Target detection method and device in severe weather based on feature fusion

By introducing inclement weather processing enhanced branching and mixed random pooling in the target detection model, the structural similarity loss is calculated, and the parameters are optimized to improve the accuracy of inclement weather detection, solving the problems of high false detection rate and missed detection rate, achieving high efficiency real-time detection.

CN119027767AActive Publication Date: 2024-11-26ZHEJIANG UNIV
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411022141.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-29
Publication Date
2024-11-26
Estimated Expiration
2044-07-29

AI Technical Summary

Technical Problem

The prior art has high target detection error and missed detection rates in severe weather scenarios, and the calculation consumption brought by the cascade network is increased, which violates the real-time requirements.

Method used

Using a bad weather object detection method based on feature fusion, by introducing inclement weather processing enhancement branches and mixed random pooling in the target detection model, structural similarity loss is calculated to optimize model parameters and improve detection accuracy without increasing calculation consumption.

Benefits of technology

It improves the target detection accuracy in inclement weather scenarios, reduces training difficulty, enhances the generalization of the model, and avoids additional computing consumption during the inference stage.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119027767A_ABST
    Figure CN119027767A_ABST
Patent Text Reader

Abstract

The invention discloses a severe weather target detection method and device based on feature fusion, and the method comprises the steps: taking a target detection model as a detection branch, extracting a shallow feature map of a real severe weather image, and constructing the detection loss of the detection branch; adding a severe weather processing model as a severe weather processing enhancement branch, and removing severe weather interference factors from a real severe weather image to obtain a processed image; performing mixed random pooling on the shallow feature map and the processed image to obtain pooled features, and performing feature fusion on the two pooled features to obtain a fused feature; calculating the similarity loss of the fused features and pooled features corresponding to the shallow feature map, and performing parameter optimization on the target detection model according to the similarity loss and the detection loss; and target detection is performed on the to-be-detected severe weather image by using the target detection model after parameter optimization, so that the precision of target detection in the severe weather scene can be improved with relatively low calculation consumption.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of combining target detection and computer vision, and specifically relates to a method and device for detecting targets in severe weather based on feature fusion. Background Art

[0002] In the field of deep learning, object detection is a hot topic in the field, and object detection in extreme weather conditions such as fog, dust, and low light has always been difficult. In foggy scenes, the commonly used method is to cascade the defogging network and the detection network, restore the clean image through the defogging network, and use the clean image as the input of the detection network, such as the following two public technical solutions:

[0003] The patent application with publication number CN118154886A discloses a method for infrared image denoising and small target detection in bad weather. First, a three-category infrared image small target data set is constructed, the data set is randomly divided into three equal parts and three types of noise, namely Gaussian noise, Poisson noise and salt and pepper noise, are added; then a network architecture for infrared image denoising and small target detection in bad weather conditions is constructed, the overall framework includes a filter denoising module, a full-resolution denoising module and a YOLOv5 detector, and the loss supervision function is used in the detector to improve the detection performance of small targets; finally, based on the three-category infrared image small target data set with noise, the proposed network architecture is trained to obtain a model for denoising and recognition of infrared images, which can remove noise and extract the area where the key features are located in the image, and then the key area is passed to the detector to obtain the location information and category information of the target, which can better improve the performance of small target detection.

[0004] The patent application with publication number CN116704284A discloses an enhanced refinement network and method for target detection in bad weather, wherein the enhanced refinement network includes an adaptive image enhancement module, a feature refinement module and a target detection module connected in sequence; the adaptive image enhancement module includes a contrast-limited adaptive histogram equalization filter, an adaptive illumination enhancement filter and a Laplacian pyramid sharpening structure connected in sequence; the feature refinement module includes a downsampling layer, a feature refinement layer and an upsampling layer connected in sequence through a residual block. Target recognition is performed after feature enhancement and feature refinement.

