An adaptive dual-path unmanned aerial vehicle detection method and system for foggy environment

Abstract

The application discloses a kind of fog environment-oriented adaptive dual-path unmanned aerial vehicle detection method and system, belong to computer vision, image processing and low-altitude safety monitoring cross technical field.The present application aims to solve the problem of high false negative rate, waste of computing resources, dehazing and detection task mismatch in the fog environment of prior art.The method comprises: receiving input image and calculating global fog concentration coefficient;With preset threshold 0.52 comparison, adaptive selection dehazing enhancement path or direct detection path;Moderate fog through ASC image enhancement network processing image;Finally, through improved YOLO P2 target detection network execution detection.The application realizes computing resource on-demand allocation, improves the fog micro-target detection precision and system real-time, applicable to low-altitude safety monitoring scene.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary technical field of computer vision, image processing and low-altitude safety monitoring, and specifically relates to an adaptive dual-path UAV detection method and system for foggy environments. Background Technology

[0002] With the booming development of the low-altitude economy, the security threat posed by unauthorized drone flights is becoming increasingly serious. Existing countermeasures systems mostly employ solutions that combine radar, radio frequency, and visible light vision. Among these, visible light-based visual detection methods are widely used due to their advantages of lower cost, providing intuitive visual evidence, and accurate spatial positioning. However, in adverse weather conditions such as fog, this technology faces three fundamental challenges: 1. Severe Image Degradation: The scattering effect of fog significantly reduces the quality of captured images, specifically manifested as reduced contrast, color distortion, and blurred details. The physical process can be described by an atmospheric scattering model:

[0003] in, For the observed foggy images, For potential clear scenarios, This represents the global atmospheric light value. This refers to the scene's transmittance. Increased fog concentration leads to... Reducing the size of the target causes the boundary between the target and the background to merge, resulting in a sharp decrease in feature discrimination.

[0004] 2. Loss of features for small targets: Mainstream detection algorithms, in order to also detect large targets, typically construct feature pyramids with strides of 8, 16, and 32. Let the input image size be... After deep downsampling by the network, the feature map size is significantly reduced. For high-altitude flight, which may only occupy a small portion of the image... For micro-drone targets with a resolution of less than one pixel, their effective features are easily lost during multiple downsampling processes, making them undetectable by deep networks and resulting in serious missed detections.

[0005] 3. Inefficient use of computational resources: The traditional cascaded approach of "full dehazing first, then detection" suffers from efficiency bottlenecks. Under light fog or clear weather conditions, performing dehazing algorithms on all video frames incurs unnecessary computational overhead, making it difficult to meet the frame rate requirements of real-time monitoring. Furthermore, general dehazing algorithms may introduce artifacts or cause image distortion, negatively impacting detection accuracy.

[0006] While the existing detection method based on radio frequency features, proposed in patent publication number CN115932751B, is unaffected by weather, it is susceptible to electromagnetic interference and has limited positioning accuracy. Therefore, there is an urgent need for a visual detection solution that can intelligently adapt to different fog concentrations, collaboratively optimize image restoration and target detection tasks, and achieve the best balance between accuracy and speed. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide an adaptive dual-path UAV detection method and system for foggy environments, which addresses the shortcomings of the prior art. This method solves the technical problems of high false negative rate, low efficiency caused by static allocation of computing resources, and mismatch between defogging results and detection tasks in the detection of small targets by UAVs in foggy environments.

[0008] The present invention adopts the following technical solution: An adaptive dual-path UAV detection method for foggy environments includes the following steps: S1. Receive the input image and calculate the global fog concentration coefficient of the input image; S2. Compare the global fog concentration coefficient with a preset threshold and perform adaptive path decision-making: if the global fog concentration coefficient is greater than or equal to the preset threshold, proceed to step S3; if the global fog concentration coefficient is less than the preset threshold, proceed to step S4. S3. The input image is dehazing and enhanced using an ASC image enhancement network optimized for detection tasks to obtain an enhanced image; S4. Input the input image or the enhanced image into the improved YOLO P2 target detection network to perform UAV target detection, and output the detection results, which include the location bounding box and category confidence of the UAV target.

