A small target detection method in haze scenes based on deep learning

Abstract

The present invention relates to the field of target detection technology, specifically a method for detecting small targets in haze scenes based on deep learning. The method comprises the following specific steps: constructing an original dataset for detecting small targets in haze scenes, or using an existing publicly available dataset for detecting small targets in haze scenes; first collecting image data of small targets in a haze environment, then using a labeling tool to select the position of the small targets and label them; preprocessing the original dataset for detecting small targets in haze scenes to obtain a preprocessed dataset for detecting small targets in haze scenes, and dividing the dataset to facilitate subsequent training of a deep learning-based model for detecting small targets in haze scenes. When processing small target detection tasks in haze scenes, the present invention can more accurately capture and identify the detailed features of small targets, effectively reducing the interference of haze on the features of small targets.

Description

Technical Field

[0001] The present invention relates to the field of target detection technology, and specifically to a small target detection method in haze scenes based on deep learning. Background Art

[0002] Haze, a serious environmental pollution problem, typically occurs in cities or industrialized areas with poor air quality and low visibility. Its high pollution concentrations and long-term impacts not only endanger human health but also pose significant challenges to traffic safety and environmental monitoring. In particular, in areas such as traffic monitoring, public safety, and environmental management, detecting small targets (such as pedestrians, vehicles, and critical equipment) in haze-filled scenes is crucial for ensuring safety and improving management efficiency. However, due to the obstructive effects of haze, traditional target detection methods exhibit significant deficiencies in detecting small targets in images under low visibility conditions. New technical approaches are urgently needed to improve the accuracy and robustness of small target detection in haze-filled scenes.

[0003] Object detection in hazy scenes faces multiple challenges. First, haze reduces contrast, blurs edges, and loses detail in target images, making it difficult for traditional feature engineering methods (such as edge detection and color segmentation) to effectively extract target information. Second, small objects in hazy environments are more easily overlooked by traditional detection methods due to their smaller size and greater integration with background information, significantly increasing the missed detection rate. Furthermore, complex ambient lighting variations and optical scattering in hazy scenes further complicate object detection.

[0004] Traditional small target detection methods, such as image processing algorithms based on hand-crafted features, often require extensive manual debugging and struggle to achieve efficient and reliable detection results in complex haze scenes. Furthermore, these methods often fail to address the detection requirements of multi-scale targets and lack adaptability to the diverse morphologies of targets and complex backgrounds. In recent years, with the rapid development of deep learning technology, particularly the widespread application of models such as convolutional neural networks (CNNs) in the field of target detection, small target detection methods based on deep learning have demonstrated significant performance advantages. By automatically extracting multi-level features from data, deep learning can better capture the texture information and spatial distribution characteristics of targets, providing a new solution for small target detection in haze scenes.

[0005] Despite significant progress in object detection using deep learning techniques, small object detection in hazy scenes still faces numerous technical bottlenecks. For example, due to the degraded image quality caused by haze, existing deep learning models may have difficulty distinguishing objects from backgrounds during feature extraction, leading to false and missed detections. Furthermore, small objects often occupy a relatively small proportion of an image, and existing methods often lose critical information due to downsampling, compromising detection accuracy. Therefore, designing deep learning models that are more adaptable to low-quality images in hazy scenes, particularly enhancing their ability to extract small object features, remains a key research topic.

[0006] Based on the above, the existing deep learning-based small target detection model in haze scenes has the following problems:

[0007] 1. Haze interferes with images, leading to reduced detection performance: Existing object detection methods perform poorly in hazy scenes, primarily because haze reduces image contrast, blurs details, and loses target features. In these conditions, traditional models struggle to extract reliable target features, and are prone to missed or false detections, especially for small objects.

[0008] 2. The interweaving of small objects and complex backgrounds leads to insufficient feature representation: Small objects are often intertwined with complex backgrounds, making it difficult for existing object detection models to distinguish between the two. Especially in complex scenes, the features of small objects are easily obscured by background noise, resulting in high false detection and missed detection rates. Existing methods have poor representation capabilities for small object features and are unable to fully extract their details and boundary information. The lack of effective feature enhancement mechanisms means that target features are easily overwhelmed by background information during the feature extraction process. Consequently, the accuracy and reliability of small object detection are insufficient, especially for tiny objects or complex scenes.

[0009] 3. Insufficient multi-scale feature integration leads to limited model recognition capabilities: Small target detection requires the integration of multi-level and multi-scale feature information, but existing models have deficiencies in feature integration and find it difficult to capture the correlation between features at different levels, especially for detecting tiny and slender targets.

[0010] 4. Traditional methods are unable to process multi-scale and multi-level features: Traditional target detection methods use fixed convolutional structures or simple feature fusion techniques, which are difficult to adapt to the characteristic changes of small targets at different scales, resulting in unsatisfactory detection results for multi-scale targets.

[0011] Therefore, a small target detection method in haze scenes based on deep learning is invented. Summary of the Invention

[0012] To solve the above technical problems, according to one aspect of the present invention, the present invention provides the following technical solutions:

[0013] A method for detecting small targets in haze scenes based on deep learning includes the following specific steps:

[0014] S1: Construct an original dataset for small target detection in haze scenes, or use an existing publicly available dataset for small target detection in haze scenes; first collect image data of small targets in haze environments, then use annotation tools to select the locations of small targets and label them;

[0015] S2: Preprocess the original dataset of small target detection in haze scenes to obtain a preprocessed dataset of small target detection in haze scenes, and divide the dataset to facilitate subsequent training of a deep learning-based small target detection model in haze scenes.

[0016] S3: Constructing a small target detection network FASOD-NET in haze scenes;

[0017] S4: Based on the constructed FASOD-Net model for small object detection in haze scenes, train and update the parameters of each layer. First, initialize all neural network parameters and set model-related hyperparameters, including training rounds, batch size, weight decay coefficient, learning rate, and total number of iterations.

[0018] S5: After the haze scene small target detection model is trained, the trained haze scene small target detection model is applied to the haze scene image to detect small targets. Finally, the small target detection results in the haze scene are output, including the small target bounding box coordinates, small target category label and confidence score.

[0019] As a preferred solution of the method for detecting small targets in haze scenes based on deep learning described in the present invention, the specific steps of S2 are as follows:

[0020] S21: Cropping the hazy scene images in the original dataset of small target detection in hazy scenes to a uniform size. Since the original dataset is small, to ensure that the dataset has sufficient samples for training, validation, and testing, and to enhance the robustness of the small target detection model in hazy scenes, the model is less sensitive to image quality.

[0021] S22: Data enhancement processing is then performed to better train the deep learning model, and finally a data-enhanced small target dataset in the haze scene is obtained, which is divided into a training set, a validation set, and a test set in a ratio of 8:1:1.

[0022] As a preferred solution of the method for detecting small targets in haze scenes based on deep learning described in the present invention, the specific steps of S3 are as follows:

[0023] S31: Input the haze scene image into the dehazing and shallow feature extraction module;

[0024] S32: Input F4 into the multi-channel convolution module for fine feature capture;

[0025] S33: Input F2, F3, and F5 into the multi-scale feature integration and enhancement detection module.