[0005] However, cascade networks bring a lot of extra computing consumption, which greatly prolongs the inference time required by the model, which is contrary to the high real-time requirements of actual engineering applications. Therefore, it is necessary to propose a method that does not consume extra computing resources while improving the inference accuracy. Summary of the invention

[0006] In view of the above, the purpose of the present invention is to provide a method and device for severe weather target detection based on feature fusion, so as to solve the technical problems of high false detection rate and missed detection rate of target detection in severe weather scenes, without bringing about additional computing consumption, while improving the detection accuracy of the model in severe weather scenes.

[0007] To achieve the above-mentioned object of the invention, an embodiment provides a method for detecting a bad weather target based on feature fusion, comprising the following steps:

[0008] The target detection model is used as the detection branch. The shallow feature map of the real severe weather image is extracted using the target detection model network, and the detection loss of the detection branch is constructed.

[0009] A bad weather processing model with fixed parameters is added as a bad weather processing enhancement branch, and the bad weather processing model is used to remove the bad weather interference factors from the real bad weather image to obtain the processed image;

[0010] The shallow feature map and the processed image are subjected to mixed random pooling to obtain the pooled features, and the two pooled features are fused to obtain the fused features;

[0011] Calculate the structural similarity between the fusion feature and the pooled feature corresponding to the shallow feature map as the similarity loss, and optimize the parameters of the target detection model based on the similarity loss and detection loss;

[0012] The target detection model with optimized parameters is used to perform target detection on the severe weather images to be detected.

[0013] Preferably, the severe weather processing model is a model for removing at least one of the severe weather interference factors such as fog, rain, snow, low light, and strong light. For the interference factor of fog, the severe weather processing model adopts the DehazeFormer dehazing model, for the interference factor of rain, the DeTformer deraining model based on the Transformer framework is adopted, and for the interference factor of low light, the diffusion-based LightenDiffusion low-light restoration model is adopted.

[0014] Preferably, the shallow feature map and the processed image are subjected to mixed random pooling to obtain pooled features, including:

[0015] The processed image I en Downsample to the shallow feature map F low The same size is obtained to obtain image I′ en , image I′ en And the shallow feature map F low Randomly perform average pooling or maximum pooling along the channel to obtain the pooled feature I e ″n and F′ low :

[0016] I e ″ n =λAvgPool(I′ en )⊕(1-λ)MaxPool(I′ en )

[0017] F′ low =λAvgPool(F low )⊕(1-λ)MaxPool(F low )

[0018] Among them, λ is a random value of 0 or 1, AvgPool(·) means finding the average value along the channel dimension, ⊕ means adding the feature maps, and MaxPool(·) means finding the maximum value along the channel dimension.

[0019] Preferably, the two pooled features are fused to obtain fused features, including:

[0020] The pooled feature F′ corresponding to the shallow feature map low And the pooled feature I corresponding to the processed image e ″ n The fusion feature F is obtained by weighted addition. fuse :

[0021] F fuse =βI e ″ n ⊕(1-β)F′ low

[0022] Among them, β is the fusion weight coefficient, which is a random value, β∈[0,1].

[0023] Preferably, calculating the structural similarity between the fused feature and the pooled feature corresponding to the shallow feature map as the similarity loss includes:

[0024] Calculate the fusion feature F separately fuse The pooled feature F′ corresponding to the shallow feature map low The mean value of the pixels in and Standard Deviation and and covariance

[0025] Based on mean and Standard Deviation and and covariance Calculate the structural similarity SSIM (Ffuse ,F′ low ):

[0026]

[0027] Based on the structural similarity SSIM (F fuse ,F′ low ) Calculate the similarity loss L ssim :

[0028] L ssim =β*(1-SSIM(F fuse ,F ′ low ))

[0029] Among them, β is the pooled feature F′ corresponding to the shallow feature map low And the pooled feature I corresponding to the processed image e ″ n The fusion weight coefficients, C1 and C2 are constants to prevent the denominator from being zero.