[0009] Preferably, in step S1, a lightweight statistical algorithm based on dark channel priors is used to calculate the global fog concentration coefficient. The specific process is as follows: Define the dark channel of the input image as:

[0010] in, It is based on pixels The local area centered on y represents the pixel value in the red, green, and blue channels; The mean value of the dark channel is calculated to obtain the global fog concentration coefficient.

[0011] Preferably, in step S2, the preset threshold is 0.52.

[0012] Preferably, in step S3, the ASC image enhancement network adopts an encoder-decoder structure, and its input processing is as follows: Original foggy image use Operator extracts edge feature map The original foggy image With the edge feature map The enhanced input is obtained by concatenating along the channel dimension. .

[0013] Preferably, a convolutional block attention module is embedded after the encoder key layer of the ASC image enhancement network, and the processing procedure of the convolutional block attention module is as follows: First, channel attention processing is performed on the intermediate feature map to generate channel weight vectors; Then, spatial attention processing is applied to the channel-weighted features to generate a spatial weight matrix; Final output features for:

[0014] in, The characteristics are those of the channel after weighting. This is the spatial weight matrix.

[0015] Preferably, the ASC image enhancement network is trained using a composite loss function, which is:

[0016] in, For structural similarity loss, To rebuild the losses, In order to perceive loss, , , These are the weighting coefficients.

[0017] Preferably, in step S4, the improved YOLO P2 target detection network uses the YOLOX-S model as the baseline, extracts shallow high-resolution features with a downsampling step size of 4 in Stage 2 of the CSPDarknet backbone network, and forms the P2 feature layer after processing by the transition layer. At the same time, it extracts the P3 feature layer with a downsampling step size of 8, the P4 feature layer with a downsampling step size of 16, and the P5 feature layer with a downsampling step size of 32, and constructs a four-layer feature pyramid containing P2, P3, P4, and P5.

[0018] Preferably, the four-layer feature pyramid performs deep feature fusion through the top-down path of the feature pyramid network and the bottom-up path of the path aggregation network, and the fusion process satisfies:

[0019] in, For convolution operations, The output features of Stage 5 of the CSPDarknet backbone network are shown. The output features of Stage 4 of the CSPDarknet backbone network are shown. The output features of Stage 3 of the CSPDarknet backbone network. The output features of Stage 2 of the CSPDarknet backbone network are shown. To perform upsampling operations on the P5 feature layer. This is to perform upsampling operations on the P4 feature layer. This is to perform upsampling operations on the P3 feature layer.

[0020] Preferably, the improved YOLO P2 target detection network is configured with four detection heads corresponding to the four-layer feature pyramid, namely the P2 detection head, P3 detection head, P4 detection head, and P5 detection head; wherein, the P2 detection head is used to detect UAV targets with a pixel area of ​​10×10 to 32×32, the P3 detection head is used to detect small-sized UAV targets, the P4 detection head is used to detect medium-sized UAV targets, and the P5 detection head is used to detect large-sized UAV targets.

[0021] Secondly, embodiments of the present invention provide an adaptive dual-path UAV detection system for foggy environments, comprising: A fog concentration calculation module is used to receive an input image and calculate the global fog concentration coefficient of the input image; An adaptive decision-making module is used to compare the global fog concentration coefficient with a preset threshold and select a processing path based on the comparison result. The image enhancement module is used to perform dehazing enhancement processing on the input image using an ASC image enhancement network when the adaptive decision module determines that the global fog concentration coefficient is greater than or equal to the preset threshold, so as to obtain an enhanced image. The target detection module is used to input the input image or the enhanced image into the improved YOLO P2 target detection network to perform UAV target detection, and output the detection result including the bounding box of the UAV target location and the category confidence.

[0022] Thirdly, a computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described adaptive dual-path UAV detection method for foggy environments.

[0023] Fourthly, embodiments of the present invention provide a computer-readable storage medium including a computer program, which, when executed by a processor, implements the steps of the above-described adaptive dual-path UAV detection method for foggy environments.

[0024] Fifthly, a chip includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described adaptive dual-path UAV detection method for foggy environments.

[0025] In a sixth aspect, embodiments of the present invention provide an electronic device including a computer program, which, when executed by the electronic device, implements the steps of the above-described adaptive dual-path UAV detection method for foggy environments.