[0026] As a preferred solution of the method for detecting small targets in haze scenes based on deep learning described in the present invention, the construction and execution process of the defogging and shallow feature extraction module in S31 is as follows:

[0027] S311: Apply a CBS operation including a 3×3 convolution, normalization, and activation function to the input haze scene image to obtain a haze shallow feature map F1;

[0028] S312: Input F1 into the multi-scale fusion large convolution kernel enhancement module MSE-LKC;

[0029] S313: Input F2 into the parallel enhanced attention module PEA;

[0030] S314: Input F3 and F1 into Fusion for channel dimension fusion, and obtain the final output feature map F4 by fusing features at different levels.

[0031] As a preferred solution of the method for detecting small targets in haze scenes based on deep learning according to the present invention, the operation and construction process of the MSE-LKC module in S312 is as follows:

[0032] S3121: Perform batch normalization on the input F1 to obtain a feature map F1_1. Then, perform a convolution operation on F1_1 with a 7×7 convolution kernel to extract features to obtain a feature map F1_2. Then, perform a convolution operation on F1_2 with a 5×5 convolution kernel to obtain the first-stage feature map F1_3.

[0033] S3122: Input F1_3 into a deep dilated convolution module with a dilation rate of 7×7, a deep dilated convolution module with a dilation rate of 13×13, and a deep dilated convolution module with a dilation rate of 19×19 to extract multi-scale features, generating three corresponding feature sub-graphs F1_4, F1_5, and F1_6 respectively; then, concatenate F1_4, F1_5, and F1_6 in the channel dimension through the C operation to obtain the fused feature graph F1_7;

[0034] S3123: Input F1_7 into a 1×1 convolution kernel for convolution operation to obtain feature map F1_8, process F1_8 through the GELU activation function to obtain feature map F1_9, and perform Conv1×1 convolution operation on F1_9 to obtain the final output feature map F1_10;

[0035] S3124: Add F1 and F1_10 through residual connection to form a feature map F2 with stronger representation ability.

[0036] As a preferred solution of the method for detecting small targets in haze scenes based on deep learning described in the present invention, the construction and execution process of the PEA module in S313 is as follows:

[0037] S3131: Perform batch normalization on F2 to standardize the feature distribution, improve the stability and convergence speed of the model, and obtain the normalized feature map F2_1;

[0038] S3132: Input F2_1 into three attention modules respectively:

[0039] S31321: The pixel-level attention module processes the spatial dimension of F2_1, extracts local saliency information, and obtains the enhanced feature subgraph F2_2;

[0040] S31322: The channel-level attention module operates on the channel dimension of F2_1 to capture global context information and obtain the weighted feature subgraph F2_3;

[0041] S31323: The simple pixel-level attention module uses a simplified mechanism to process F2_1, supplementing the feature extraction capabilities of other modules to obtain the feature subgraph F2_4;

[0042] S3133: Concatenate F2_2, F2_3, and F2_4 in the channel dimension to obtain the fused feature map F2_5;

[0043] S3134: Perform a Conv1×1 convolution operation on F2_5 to achieve feature dimensionality reduction, and perform a nonlinear transformation on the feature map using the GELU activation function to obtain the processed feature map F2_6. Then perform a Conv1×1 convolution operation on F2_6 to further extract features and adapt them to subsequent tasks, obtaining the feature map F2_7.

[0044] S3135: Add F2 and F2_7 through residual connection to obtain the final output feature map F3;

[0045] Among them, the process of applying PEA to F2 to obtain F3 is shown in the following formula:

[0046] F2_1=BatchNorm(F2)

[0047] F3=F2+Conv1_1(GELU(Conv1_1(C(PA(F2_1),CA(F2_1),SPA(F2_1)))))

[0048] Among them, BatchNorm represents the batch normalization operation; Conv1_1 represents the convolution operation of size 1×1; GELU represents the application of the activation function GELU; SPA represents the simple pixel-level attention module; CA represents the channel-level attention module; PA represents the pixel-level attention module; C represents the splicing operation in the channel dimension; + represents the residual connection operation.

[0049] As a preferred solution of the method for detecting small targets in haze scenes based on deep learning described in the present invention, the construction and execution process of the pixel-level attention module in S31321 is as follows:

[0050] S313211: Perform a Conv1×1 convolution operation on F2_1 to adjust the number of channels of the input feature map; then, apply the activation function GELU to further process F2_1 to obtain the activated feature map F2_1_1;

[0051] S313212: Perform a Conv1×1 convolution operation on F2_1_1 to obtain the intermediate feature map F2_1_2; then, apply the Sigmoid activation function to normalize F2_1_2 to obtain the pixel-level attention weight;

[0052] S313213: Performs element-wise multiplication of F2_1 and pixel attention weights to weight the feature map and obtain the final feature map F2_2. This module can enhance the representation of small target features while suppressing irrelevant background information, improving the model's ability to detect small targets in haze scenes.

[0053] Among them, the process of applying the pixel-level attention module to process F2_1 and obtain F2_2 is shown in the following formula:

[0054] F2_2=F2_1×S(Conv1_1(GELU(Conv1_1(F2_1))))

[0055] Among them, Conv1_1 represents a convolution operation of size 1×1; GELU represents the application of the activation function GELU; S represents the Sigmoid activation function; × represents the element-by-element multiplication operation;

[0056] The construction and execution process of the channel-level attention module in S31322 is as follows:

[0057] S313221: Perform global average pooling on F2_1 to generate global information in the channel dimension; then, perform a Conv1×1 convolution operation on the pooled result to obtain the feature map T1, and perform nonlinear processing on T1 through the activation function GELU to generate the intermediate expression feature map T2;

[0058] S313222: Perform a Conv1×1 convolution operation on T2 and then use the Sigmoid activation function to generate channel attention weights. Perform element-wise multiplication of F2_1 and the channel attention weights to achieve weighted output on the channel and obtain the enhanced output feature map F2_3.

[0059] Among them, the process of applying the channel-level attention module to process F2_1 to obtain F2_3 is shown in the following formula:

[0060] F2_3=F2_1×S(Conv1_1(GELU(Conv1_1(GAP(F2_1)))))

[0061] Among them, GAP represents the global average pooling operation; Conv1_1 represents the convolution operation of size 1×1; GELU represents the application of the activation function GELU; S represents the Sigmoid activation function; × represents the element-by-element multiplication operation;

[0062] The construction and execution process of the simple pixel-level attention module in S31323 is as follows:

[0063] S313231: Perform a Conv1×1 convolution operation on F2_1 to obtain a preliminary pixel feature map X1; then, X1 is convolved with another 3×3 convolution kernel to further extract local pixel correlation information to obtain a feature map X2;

[0064] S313232: Pass F2_1 through another branch to directly perform a Conv1×1 convolution operation to obtain the feature map X3; then, add X2 and X3 and generate pixel attention weights through the Sigmoid activation function;

[0065] S313233: Perform element-wise multiplication of F2_1 and pixel attention weight to obtain the enhanced output feature map F2_4;

[0066] Among them, the process of applying a simple pixel-level attention module to process F2_1 to obtain F2_4 is shown in the following formula:

[0067] F2_4=F2_1×s(Conv3_3(Conv1_1(F2_1))+Conv1_1(F2_1))

[0068] Among them, Conv1_1 represents a convolution operation of size 1×1; Conv3_3 represents a convolution operation of size 3×3; S represents the Sigmoid activation function; + represents the addition operation of the results of the two branches; × represents the element-by-element multiplication operation.