[0030] Preferably, the detection loss of the detection branch includes a regression loss L representing the distance between the target prediction box and the target real box. CIoU , the target confidence loss L indicating the reliability of the target prediction box obj , the classification loss L, which indicates the reliability of predicting the correct category cls ;

[0031] The total loss L for parameter optimization of the target detection model is:

[0032] L=α box L CIoU +α obj L obj +α cls L cls +α ssim L ssim

[0033] Among them, α box , α obj , α cls , α ssim are the balance factors for balancing various losses, L ssim is the similarity loss.

[0034] Preferably, the target detection model after parameter optimization is used to perform target detection on the severe weather image to be detected, including:

[0035] The detected severe weather images are only input into the target detection model after parameter optimization, and the confidence and coordinates of the target box are obtained through forward reasoning to achieve target detection.

[0036] Preferably, the target detection model adopts a YOLOv5 model, and uses the backbone network of YOLOv5 to extract shallow feature maps.

[0037] To achieve the above-mentioned purpose of the invention, an embodiment also provides a severe weather target detection device based on feature fusion, including a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the above-mentioned severe weather target detection method based on feature fusion.

[0038] To achieve the above-mentioned purpose of the invention, an embodiment further provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the above-mentioned severe weather target detection method based on feature fusion.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The present invention adds a severe weather processing enhancement branch, and guides the detection branch to learn by introducing a new loss, namely, the structural similarity SSIM loss, thereby improving the target detection accuracy in severe weather scenarios; the severe weather processing enhancement branch of the present invention only works in the training stage and does not participate in the reasoning stage, so it does not bring additional computing consumption; when pooling images, the present invention designs a hybrid random pooling scheme, and replaces the conventional deterministic pooling with hybrid random pooling, which can solve the overfitting problem to a certain extent and improve the accuracy of the model; the present invention fuses images to different degrees and performs different degrees of loss penalties, one advantage of which is that the training difficulty is reduced, and another advantage is that it is equivalent to data enhancement at the feature level, generating different feature maps, and improving the generalization of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, 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 invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0042] Figure 1 is a flow chart of a method for detecting a target in severe weather based on feature fusion provided in an embodiment;

[0043] Figure 2 Schematic diagram of the severe weather processing enhancement branch provided by the embodiment. DETAILED DESCRIPTION

[0044] To make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation methods described herein are only used to explain the present invention and do not limit the scope of protection of the present invention.

[0045] like Figure 1 As shown, an embodiment provides a method for detecting a bad weather target based on feature fusion, comprising the following steps:

[0046] S1, takes the target detection model as the detection branch, uses the target detection model network to extract the shallow feature map of the real severe weather image, and constructs the detection loss of the detection branch.

[0047] In the embodiment, the target detection model of the detection branch can adopt YOLOv5, and the first layer is the backbone network of YOLOv5, specifically the Conv convolution layer, which includes three parts: a two-dimensional convolution layer with a convolution kernel size of 2 and a stride of 2, a BN layer, and a SiLU activation function layer. 3×H×W After this layer, the height and width become Get the shallow feature map F low , the shallow feature map F low The input to the second layer for target detection is and is used for subsequent calculations to obtain the detection loss, where the detection loss includes the regression loss L representing the distance between the target prediction box and the target real box. CIoU , the target confidence loss L indicating the reliability of the target prediction box obj , the classification loss L, which indicates the reliability of predicting the correct category cls , this detection loss is used for back-propagation.

[0048] S2, adding a bad weather processing model with fixed parameters as a bad weather processing enhancement branch, and using the bad weather processing model to remove bad weather interference factors from the real bad weather image to obtain a processed image.

[0049] In an embodiment, the added severe weather processing enhancement branch is used to guide the detection branch to learn the features of the image after removing the severe weather interference factors. Among them, the severe weather processing model is a model that removes at least one of the severe weather interference factors such as fog, rain, snow, low light, and strong light. For example, for the interference factor of fog, the severe weather processing model adopts the DehazeFormer dehazing model, for the interference factor of rain, the DeTformer deraining model based on the Transformer framework is adopted, and for the interference factor of low light, the diffusion-based LightenDiffusion low-light recovery model is adopted. These severe weather processing model parameters are fixed, and when applied, they are similar to black box input and output, and the model parameters are not updated. Such a design facilitates the replacement of various processing models, such as other dehazing, deraining, low-light enhancement models, etc., and has better versatility.