[0026] Compared with the prior art, the present invention has at least the following beneficial effects: An adaptive dual-path UAV detection method for foggy environments is proposed. By first calculating the global fog density coefficient for path selection, it achieves on-demand allocation of computational resources: in moderate to heavy fog, a defogging enhancement path is initiated to avoid missed detections caused by image degradation; in light fog or clear weather, the detection path is directly entered, avoiding unnecessary computational overhead and perfectly balancing real-time performance and detection accuracy. Simultaneously, a detection-optimized ASC image enhancement network and an improved YOLO P2 target detection network are specifically integrated. These two networks work together to adapt to the characteristics of foggy scenes—the defogging network outputs high-quality images suitable for detection, while the detection network accurately captures minute targets. This method fundamentally solves the problems of image degradation in foggy weather, loss of features of minute targets, and mismatch between defogging and detection tasks. It is suitable for scenarios with stringent requirements for both real-time performance and accuracy, such as low-altitude safety monitoring.

[0027] Furthermore, the dark channel prior conforms to the physical characteristics of foggy images—fog scattering leads to lower pixel values ​​in the dark channel. By defining the dark channel and calculating its mean, the global fog concentration level can be intuitively reflected, avoiding misjudgments of single pixels or local areas. Compared to complex fog concentration detection algorithms, this method requires no heavy computation and can quickly process each frame of the surveillance video stream, meeting real-time decision-making needs. Simultaneously, the combination of local area sampling and global mean calculation in the dark channel takes into account both local fog condition differences and overall fog concentration levels, resulting in accurate and reliable calculation results that provide a solid basis for subsequent path decisions.

[0028] Furthermore, extensive experiments using a foggy day dataset validated the pre-defined threshold: the system analyzed the detection accuracy gain and computational latency brought about by defogging operations at different thresholds, ultimately determining 0.52 as the global performance balance point. This threshold accurately distinguishes between light fog / clear days and moderate to heavy fog scenarios: when the fog concentration coefficient ≥ 0.52, the improvement in detection accuracy from defogging significantly outweighs the computational cost; when the fog concentration coefficient < 0.52, skipping the defogging step does not affect the detection effect and can even reduce latency. By fixing this optimal threshold, decision fluctuations caused by dynamic threshold adjustments are avoided, ensuring the consistency and reliability of path selection, and enabling the system to maintain stable performance under complex and changing weather conditions.

[0029] Furthermore, one of the core problems with foggy images is edge blurring, leading to the merging of targets with the background boundaries and a decrease in feature discrimination. By extracting edge feature maps from the input fog image using the Sobel operator and then performing channel concatenation with the original fog image, high-frequency contour information is explicitly injected into the network. The edge feature maps contain the contour structure of the targets, guiding the network to focus on restoring object boundaries during defogging, avoiding the smoothing or loss of details common in traditional defogging algorithms. Simultaneously, the channel concatenation method requires no complex preprocessing and is naturally compatible with the encoder-decoder network structure, allowing the network to learn fog image restoration and edge preservation simultaneously during training. The enhanced output image not only has improved clarity but also highlights the contour features of targets such as drones, providing better input data for subsequent detection networks.

[0030] Furthermore, in foggy images, invalid information such as a uniform sky background can easily interfere with feature extraction, while the features of drone targets may be obscured by fog. The CBAM module first generates a weight vector through channel attention processing, emphasizing effective feature channels containing target information and weakening invalid channels; then, it generates a weight matrix through spatial attention processing, focusing on key areas in the image where drones may exist and suppressing background noise. This channel-spatial dual attention mechanism can adaptively distinguish between effective features and interfering information, sharpen target region features, and improve the signal-to-noise ratio of the feature map. Compared to dehazing networks without attention mechanisms, it makes the target features of the enhanced image more prominent, significantly reducing the recognition difficulty of subsequent detection networks and reducing the risk of missed detections.