[0069] As a preferred solution of the method for detecting small targets in haze scenes based on deep learning described in the present invention, the construction and execution process of the fine feature capture multi-channel convolution module in S32 is as follows:

[0070] S321: Perform a Conv1×1 convolution operation on F4 to adjust the number of channels and extract preliminary features to generate a feature map F4_1;

[0071] S322: Feature splitting is performed on F4_1, which is divided into three feature processing paths: in the first path, two Conv 3×3 convolution operations are performed to extract local deep features and generate feature map F4_2; in the second path, Conv3×3 convolution operations and Conv5×5 convolution operations are performed in sequence to further extract features with a larger receptive field and generate feature map F4_3; in the third path, Conv5×5 convolution operations are performed to further extract features with a larger receptive field and generate feature map F4_4;

[0072] S323: Concatenate F4_2, F4_3, and F4_4 output from the three paths in the channel dimension to obtain the fused feature map F4_5; perform channel integration on F4_5 through the Conv1×1 convolution operation to generate the final small target deep feature map F4_6;

[0073] S324: Perform a Conv3×3 convolution operation on F4_6 to further integrate cross-path information and optimize local feature expression to generate feature map F5. Through the above steps, its module combines multi-scale convolution kernels and diversion structures to effectively extract deep semantic features of small targets and enhance the model's detection ability for small targets.

[0074] As a preferred solution of the method for detecting small targets in haze scenes based on deep learning described in the present invention, the construction and execution process of the multi-scale feature integration enhancement detection module in S33 is as follows:

[0075] S331: Apply a deformable convolution operation to the input F2 to dynamically adapt to the morphological characteristics of small targets and complex background noise in the haze scene, generating an enhanced feature map F6; apply a deformable convolution operation to the input F3 to generate an enhanced feature map F7;

[0076] S332: Concatenate F5 and F7 in the channel dimension to form a fused feature map F8, which enhances the multi-scale expression capability of the features and takes into account the unification of details and global features. Subsequently, a Conv 3×3 convolution operation is performed on F8 to further integrate cross-path information and optimize the multi-level expression of features to generate an optimized feature map F9. Finally, concatenate F9 and F6 in the channel dimension to form a fused feature map F10.

[0077] S333: Perform a Conv3×3 operation on F10 to further integrate cross-path information and optimize the multi-level expression of features to obtain a feature map F11; input F11 into the target detection head Head, use Head to detect F11, and output a tensor containing detection information. Each row corresponds to a detection result, which is a small target detection result in the haze scene, including the bounding box coordinates, small target category label and confidence score.

[0078] Compared with existing technologies:

[0079] When processing small target detection tasks in haze scenes, the present invention can more accurately capture and identify the detailed features of small targets, effectively reduce the interference of haze on the features of small targets, and solve the problem of unclear expression of small target features caused by haze in the existing technology; the present invention can more accurately distinguish small targets from the background through multi-scale feature extraction and multi-level feature fusion, and solve the problem that small targets are easily confused with background information in complex haze scenes; in addition, the present invention enhances the detection capability of tiny and slender small targets, significantly reduces the problem of small target feature loss caused by excessive pooling and convolution operations in the existing technology, and significantly improves the accuracy and robustness of small target detection in haze scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 It is a schematic diagram of the process of the present invention;

[0081] Figure 2 Schematic diagram of the main structure of FASOD-Net of the present invention;

[0082] Figure 3 Schematic diagram of the defogging and shallow feature extraction module of the present invention;

[0083] Figure 4 This is a schematic diagram of the MSE-LKC structure of the present invention;

[0084] Figure 5 Schematic diagram of the PEA structure of the present invention;

[0085] Figure 6 Schematic diagram of the pixel-level attention module structure of the present invention;

[0086] Figure 7 Schematic diagram of the channel-level attention module structure of the present invention;

[0087] Figure 8 This is a schematic diagram of the structure of a simple pixel-level attention module of the present invention;

[0088] Figure 9 Schematic diagram of the multi-channel convolution module for capturing fine features of the present invention;

[0089] Figure 10 Schematic diagram of the multi-scale feature integration and enhanced detection module of the present invention. DETAILED DESCRIPTION

[0090] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0091] The present invention provides a small target detection method in haze scenes based on deep learning, which is used to detect and identify small targets in haze environments. The present invention designs a small target detection network in haze scenes called FASOD-Net (Fog-Aware Small Object Detection Network). The network can effectively remove the interference of haze on the image, deeply analyze the foggy image, capture and extract the features of small targets, and finally generate a target detection result image. In the target detection result image, the detected small targets are marked with bounding boxes, and the background area remains unchanged; please refer to the detailed description of the network. Figures 1-10 ;

[0092] The specific steps are as follows:

[0093] S1: Construct an original dataset for small target detection in haze scenes, or use an existing publicly available dataset for small target detection in haze scenes. First, collect image data of small targets in haze environments (e.g., using drones or cameras to take field photos at different times and locations, or using an existing publicly available dataset). Then, use annotation tools to select the location of the small targets (i.e., determine the horizontal and vertical coordinates of the upper left and lower right corners of the box) and label them.

[0094] S1 includes but is not limited to the following embodiments: the richer the types of small targets contained in the dataset, the more small targets the model can detect; based on the original dataset of small target detection in haze scenes collected by ourselves, a total of 2000 images, including 800 small targets of vehicles, 600 small targets of people, 350 small targets of animals, and 250 small targets of building details; the labels here can be divided into four categories: vehicles, people, animals, and building details; use the LabelImg tool to annotate the dataset and export the xml label file, which contains the label name and annotation box information of the target small targets;

[0095] Among them, the LabelImg tool is an open source image annotation tool, mainly used for image dataset annotation for tasks such as target detection and image segmentation;

[0096] S2: Preprocess the original dataset of small target detection in haze scenes to obtain a preprocessed dataset of small target detection in haze scenes, and divide the dataset to facilitate subsequent training of a deep learning-based small target detection model in haze scenes.

[0097] The specific steps of S2 are as follows:

[0098] S21: Cropping the hazy scene images in the original dataset of small target detection in hazy scenes to a uniform size. Since the original dataset is small, to ensure that the dataset has sufficient samples for training, validation, and testing, and to enhance the robustness of the small target detection model in hazy scenes, the model is less sensitive to image quality.

[0099] S22: Data augmentation is then performed to better train the deep learning model, ultimately obtaining a data-enhanced dataset of small objects in haze scenarios. This dataset is then divided into a training set, a validation set, and a test set in a ratio of 8:1:1.

[0100] Its S2 includes but is not limited to the following embodiments: the original data set for small target detection in haze scenes has a total of 2000 images, and the images are first cropped to a uniform size of 512×512 pixels; the original data set is expanded by a data enhancement method, specifically including adding random pixels, Gaussian noise, salt and pepper noise, image contrast adjustment, image flipping, etc., which can be implemented in Python language; random pixels use the cv2.randu() function, Gaussian noise uses the ImageFilter.GaussianBlur() method in the PIL library, salt and pepper noise can use the cv2.randSalt() function, image contrast adjustment can use the ImageEnhance.Contrast method, and flip operation is implemented by the cv2.flip() function; the final data set is enhanced and expanded to 12,000 images, and divided according to the ratio of 8:1:1 to obtain 9600 training sets, 1200 validation sets, and 1200 test sets;

[0101] S3: Constructing a small target detection network FASOD-NET in haze scenes;

[0102] Input the haze scene images from the small target dataset under haze scenes constructed in Step 2 into FASOD-Net;

[0103] This paper designs a small target detection network FASOD-Net (Fog-Aware Small Object Detection Network) in haze scenes. The process of using FASOD-Net to detect small targets in haze scene images is divided into three steps: defogging and small target shallow feature extraction; fine feature capture; multi-level feature fusion and small target detection. The above three main steps correspond to the three modules of the FASOD-Net network model: defogging and shallow feature extraction module, fine feature capture multi-channel convolution module, and multi-scale feature integration and enhanced detection module.