[0050] S3, the shallow feature map and the processed image are subjected to mixed random pooling to obtain pooled features, and the two pooled features are fused to obtain fused features.

[0051] In the embodiment, Figure 2 As shown, the pooling process includes: processing the image I en Downsample to the shallow feature map F low The same size is obtained to obtain image I′ en , image I′ en and shallow feature map F low Randomly perform average pooling or maximum pooling along the channel to obtain the pooled feature I e ″ n and F′ low :

[0052] I e ″ n =λAvgPool(I′ en )⊕(1-λ)MaxPool(I′ en )

[0053] F′ low =λAvgPool(F low )⊕(1-λ)MaxPool(F low )

[0054] Among them, λ is a random value of 0 or 1. During training, a random λ is generated for each batch of images. Assuming that each batch of images is n, the first batch of images will generate a λ1, and the second batch of images will generate a λ2. The λ in each batch is the same. AvgPool(·) means finding the average value along the channel dimension, ⊕ means adding the feature maps, and MaxPool(·) means finding the maximum value along the channel dimension.

[0055] In the embodiment, Figure 2 As shown, feature fusion includes: pooling the feature F′ corresponding to the shallow feature map low And the pooled feature I corresponding to the processed image e ″ n The fusion feature F is obtained by weighted addition. fuse :

[0056] F fuse =βI e ″ n ⊕(1-β)F′ low

[0057] Among them, β is the fusion weight coefficient, which is a random value, β∈[0,1]. During training, a random β is generated for each batch of images. Assuming that each batch of images is n, the first batch of images will generate a β1, and the second batch of images will generate β2. The β in each batch is the same.

[0058] S4, calculates the structural similarity between the fused features and the pooled features corresponding to the shallow feature map as the similarity loss, and optimizes the parameters of the target detection model based on the similarity loss and the detection loss.

[0059] In the embodiment, calculating the similarity loss includes:

[0060] S4-1, traverse image F fuse and F′ low All pixels, get the values ​​of all pixels, after mixed random pooling, F fuse and F′ low The number of channels is 1, so we only need to traverse the pixel values ​​of this channel. For convenience of representation, F fuse and F′ low Denote by x and y respectively, calculate the mean μ of image x and y x and μ y :

[0061]

[0062] Where N is the number of image pixels, x i Figure F fuse The value of the i-th pixel in , y i For graph F′ low The value of the i-th pixel in ;

[0063] S4-2, mean μ x and μ y Calculate image F fuse and F′ low The standard deviation σ x and σ y :

[0064]

[0065] S4-3, using the mean μ x and μ y Calculate image F fuse and F′ low The covariance σ xy :

[0066]

[0067] S4-4, using μ x , μ y , σ x , σ y and σ xy Calculate the structural similarity SSIM:

[0068]

[0069] Among them, C1 and C2 are constants to prevent the denominator from being zero;

[0070] S4-5, replace x and y with F fuse and F′ low The resulting SSIM loss is defined as:

[0071] L ssim =β*(1-SSIM(F fuse ,F′ low ))

[0072] Where β is the en With Figure F low The weight coefficient of fusion.

[0073] The target detection model is optimized based on the similarity loss and detection loss. The total loss L is used and the gradient is calculated using the back propagation algorithm. Then the SGD optimizer is used to update the parameters to minimize the loss function. The gradient back propagation is as follows: Figure 1 The dotted line indicates the total loss L:

[0074] L=α box L CIoU +α obj L obj +α cls L cls +α ssim L ssim

[0075] Among them, α box , α obj , α cls , α ssimare the balancing factors for balancing various losses respectively.

[0076] S5, using the target detection model with optimized parameters to perform target detection on the severe weather image to be detected.

