioned deep learning network model, a loss function is used to constrain the output results. This embodiment addresses the problem that traditional loss functions cannot effectively handle complex regions by designing an adaptive weight loss function. This loss function dynamically adjusts the weights based on the importance of the artillery barrel image region and the fog concentration, particularly focusing on processing the edges and details of the target region, thereby improving the image quality after defogging.

[0091] The loss function is expressed as follows:

[0092]

[0093] Here, w(x) is an adaptive weight that is adjusted based on the importance of the region and the concentration of haze. For example, for edge regions of the target image, the value of w(x) is larger, indicating that the region has a higher weight in the loss calculation; while for background regions, the value of w(x) is smaller.

[0094] A defogging enhancement network model based on the cascaded structure of AOD-Net and U-Net was used to process foggy artillery barrel target images to obtain defogging images. Figure 3 A comparison of the dehazing effects of several commonly used algorithms with the algorithm of this invention is presented. As can be clearly seen from the figures, the dehazing method using the concatenated AOD-Net and U-Net structures provided by this invention can clearly recover the edge details of the artillery barrel target, while effectively removing the haze in the background, resulting in a significant improvement in the overall image clarity and contrast.

[0095] Example 2

[0096] This embodiment discloses a deep learning-based dehazing system for artillery barrel target images, including:

[0097] The image acquisition module is configured to acquire raw images of the artillery barrel target.

[0098] The image dehazing module is configured to input the original artillery barrel target image into a pre-trained dehazing enhancement network model for processing, and obtain a dehazed artillery barrel target image.

[0099] The defogging enhancement network model includes an improved U-Net module, an AOD-Net module, and an atmospheric scattering model layer connected in sequence. The improved U-Net module extracts multi-scale features from the original artillery barrel target image and outputs a multi-scale feature map to the AOD-Net module. The AOD-Net module extracts features from different receptive fields of the multi-scale feature map through different convolutional layers to achieve multi-scale feature fusion and outputs the input value of the atmospheric scattering model layer. The atmospheric scattering model layer obtains the defogging artillery barrel target image based on the input value.

[0100] Example 3

[0101] The purpose of this embodiment is to provide a computing device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method of Embodiment 1.

[0102] Example 3

[0103] The purpose of this embodiment is to provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the steps of the method of Embodiment 1.

[0104] The steps and methods involved in the apparatuses of Embodiments 3 and 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0105] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0106] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0107] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for dehazing artillery barrel target images based on deep learning, characterized in that, include: Acquire the original image of the artillery barrel target; The original gun barrel target image is input into a pre-trained defogging enhancement network model for processing to obtain a defogging gun barrel target image. The defogging enhancement network model includes an improved U-Net module, an AOD-Net module, and an atmospheric scattering model layer connected in sequence. The improved U-Net module extracts multi-scale features from the original artillery barrel target image and outputs the multi-scale feature map to the AOD-Net module. The AOD-Net module extracts features from different receptive fields of the multi-scale feature map through different convolutional layers to achieve multi-scale feature fusion and outputs the input value of the atmospheric scattering model layer. The atmospheric scattering model layer obtains the defogging artillery barrel target image based on the input value. An optimizer is designed to shorten the training time during the training of the defogging enhancement network model; The optimizer is represented as: in, Let be the slope of the gradient descent curve. Step size, For threshold; The defogging enhancement network model is designed with an adaptive weight loss function to constrain the output. This loss function dynamically adjusts the weights based on the importance of the artillery barrel image region and the fog concentration.

2. The method for dehazing artillery barrel target images based on deep learning as described in claim 1, characterized in that, The improved U-Net module features a lightweight design for the U-Net network structure, reducing its sampling process.

3. The method for dehazing artillery barrel target images based on deep learning as described in claim 1, characterized in that, The atmospheric scattering model layer is implemented by an atmospheric scattering model, which is represented as follows: in, The constant deviation is 1, which is the default value. This represents the pixel value of a foggy image captured by the imaging sensor at each pixel. Represents the pixel values ​​of a real-world scene image without fog; A represents atmospheric light. This indicates transmittance.

4. The method for dehazing artillery barrel target images based on deep learning as described in claim 1, characterized in that, The loss function is expressed as: in, It is an adaptive weight. Represents the pixel values ​​of a real-world scene image without fog.

5. A deep learning-based dehazing system for artillery barrel target images, employing the deep learning-based dehazing method for artillery barrel target images as described in any one of claims 1-4, characterized in that, include: The image acquisition module is configured to acquire raw images of the artillery barrel target. The image dehazing module is configured to: input the original artillery barrel target image into a pre-trained dehazing enhancement network model for processing, and obtain a dehazed artillery barrel target image; The defogging enhancement network model includes an improved U-Net module, an AOD-Net module, and an atmospheric scattering model layer connected in sequence. The improved U-Net module extracts multi-scale features from the original artillery barrel target image and outputs a multi-scale feature map to the AOD-Net module. The AOD-Net module extracts features from different receptive fields of the multi-scale feature map through different convolutional layers to achieve multi-scale feature fusion and outputs the input value of the atmospheric scattering model layer. The atmospheric scattering model layer obtains the defogging artillery barrel target image based on the input value.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the deep learning-based dehazing method for artillery barrel target images as described in any one of claims 1-4.

7. A computer 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 program, it implements the steps in the deep learning-based dehazing method for artillery barrel target images as described in any one of claims 1-4.

Images