[0104] The specific steps for S3 are as follows:

[0105] S31: Input the haze scene image into the defogging and shallow feature extraction module. The defogging and shallow feature extraction module is as follows: Figure 3 As shown;

[0106] The construction and execution process of the dehazing and shallow feature extraction modules in S31 are as follows:

[0107] The haze scene image is input into the dehazing and shallow feature extraction module. A CBS operation is performed on the haze scene image to obtain the haze shallow feature map F1. Then, F1 is input into the MSE-LKC module to obtain the feature map F2. Next, F2 is input into the PEA module to obtain the feature map F3. Finally, F3 and F1 are input into the Fusion module to obtain the final output feature map F4.

[0108] S311: Apply a CBS operation, consisting of a 3×3 convolution, normalization, and activation function, to the input haze scene image to obtain a haze shallow feature map F1. The CBS operation is a basic convolution module that extracts local image features using a 3×3 convolution kernel. Batch normalization (BatchNorm) is then performed to stabilize the feature distribution. The activation function introduces nonlinearity to enhance the model's feature representation capabilities.

[0109] S312: Input F1 into the multi-scale fusion large convolution kernel enhancement module MSE-LKC. The structure of the MSE-LKC module is as follows: Figure 4 As shown;

[0110] The operation and construction process of the MSE-LKC module in S312 is as follows:

[0111] S3121: Perform batch normalization on input F1 (i.e. Figure 4 The BatchNorm operation in the image is used to obtain the feature map F1_1, and then a convolution operation is performed on F1_1 using a convolution kernel of size 7×7 to extract features (i.e. Figure 4 The feature map F1_2 is obtained by using the Conv7×7 in the image, and then the convolution kernel of size 5×5 is used to perform a convolution operation on F1_2 to obtain the feature map F1_3 of the first stage;

[0112] S3122: Input F1_3 into the depth expansion convolution module with an expansion rate of 7×7 (i.e. Figure 4 DWDConv7 in ), a deep dilated convolution module with a dilation rate of 13×13 (i.e. Figure 4 DWDConv13 in ), a deep dilated convolution module with a dilation rate of 19×19 (i.e. Figure 4 DWDConv19 in

[19] is used to extract multi-scale features and generate three corresponding feature sub-graphs F1_4, F1_5 and F1_6 respectively; then, F1_4, F1_5 and F1_6 are concatenated in the channel dimension through the C operation (i.e., the Concat operation) to obtain the fused feature graph F1_7;

[0113] S3123: Input F1_7 into a 1×1 convolution kernel for convolution operation (i.e., Conv1×1 operation in the figure) to obtain feature map F1_8. F1_8 is processed by GELU activation function to obtain feature map F1_9. F1_9 is subjected to Conv1×1 convolution operation to obtain the final output feature map F1_10.

[0114] S3124: Add F1 and F1_10 through residual connection (i.e., the + operation in the figure) to form a feature map F2 with stronger representation ability;

[0115] BatchNorm is a commonly used deep learning technique that stabilizes the activation distribution of a neural network by standardizing the features of each batch of data to a mean of 0 and a variance of 1. It dynamically adjusts the scale of network inputs during training, alleviating vanishing or exploding gradients, accelerating model convergence, and improving training stability and generalization. Furthermore, BatchNorm has a certain regularization effect, helping to reduce reliance on other regularization methods.

[0116] The GELU (Gaussian Error Linear Unit) activation function is a smooth nonlinear activation function that combines the advantages of ReLU and Sigmoid. Its mathematical expression is: GELU(x) = x·Φ(x), where Φ(x) is the cumulative distribution function of the standard normal distribution. Compared with ReLU and Sigmoid, GELU smoothly adjusts the activation according to the size of the input value, allowing larger values ​​to pass and suppressing smaller values, providing a more natural gradient flow. It is widely used in deep learning models (such as Transformer) to improve training effects and performance.

[0117] S313: Input F2 into the parallel enhanced attention module PEA. The structure of the PEA module is as follows: Figure 5 As shown;

[0118] The construction and execution process of the PEA module in S313 is as follows:

[0119] S3131: Batch normalization of F2 (i.e. Figure 5 Batch Norm operation in is used to standardize the feature distribution, improve the stability and convergence speed of the model, and obtain the normalized feature map F2_1;

[0120] S3132: Input F2_1 into three attention modules respectively:

[0121] S31321: The pixel-level attention module processes the spatial dimension of F2_1, extracts local saliency information, and obtains the enhanced feature subgraph F2_2;

[0122] The construction and execution process of the pixel-level attention module in S31321 is as follows:

[0123] Input F2_1 into the pixel-level attention module (Pixel Attention, PA), the structure of the Pixel Attention module is as follows Figure 6 As shown;

[0124] S313211: Perform a Conv1×1 convolution operation on F2_1 to adjust the number of channels of the input feature map; then, apply the activation function GELU to further process F2_1 to obtain the activated feature map F2_1_1;

[0125] S313212: Perform a Conv1×1 convolution operation on F2_1_1 to obtain the intermediate feature map F2_1_2; then, apply the Sigmoid activation function to normalize F2_1_2 to obtain the pixel-level attention weight;

[0126] S313213: Performs element-wise multiplication of F2_1 and pixel attention weights to weight the feature map and obtain the final feature map F2_2. This module can enhance the representation of small target features while suppressing irrelevant background information, improving the model's ability to detect small targets in haze scenes.

[0127] Among them, the process of applying the pixel-level attention module to process F2_1 and obtain F2_2 is shown in the following formula:

[0128] F2_2=F2_1×S(Conv1_1(GELU(Conv1_1(F2_1))))

[0129] Among them, Conv1_1 represents a convolution operation of size 1×1; GELU represents the application of the activation function GELU; S represents the Sigmoid activation function; × represents the element-by-element multiplication operation;

[0130] S31322: The channel-level attention module (Channel Attention) operates on the channel dimension of F2_1 to capture global context information and obtain the weighted feature subgraph F2_3;

[0131] The construction and execution process of the channel-level attention module in S31322 is as follows:

[0132] Input F2_1 into the channel-level attention module (Channel Attention, CA), the structure of the ChannelAttention module is as follows Figure 7 As shown;

[0133] S313221: Perform global average pooling (i.e., GAP operation in the figure) on F2_1 to generate global information in the channel dimension; then, perform Conv1×1 convolution operation on the pooled result to obtain feature map T1, and perform nonlinear processing on T1 through the activation function GELU to generate the intermediate expression feature map T2;

[0134] S313222: After performing a Conv1×1 convolution operation on T2, a Sigmoid activation function is used to generate channel attention weights. F2_1 is element-wise multiplied with the channel attention weights to achieve weighted output on the channel (i.e., the multiplication operation in the figure), and the enhanced output feature map F2_3 is obtained.