[0077] In the embodiment, when applying detection, the severe weather image to be identified is input into the target detection model after parameter optimization. During reasoning, there is no need to extract shallow feature maps, nor is there any need to input the image into the severe weather processing enhancement branch. Only the target detection model is input for forward reasoning to obtain the confidence and coordinates of the target box.

[0078] Based on the same inventive concept, an embodiment further provides a device for detecting a target in severe weather based on feature fusion, including a memory and one or more processors, wherein executable code is stored in the memory, and when the one or more processors execute the executable code, the device is used to implement the above-mentioned method for detecting a target in severe weather based on feature fusion, which specifically includes the following steps:

[0079] S1, taking the target detection model as the detection branch, using the target detection model network to extract the shallow feature map of the real severe weather image, and constructing the detection loss of the detection branch;

[0080] S2, adding a bad weather processing model with fixed parameters as a bad weather processing enhancement branch, and using the bad weather processing model to remove bad weather interference factors from the real bad weather image to obtain a processed image;

[0081] S3, performing mixed random pooling on the shallow feature map and the processed image to obtain pooled features, and performing feature fusion on the two pooled features to obtain fused features;

[0082] S4, calculating the structural similarity between the fusion feature and the pooled feature corresponding to the shallow feature map as the similarity loss, and optimizing the parameters of the target detection model based on the similarity loss and the detection loss;

[0083] S5, using the target detection model with optimized parameters to perform target detection on the severe weather image to be detected.

[0084] The computing device provided in the embodiment, in addition to the processor and memory, also includes hardware required for other services such as internal bus, network interface, memory, etc. at the hardware level. The memory is a non-volatile memory, and the processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the severe weather target detection method based on feature fusion described in S1-S5 above. Of course, in addition to the software implementation, the present invention does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0085] Based on the same inventive concept, an embodiment further provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the above-mentioned method for detecting a target in severe weather based on feature fusion is implemented, which specifically includes the following steps:

[0086] S1, taking the target detection model as the detection branch, using the target detection model network to extract the shallow feature map of the real severe weather image, and constructing the detection loss of the detection branch;

[0087] S2, adding a bad weather processing model with fixed parameters as a bad weather processing enhancement branch, and using the bad weather processing model to remove bad weather interference factors from the real bad weather image to obtain a processed image;

[0088] S3, performing mixed random pooling on the shallow feature map and the processed image to obtain pooled features, and performing feature fusion on the two pooled features to obtain fused features;

[0089] S4, calculating the structural similarity between the fusion feature and the pooled feature corresponding to the shallow feature map as the similarity loss, and optimizing the parameters of the target detection model based on the similarity loss and the detection loss;

[0090] S5, using the target detection model with optimized parameters to perform target detection on the severe weather image to be detected.

[0091] In the embodiment, computer-readable media include permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. The information can be computer-readable instructions, data structures, program modules or other data.

[0092] The specific implementation methods described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for detecting targets in severe weather conditions based on feature fusion. It is characterized in that The following steps are involved: The target detection model is used as the detection branch. The shallow feature map of the real severe weather image is extracted using the target detection model network, and the detection loss of the detection branch is constructed. A bad weather processing model with fixed parameters is added as a bad weather processing enhancement branch, and the bad weather processing model is used to remove the bad weather interference factors from the real bad weather image to obtain the processed image; The shallow feature map and the processed image are subjected to mixed random pooling to obtain the pooled features, and the two pooled features are fused to obtain the fused features; Calculate the structural similarity between the fusion feature and the pooled feature corresponding to the shallow feature map as the similarity loss, and optimize the parameters of the target detection model based on the similarity loss and detection loss; The target detection model with optimized parameters is used to perform target detection on the severe weather images to be detected.

2. The method for detecting bad weather targets based on feature fusion according to claim 1, It is characterized in that The severe weather processing model is a model for removing at least one of the severe weather interference factors such as fog, rain, snow, low light, and strong light, including: for the interference factor of fog, the severe weather processing model adopts the DehazeFormer dehazing model, for the interference factor of rain, the DeTformer deraining model based on the Transformer framework is adopted, and for the interference factor of low light, the diffusion-based LightenDiffusion low-light restoration model is adopted.

3. The method for detecting bad weather targets based on feature fusion according to claim 1, It is characterized in that The shallow feature map and the processed image are subjected to mixed random pooling to obtain the pooled features, including: The processed image I en Downsample to the shallow feature map F low The same size is obtained as image I e ' n , image I e ' n and shallow feature map F low Randomly perform average pooling or maximum pooling along the channel to obtain the pooled feature I e ″ n and F l ' ow : I e ″ n =λAvgPool(I e ′ n )⊕(1-λ)MaxPool(I e ′ n ) F l ′ ow =λAvgPool(F low )⊕(1-λ)MaxPool(F low ) Among them, λ is a random value of 0 or 1, AvgPool(·) means finding the average value along the channel dimension, ⊕ means adding the feature maps, and MaxPool(·) means finding the maximum value along the channel dimension.

4. The method for detecting bad weather targets based on feature fusion according to claim 1, It is characterized in that The two pooled features are fused to obtain fused features, including: The pooled feature F′ corresponding to the shallow feature map low And the pooled feature I″ corresponding to the processed image en The fusion feature F is obtained by weighted addition. fuse : F fuse =βI″ en ⊕(1-β)F′ low Among them, β is the fusion weight coefficient, which is a random value, β∈[0,1].

5. The method for detecting bad weather targets based on feature fusion according to claim 1, It is characterized in that Calculate the structural similarity between the fused features and the pooled features corresponding to the shallow feature map as the similarity loss, including: Calculate the fusion feature F separately fuse The pooled feature F′ corresponding to the shallow feature map low The mean value of the pixels in and Standard Deviation and and covariance Based on mean and Standard Deviation and and covariance Calculate the structural similarity SSIM (F fuse ,F′ low ): Based on the structural similarity SSIM (F fuse ,F′ low ) Calculate the similarity loss L ssim : L ssin =β*(1-SSIM(F fuse ,F′ low )) Among them, β is the pooled feature F′ corresponding to the shallow feature map low And the pooled feature I″ corresponding to the processed image en The fusion weight coefficient, C 1 and C 2 is a constant to prevent the denominator from being zero.

6. The method for detecting bad weather targets based on feature fusion according to claim 1, It is characterized in that The detection loss of the detection branch includes a regression loss L representing the distance between the target prediction box and the target real box CIoU , the target confidence loss L indicating the reliability of the target prediction box obj , the classification loss L, which indicates the reliability of predicting the correct category cls ; The total loss L for parameter optimization of the target detection model is: L=a box L CIoU +a obj L obj +a cls L cls +a ssim L ssim Among them, α box , α obj , α cls , α ssim are the balance factors for balancing various losses, L ssim is the similarity loss.

7. The method for detecting bad weather targets based on feature fusion according to claim 1, It is characterized in that The target detection model with optimized parameters is used to detect the bad weather images to be detected, including: The detected severe weather images are only input into the target detection model after parameter optimization, and the confidence and coordinates of the target box are obtained through forward reasoning to achieve target detection.

8. The method for detecting bad weather targets based on feature fusion according to claim 1, It is characterized in that The target detection model adopts the YOLOv5 model, and uses the backbone network of YOLOv5 to extract shallow feature maps.

9. A device for detecting a target in severe weather based on feature fusion, comprising a memory and one or more processors, wherein the memory stores executable code, It is characterized in that When the one or more processors execute the executable code, they are used to implement the severe weather target detection method based on feature fusion as described in any one of claims 1-8.

10. A computer-readable storage medium, It is characterized in that A program is stored thereon, and when the program is executed by a processor, the method for detecting a target in severe weather based on feature fusion described in any one of claims 1 to 8 is implemented.

Citation Information

Patent Citations

  • Target detection and image definition combined learning method based on multi-scale feature fusion

    CN113537397A

  • Enhanced and refined network and method for target detection in severe weather

    CN116704284A

  • Atmospheric scattering model-based foggy image target detection method and system

    CN118279190A