[0135] Among them, the process of applying the channel-level attention module to process F2_1 to obtain F2_3 is shown in the following formula:

[0136] F2_3=F2_1×s(Conv1_1(GELU(conv1_1(GAP(F2_1)))))

[0137] Among them, GAP represents the global average pooling operation; Conv1_1 represents the convolution operation of size 1×1; GELU represents the application of the activation function GELU; S represents the Sigmoid activation function; × represents the element-by-element multiplication operation;

[0138] S31323: Simple Pixel Attention module uses a simplified mechanism to process F2_1, supplementing the feature extraction capabilities of other modules to obtain the feature subgraph F2_4;

[0139] The construction and execution process of the simple pixel-level attention module in S31323 is as follows:

[0140] Input F2_1 into the Simple Pixel Attention (SPA) module. The structure of the SimplePixel Attention module is as follows: Figure 8 As shown;

[0141] S313231: Perform a Conv1×1 convolution operation on F2_1 to obtain a preliminary pixel feature map X1; then, X1 is convolved with another convolution kernel of size 3×3 (i.e., the Conv3×3 operation in the figure) to further extract local pixel correlation information and obtain the feature map X2;

[0142] S313232: Pass F2_1 through another branch to directly perform a Conv1×1 convolution operation to obtain the feature map X3; then, add X2 and X3 and generate pixel attention weights through the Sigmoid activation function;

[0143] S313233: Perform element-wise multiplication of F2_1 and pixel attention weight (i.e. Figure 8 The multiplication operation in is performed to obtain the enhanced output feature map F2_4;

[0144] Among them, the process of applying a simple pixel-level attention module to process F2_1 to obtain F2_4 is shown in the following formula:

[0145] F2_4=F2_1×s(Conv3_3(Conv1_1(F2_1))+Conv1_1(F2_1))

[0146] Among them, Conv1_1 represents a convolution operation of size 1×1; Conv3_3 represents a convolution operation of size 3×3; S represents the Sigmoid activation function; + represents the addition operation of the results of the two branches; × represents the element-by-element multiplication operation;

[0147] S3133: Concatenate F2_2, F2_3, and F2_4 in the channel dimension (i.e. Figure 5 C operation in the figure), and obtain the fused feature map F2_5;

[0148] S3134: Perform a Conv1×1 convolution operation on F2_5 to achieve feature dimensionality reduction, and perform a nonlinear transformation on the feature map using the GELU activation function to obtain the processed feature map F2_6. Then perform a Conv1×1 convolution operation on F2_6 to further extract features and adapt them to subsequent tasks, obtaining the feature map F2_7.

[0149] S3135: Through residual connection (i.e. Figure 5 The "+" operation in the function adds F2 and F2_7 to obtain the final output feature map F3;

[0150] Among them, the process of applying PEA to F2 to obtain F3 is shown in the following formula:

[0151] F2_1=BatchNorm(F2)

[0152] F3=F2+Conv1_1(GELU(Conv1_1(C(PA(F2_1),CA(F2_1),SPA(F2_1)))))

[0153] Among them, BatchNorm represents the batch normalization operation; Conv1_1 represents the convolution operation of size 1×1; GELU represents the application of the activation function GELU; SPA represents the simple pixel-level attention module; CA represents the channel-level attention module; PA represents the pixel-level attention module; C represents the splicing operation in the channel dimension; + represents the residual connection operation;

[0154] S314: Input F3 and F1 into Fusion for channel dimension fusion, and obtain the final output feature map F4 by fusing features at different levels;

[0155] S32: Input F4 into the fine feature capture multi-channel convolution module. The structure of the fine feature capture multi-channel convolution module is as follows: Figure 9 As shown;

[0156] The construction and execution process of the fine feature capture multi-channel convolution module in S32 is as follows:

[0157] S321: Perform a Conv1×1 convolution operation on F4 to adjust the number of channels and extract preliminary features to generate a feature map F4_1;

[0158] S322: Feature splitting is performed on F4_1, which is divided into three feature processing paths: in the first path, two Conv 3×3 convolution operations are performed to extract local deep features and generate feature map F4_2; in the second path, Conv3×3 convolution operations and Conv5×5 convolution operations are performed in sequence to further extract features with a larger receptive field and generate feature map F4_3; in the third path, Conv5×5 convolution operations are performed to further extract features with a larger receptive field and generate feature map F4_4;

[0159] S323: Concatenate F4_2, F4_3 and F4_4 output from the three paths in the channel dimension (i.e. Figure 9 C operation in the image), and obtain the fused feature map F4_5; perform channel integration on F4_5 through the Conv1×1 convolution operation to generate the final small target deep feature map F4_6;

[0160] S324: Perform a Conv3×3 convolution operation on F4_6 to further integrate cross-path information and optimize local feature expression to generate feature map F5. Through the above steps, the module combines multi-scale convolution kernels and shunt structure to effectively extract deep semantic features of small objects and enhance the model's detection ability for small objects.

[0161] S33: Input F2, F3, and F5 into the multi-scale feature integration enhancement detection module. The structure of the multi-scale feature integration enhancement detection module is as follows: Figure 10 As shown;

[0162] The construction and execution process of the multi-scale feature integration enhanced detection module in S33 is as follows:

[0163] S331: Apply a deformable convolution operation to the input F2 to dynamically adapt to the morphological characteristics of small targets and complex background noise in the haze scene, generating an enhanced feature map F6; apply a deformable convolution operation to the input F3 to generate an enhanced feature map F7;

[0164] S332: Concatenate F5 and F7 in the channel dimension to form a fused feature map F8, which enhances the multi-scale expression capability of the features and takes into account the unification of details and global features. Subsequently, a Conv 3×3 convolution operation is performed on F8 to further integrate cross-path information and optimize the multi-level expression of features to generate an optimized feature map F9. Finally, concatenate F9 and F6 in the channel dimension to form a fused feature map F10.

[0165] S333: Perform a Conv3×3 operation on F10 to further integrate cross-path information and optimize the multi-level expression of features to obtain a feature map F11; input F11 into the target detection head Head, use the Head to detect F11, and output a tensor containing detection information. Each row corresponds to a detection result. The detection result is the detection result of a small target in the haze scene, including the bounding box coordinates, the small target category label, and the confidence score.

[0166] The head is responsible for directly detecting the object's category, location (including the coordinates of the bounding box), and confidence level of existence from the feature map. The output of the detection head is a three-dimensional tensor, where each row contains the category probability, bounding box coordinates, and confidence level of existence of each detection box at each spatial location (relative to the feature map).

[0167] Bounding box prediction: Each cell is responsible for predicting the coordinates and confidence of several bounding boxes; the coordinates of the bounding box include: the center point (x, y) (relative to the position of the feature map cell), width and height (w, h) (scaled relative to the width and height of the entire image);

[0168] Category and confidence prediction: In addition to coordinate predictions, each bounding box also has a confidence prediction, which is used to indicate the probability of an object of a specific category existing within the box.

[0169] S3 includes but is not limited to the following embodiments: inputting a haze scene image under a haze scene into a dehazing and shallow feature extraction module; the size of the haze scene image is 512×512 pixels and 3 channels; using a convolution kernel of size 3×3 to perform a convolution operation on the haze image, and then performing batch normalization and activation function processing (i.e., the CBS operation in the figure) to obtain a preliminary haze feature map F1, the size of F1 is 256×256 pixels and 32 channels; inputting F1 into a multi-scale fusion large convolution kernel enhancement module MSE-LKC to obtain a generated feature map F2, the size of which is 256×256 pixels and 32 channels; inputting F2 into a parallel enhanced attention module PEA to obtain a feature map F3, the size of which is 256×256 pixels and 32 channels; then inputting F1 and F3 into a feature fusion module Fusion, and obtaining an enhanced feature map F4, the size of which is 256×256 pixels and 64 channels through cross-path information integration;

[0170] In the MSE-LKC module, the input feature map F1 undergoes batch normalization (BatchNorm), and then uses a convolution kernel of size 7×7 to perform convolution (Conv7×7) to extract preliminary features; then uses a convolution kernel of size 5×5 to perform convolution (Conv5) to generate the first stage feature map F1_3, which has a size of 256×256 pixels and 32 channels; F1_3 is input to the depth expansion convolution module with an expansion rate of 7×7 (i.e. Figure 4 DWDConv7 in ), a deep dilated convolution module with a dilation rate of 13×13 (i.e. Figure 4 DWDConv13 in ), a deep dilated convolution module with a dilation rate of 19×19 (i.e. Figure 4 DWDConv19 in

[19] is used to extract multi-scale features and generate three corresponding feature sub-maps F1_4, F1_5 and F1_6, the size of which are all 256×256 pixels and 32 channels. The splicing operation (C) is used to splice these sub-maps in the channel dimension to obtain the fused feature map F1_7, which has a size of 256×256 pixels and 96 channels. The fused feature map F1_7 is input into a convolution kernel of size 1×1 for convolution operation, and then processed by GELU activation function, and then convolved again with a 1×1 convolution kernel to generate the output feature map F1_10, which has a size of 256×256 pixels and 32 channels. At the same time, the initial input feature map F1 is added to F1_10 through residual connection to obtain the feature map F2, which has a size of 256×256 pixels and 32 channels.

[0171] In the PEA module, first batch normalization operation (BatchNorm) is performed on F2 to generate the normalized feature map F2_1; the feature map F2_1 is input into the pixel-level attention module for processing, and a convolution kernel of size 1×1 is used to convolve F2_1 (Conv1×1), the number of channels is adjusted, and the intermediate feature map is generated. After that, the intermediate feature map is nonlinearly transformed by the GELU activation function to generate the feature map F2_1_1; a 1×1 convolution kernel is used to convolve F2_1_1 again to generate the intermediate feature map F2_1_2, and it is normalized by the Sigmoid activation function to generate the pixel-level attention weight; the input feature map F2_ 1 and the pixel attention weight are multiplied element-by-element (×) to generate the enhanced feature sub-graph F2_2; the feature graph F2_1 is input into the channel-level attention module for processing. First, global average pooling (GAP) is performed on F2_1 to generate global information in the channel dimension. Then, a convolution kernel of size 1×1 is used to convolve the pooling result to obtain the feature graph T1, and the intermediate feature graph T2 is generated by the GELU activation function; the intermediate features of the feature graph T2 are processed by the 1×1 convolution kernel, and the channel attention weight is generated by the Sigmoid activation function; the channel attention weight is multiplied element-by-element with F2_1 in the channel dimension (×) to generate the weight-adjusted feature sub-graph F2_3; The feature map F2_1 is input into the simple pixel attention module for processing. A convolution kernel of size 1×1 is used to convolve F2_1 to generate a preliminary pixel feature map X1. A convolution kernel of size 3×3 is used to further extract local pixel correlation from the feature map X1 to generate a local feature map X2. Through another branch, a 1×1 convolution kernel is directly used to convolve F2_1 to obtain a feature map X3. After adding the results of feature maps X2 and X3, a simple pixel attention weight is generated through the Sigmoid activation function. The simple pixel attention weight is multiplied element-by-element by F2_1 (×) to generate a feature sub-map F2_4. F2_2, F2_3, F2_4 is concatenated (Concat operation) in the channel dimension to generate the fused feature map F2_5, which is 256×256 pixels and 96 channels. F2_5 is subjected to a 1×1 convolution operation (Conv1×1) to reduce the dimension to 32 channels and then undergoes nonlinear processing through the GELU activation function to generate the feature map F2_6. F2_6 is convolved again with a 1×1 convolution kernel to generate the feature map F2_7. The initial input feature map F2 is added to F2_7 through the residual connection (ResidualConnection) to generate the final output feature map F3, which is 256×256 pixels and 32 channels.

[0172] In the fine feature capture multi-channel convolution module, the input feature map F4 is first convolved with a convolution kernel (Conv1×1) of size 1×1 to generate a feature map F4_1 of size 256×256 pixels and 32 channels. Subsequently, F4_1 is split into three feature processing paths: the first path: F4_1 is sequentially convolved with two convolution kernels (Conv3×3) of size 3×3 to generate a feature map F4_2 of size 256×256 pixels and 32 channels. The second path: F4_1 is sequentially convolved with a convolution kernel of size 3×3 and a convolution kernel of size 5×5 (Conv5×5) to generate a feature map F4_3 of size 256×256 pixels and 32 channels. The third path: Path: F4_1 is directly convolved through a 5×5 convolution kernel (Conv5×5) to generate a feature map F4_4 with a size of 256×256 pixels and 32 channels; the output feature maps F4_2, F4_3 and F4_4 of the three paths are spliced ​​in the channel dimension (C) to obtain the fused feature map F4_5 with a size of 256×256 pixels and 160 channels; Subsequently, F4_5 is convolved through a 1×1 convolution kernel (Conv1×1) for channel integration to generate a feature map F4_6 with a size of 256×256 pixels and 64 channels; Finally, F4_6 is input into a 3×3 convolution kernel (Conv3×3) to generate the final output feature map F5 with a size of 256×256 pixels and 64 channels;

[0173] In the multi-scale feature integration enhancement detection module, a 3×3 deformable convolution (DConv) operation is applied to the feature map F2 to obtain an enhanced feature map F6, which has a size of 256×256 pixels and 32 channels. A 3×3 deformable convolution (DConv) operation is applied to the feature map F3 to obtain an enhanced feature map F7, which has a size of 256×256 pixels and 32 channels. The feature maps F5 and F7 are spliced ​​in the channel dimension to form a fused feature map F8, which has a size of 256×256 pixels and 96 channels. A 3×3 convolution kernel is applied to F8 to obtain an enhanced feature map F7, which has a size of 256×256 pixels and 32 channels. Convolution operation is performed to obtain the optimized feature map F9, which is 256×256 pixels and 64 channels. F9 is concatenated with F6 in the channel dimension to form a fused feature map F10, which is 256×256 pixels and 96 channels. F10 is convolved with a 3×3 convolution kernel to obtain the final optimized feature map F11, which is 256×256 pixels and 64 channels. F11 is input into the target detection head (Head), which outputs a tensor containing detection information. Each row corresponds to a detection result, including bounding box coordinates, small object category label and confidence score.

[0174] S4: Based on the constructed FASOD-Net model for small object detection in haze scenes, train and update the parameters of each layer. First, initialize all neural network parameters and set model-related hyperparameters, including training rounds, batch size, weight decay coefficient, learning rate, and total number of iterations.

[0175] S4 includes but is not limited to the following embodiments: training the small target detection network FASOD-Net in the haze scene, the entire training process is set to 100 epochs, the weight decay coefficient is set to 0.01, the learning rate is set to 0.00001, and the batch size is set to 8; the number of iterations of the first 50 epochs is 300, the number of iterations of the next 50 epochs is 400, and the total number of iterations is 35,000 times; in each round of training, Mosaic data enhancement is turned on, and the brightness and contrast of the image are randomly cropped, rotated, scaled, and adjusted to enhance the generalization ability of the model; the resolution of the input image is uniformly adjusted to a format of 512×512 to meet the network input requirements; after initializing the parameters, the small target training set and verification set data of the haze scene processed by Step 2 are divided into multiple batches, and each batch of data is input into FASOD-Net for training, the training loss value (loss) of the batch is calculated, and the parameters of the model are updated according to the loss value; after completing the training of all batches of data in the entire training set, the verification stage is entered;

[0176] During the validation phase, the validation set data is fed into FASOD-Net in batches, and the validation loss (batch_loss) for each batch is calculated. Based on the training loss and validation batch_loss values, FASOD-Net dynamically adjusts the learning rate and other optimization parameters to improve training efficiency and model performance. The entire training process continues through multiple rounds of iterations until the validation loss (batch_loss) converges, indicating that the model has good detection capabilities and training is complete. Ultimately, the trained FASOD-Net can achieve high-precision small target detection in complex haze scenarios.

[0177] S5: After the haze scene small target detection model is trained, the trained haze scene small target detection model is applied to the haze scene image to detect small targets. Finally, the small target detection results in the haze scene are output, including the small target bounding box coordinates, small target category label and confidence score.

[0178] Based on the above, the present invention designs a deep learning network FASOD-Net (Fog-Aware Small Object Detection Network) for small target detection in haze scenes;

[0179] (1) In the dehazing and shallow feature extraction module of FASOD-Net, a multi-scale fusion large convolution kernel enhancement module MSE-LKC and a parallel enhanced attention module PEA are designed; the MSE-LKC module extracts multi-scale shallow features by combining multi-scale dilated convolution kernels with batch normalization and activation functions; the PEA module combines pixel-level attention, channel-level attention mechanism and simple pixel-level attention to enhance the shallow feature expression ability from the spatial and channel dimensions, effectively remove haze interference and improve the ability to capture shallow features of small targets;

[0180] (2) In the fine feature capture multi-channel convolution module of FASOD-Net, a fine feature capture method combining multi-channel feature diversion and multi-scale convolution kernel is designed; this module enhances the local feature expression ability of small targets through multi-path feature extraction, while optimizing the global semantic expression of the target, effectively improving the detection accuracy of small targets in complex backgrounds;

[0181] (3) In the multi-scale feature integration enhanced detection module of FASOD-Net, a feature integration method combining deformable convolution and multi-path feature fusion is designed; this module optimizes the expression of target boundaries and texture features by effectively integrating features of different scales and levels, significantly improving the accuracy and robustness of small target detection in complex haze scenes.

[0182] Although the present invention has been described above with reference to embodiments, various modifications may be made thereto and equivalent components may be substituted without departing from the scope of the present invention. In particular, as long as there are no structural conflicts, the various features of the embodiments disclosed herein may be combined with each other in any manner, and the omission of an exhaustive description of such combinations in this specification is solely for the sake of space and resource conservation. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A small target detection method in haze scenes based on deep learning, characterized by: The specific steps are as follows: S1: Construct an original dataset for small target detection in haze scenes, or use an existing publicly available dataset for small target detection in haze scenes; first collect image data of small targets in haze environments, then use annotation tools to select the locations of small targets and label them; S2: Preprocess the original dataset of small target detection in haze scenes to obtain a preprocessed dataset of small target detection in haze scenes, and divide the dataset to facilitate subsequent training of a deep learning-based small target detection model in haze scenes. S3: Construct a small target detection network FASOD-NET in haze scenes; the specific steps are as follows: S31: Input the haze scene image into the defogging and shallow feature extraction module. The construction and execution process of the defogging and shallow feature extraction module is as follows: S311: Apply a CBS operation including a 3×3 convolution, normalization, and activation function to the input haze scene image to obtain a haze shallow feature map F1; S312: Input F1 into the multi-scale fusion large convolution kernel enhancement module MSE-LKC to obtain F2; S313: Input F2 into the parallel enhanced attention module PEA to obtain F3; S314: Input F3 and F1 into Fusion for channel dimension fusion, and obtain the final output feature map F4 by fusing features at different levels; S32: Input F4 into the fine feature capture multi-channel convolution module to obtain F5; S33: Input F2, F3, and F5 into the multi-scale feature integration and enhancement detection module to obtain a detection result image with small object annotations; The operation and construction process of the MSE-LKC module in S312 is as follows: S3121: Perform batch normalization on the input F1 to obtain a feature map F1_1. Then, perform a convolution operation on F1_1 with a 7×7 convolution kernel to extract features to obtain a feature map F1_2. Then, perform a convolution operation on F1_2 with a 5×5 convolution kernel to obtain the first-stage feature map F1_3. S3122: Input F1_3 into a deep dilated convolution module with a dilation rate of 7×7, a deep dilated convolution module with a dilation rate of 13×13, and a deep dilated convolution module with a dilation rate of 19×19 to extract multi-scale features, generating three corresponding feature sub-graphs F1_4, F1_5, and F1_6 respectively; then, concatenate F1_4, F1_5, and F1_6 in the channel dimension through the C operation to obtain the fused feature graph F1_7; S3123: Input F1_7 into a 1×1 convolution kernel for convolution operation to obtain feature map F1_8, process F1_8 through the GELU activation function to obtain feature map F1_9, and perform Conv1×1 convolution operation on F1_9 to obtain the final output feature map F1_10; S3124: Add F1 and F1_10 through residual connection to form a feature map F2 with stronger representation ability; The construction and execution process of the PEA module in S313 is as follows: S3131: Perform batch normalization on F2 to standardize the feature distribution, improve the stability and convergence speed of the model, and obtain the normalized feature map F2_1; S3132: Input F2_1 into three attention modules respectively: S31321: The pixel-level attention module processes the spatial dimension of F2_1, extracts local saliency information, and obtains the enhanced feature subgraph F2_2; S31322: The channel-level attention module operates on the channel dimension of F2_1 to capture global context information and obtain the weighted feature subgraph F2_3; S31323: The simple pixel-level attention module uses a simplified mechanism to process F2_1, supplementing the feature extraction capabilities of other modules to obtain the feature subgraph F2_4; S3133: Concatenate F2_2, F2_3, and F2_4 in the channel dimension to obtain the fused feature map F2_5; S3134: Perform a Conv1×1 convolution operation on F2_5 to achieve feature dimensionality reduction, and perform a nonlinear transformation on the feature map using the GELU activation function to obtain the processed feature map F2_6. Then perform a Conv1×1 convolution operation on F2_6 to further extract features and adapt them to subsequent tasks, obtaining the feature map F2_7. S3135: Add F2 and F2_7 through residual connection to obtain the final output feature map F3; Among them, the process of applying PEA to F2 to obtain F3 is shown in the following formula: F2_1=BatchNorm(F2) F3=F2+Conv1_1(GELU(Conv1_1(C(PA(F2_1),CA(F2_1),SPA(F2_1))))) Among them, BatchNorm represents the batch normalization operation; Conv1_1 represents the convolution operation of size 1×1; GELU represents the application of the activation function GELU; SPA represents the simple pixel-level attention module; CA represents the channel-level attention module; PA represents the pixel-level attention module; C represents the splicing operation in the channel dimension; + represents the residual connection operation; S4: Based on the constructed FASOD-Net model for small object detection in haze scenes, train and update the parameters of each layer. First, initialize all neural network parameters and set model-related hyperparameters, including training rounds, batch size, weight decay coefficient, learning rate, and total number of iterations. S5: After the haze scene small target detection model is trained, the trained haze scene small target detection model is applied to the haze scene image to detect small targets. Finally, the small target detection results in the haze scene are output, including the small target bounding box coordinates, small target category label and confidence score.

2. The method for detecting small targets in haze scenes based on deep learning according to claim 1, characterized in that: The specific steps of S2 are as follows: S21: Cropping the hazy scene images in the original dataset of small target detection in hazy scenes to a uniform size. Since the original dataset is small, to ensure that the dataset has sufficient samples for training, validation, and testing, and to enhance the robustness of the small target detection model in hazy scenes, the model is less sensitive to image quality. S22: Data enhancement processing is then performed to better train the deep learning model, and finally a data-enhanced small target dataset in the haze scene is obtained, which is divided into a training set, a validation set, and a test set in a ratio of 8:1:

1.

3. The method for detecting small targets in haze scenes based on deep learning according to claim 1, characterized in that: The construction and execution process of the pixel-level attention module in S31321 is as follows: S313211: Perform a Conv1×1 convolution operation on F2_1 to adjust the number of channels of the input feature map; then, apply the activation function GELU to further process F2_1 to obtain the activated feature map F2_1_1; S313212: Perform a Conv1×1 convolution operation on F2_1_1 to obtain the intermediate feature map F2_1_2; then, apply the Sigmoid activation function to normalize F2_1_2 to obtain the pixel-level attention weight; S313213: Perform element-wise multiplication of F2_1 and pixel attention weight to weight the feature map and obtain the final feature map F2_2; Its module can enhance the expression of small target features while suppressing irrelevant background information, improving the model's ability to detect small targets in haze scenes; Among them, the process of applying the pixel-level attention module to process F2_1 and obtain F2_2 is shown in the following formula: F2_2=F2_1×S(Conv1_1(GELU(Conv1_1(F2_1)))) Among them, Conv1_1 represents a convolution operation of size 1×1; GELU represents the application of the activation function GELU; S represents the Sigmoid activation function; × represents the element-by-element multiplication operation; The construction and execution process of the channel-level attention module in S31322 is as follows: S313221: Perform global average pooling on F2_1 to generate global information in the channel dimension; then, perform a Conv1×1 convolution operation on the pooled result to obtain the feature map T1, and perform nonlinear processing on T1 through the activation function GELU to generate the intermediate expression feature map T2; S313222: Perform a Conv1×1 convolution operation on T2 and then use the Sigmoid activation function to generate channel attention weights. Perform element-wise multiplication of F2_1 and the channel attention weights to achieve weighted output on the channel and obtain the enhanced output feature map F2_3. Among them, the process of applying the channel-level attention module to process F2_1 to obtain F2_3 is shown in the following formula: F2_3=F2_1×S(Conv1_1(GELU(Conv1_1(GAP(F2_1))))) Among them, GAP represents the global average pooling operation; Conv1_1 represents the convolution operation of size 1×1; GELU represents the application of the activation function GELU; S represents the Sigmoid activation function; × represents the element-by-element multiplication operation; The construction and execution process of the simple pixel-level attention module in S31323 is as follows: S313231: Perform a Conv1×1 convolution operation on F2_1 to obtain a preliminary pixel feature map X1; then, X1 is convolved with another 3×3 convolution kernel to further extract local pixel correlation information to obtain a feature map X2; S313232: Pass F2_1 through another branch to directly perform a Conv1×1 convolution operation to obtain the feature map X3; then, add X2 and X3 and generate pixel attention weights through the Sigmoid activation function; S313233: Perform element-wise multiplication of F2_1 and pixel attention weight to obtain the enhanced output feature map F2_4; Among them, the process of applying a simple pixel-level attention module to process F2_1 to obtain F2_4 is shown in the following formula: F2_4=F2_1×S(Conv3_3(Conv1_1(F2_1))+Conv1_1(F2_1)) Among them, Conv1_1 represents a convolution operation of size 1×1; Conv3_3 represents a convolution operation of size 3×3; S represents the Sigmoid activation function; + represents the addition operation of the results of the two branches; × represents the element-by-element multiplication operation.

4. The method for detecting small targets in haze scenes based on deep learning according to claim 1, characterized in that: The construction and execution process of the fine feature capture multi-channel convolution module in S32 is as follows: S321: Perform a Conv1×1 convolution operation on F4 to adjust the number of channels and extract preliminary features to generate a feature map F4_1; S322: Feature splitting is performed on F4_1, which is divided into three feature processing paths: in the first path, two Conv 3×3 convolution operations are performed to extract local deep features and generate feature map F4_2; in the second path, Conv3×3 convolution operations and Conv5×5 convolution operations are performed in sequence to further extract features with a larger receptive field and generate feature map F4_3; in the third path, Conv5×5 convolution operations are performed to further extract features with a larger receptive field and generate feature map F4_4; S323: Concatenate F4_2, F4_3, and F4_4 output from the three paths in the channel dimension to obtain the fused feature map F4_5; perform channel integration on F4_5 through the Conv1×1 convolution operation to generate the final small target deep feature map F4_6; S324: Perform a Conv3×3 convolution operation on F4_6 to further integrate cross-path information and optimize local feature expression to generate feature map F5. Through the above steps, its module combines multi-scale convolution kernels and diversion structures to effectively extract deep semantic features of small targets and enhance the model's detection ability for small targets.

5. The method for detecting small targets in haze scenes based on deep learning according to claim 1, characterized in that: The construction and execution process of the multi-scale feature integration enhancement detection module in S33 is as follows: S331: Apply a deformable convolution operation to the input F2 to dynamically adapt to the morphological characteristics of small targets and complex background noise in the haze scene, generating an enhanced feature map F6; apply a deformable convolution operation to the input F3 to generate an enhanced feature map F7; S332: Concatenate F5 and F7 in the channel dimension to form a fused feature map F8, which enhances the multi-scale expression capability of features and takes into account the unification of details and global features; Subsequently, a Conv3×3 convolution operation is performed on F8 to further integrate cross-path information and optimize the multi-level expression of features to generate an optimized feature map F9. F9 is then concatenated with F6 in the channel dimension to form a fused feature map F10. S333: Perform a Conv3×3 operation on F10 to further integrate cross-path information and optimize the multi-level expression of features to obtain a feature map F11; input F11 into the target detection head Head, use Head to detect F11, and output a tensor containing detection information. Each row corresponds to a detection result, which is a small target detection result in the haze scene, including the bounding box coordinates, small target category label and confidence score.

Images