[0031] Furthermore, traditional single loss functions can easily lead to color distortion, texture smoothing, or structural deformation in dehazed images. This composite loss function, however, complements each other through three constraints: the reconstruction loss ensures the dehazing process follows an atmospheric scattering model, guaranteeing pixel-level accuracy; the perceptual loss utilizes the feature extraction capabilities of the pre-trained VGG network to constrain the restored image to approximate a truly clear image in terms of high-level semantic features, addressing the image graying problem; and the structural similarity loss maintains the overall structure and texture details of the image, avoiding target contour deformation. These three factors work synergistically through weighted coefficient adjustments, driving the network to output physically accurate, textured, and structurally complete images, rather than simply pursuing visual enhancement, perfectly matching the feature requirements of subsequent detection networks for the input image.

[0032] Furthermore, traditional detection networks typically use P3-P5 layers (downsampling strides of 8, 16, and 32). After multiple downsampling steps, the features of tiny targets (occupying only 10×10-32×32 pixels) from high-altitude UAVs are easily discarded. In the CSPDarknet backbone network, Stage 2 extracts shallow high-resolution features with a downsampling stride of 4, which are then processed to form the P2 layer, constructing a four-layer feature pyramid from P2 to P5. The P2 layer retains rich spatial detail information, fully capturing the effective features of tiny targets. Simultaneously, the four-layer structure covers a multi-scale range from extremely small to large targets, adapting to UAVs of different flight altitudes and sizes. This provides assurance for tiny target detection from the feature extraction level, significantly reducing the false negative rate.

[0033] Furthermore, through the top-down path of the feature pyramid network and the bottom-up path of the path aggregation network, features from layers P2 to P5 are fused bidirectionally: high-level features provide rich semantic information to help the network identify target categories; low-level features provide high-resolution detailed information to help accurately locate target positions. The fusion process adjusts feature dimensions through convolution operations and matches feature sizes through upsampling operations, ensuring the compatibility and effectiveness of feature fusion across layers. This fusion structure avoids the limitations of unidirectional feature transmission, ensuring that each feature layer contains both semantic and detailed information. Whether the target is small or large, its features are fully enhanced, laying the foundation for accurate identification and localization by the subsequent detection head.

[0034] Furthermore, considering the characteristics of the four feature layers, the detection heads have clearly defined roles: the P2 detection head is specifically adapted to the high-resolution features of the P2 layer, focusing on extremely small UAV targets of 10×10-32×32 pixels, filling the detection gap of traditional detection networks for targets of this scale; the P3-P5 detection heads correspond to small, medium, and large-sized targets respectively, ensuring the detection accuracy of targets of various scales by adapting to feature layers of different resolutions. This one-to-one correspondence between feature layers and detection heads allows each detection head to focus on feature recognition of targets of a specific scale, avoiding performance loss caused by cross-scale detection, and significantly improving the detection network's accuracy in recognizing and locating UAV targets of different scales in foggy weather.

[0035] It is understood that the beneficial effects of the second to sixth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0036] In summary, the method of this invention achieves an adaptive optimal balance between detection accuracy and processing efficiency, and significantly improves the detection rate of targets, especially small UAVs, in foggy weather. It effectively solves key problems in existing technologies, such as high false negative rate, low resource utilization, and mismatch between defogging and detection tasks.

[0037] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0038] Figure 1 This is a flowchart illustrating the overall system framework and adaptive decision-making process of the present invention. Figure 2 This is a detailed structural diagram of the ASC defogging network of the present invention; Figure 3 This is a schematic diagram of the feature pyramid structure of the improved YOLO P2 target detection network of this invention; Figure 4 A schematic diagram of a computer device provided in an embodiment of the present invention; Figure 5 This is a block diagram of a chip provided according to an embodiment of the present invention.

[0039] Among them, 60. Computer equipment; 61. Processor; 62. Memory; 63. Computer program; 600. Electronic device; 610. Processing unit; 620. Storage unit; 6201. Random access memory unit; 6202. Cache memory unit; 6203. Read-only memory unit; 6204. Program / utility; 6205. Program module; 630. Bus; 640. Display unit; 650. Input / output interface; 660. Network adapter; 700. External device. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] In the description of this invention, it should be understood that the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0042] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0043] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Additionally, the character " / " in this invention generally indicates that the preceding and following objects have an "or" relationship.

[0044] It should be understood that although terms such as first, second, third, etc., may be used in the embodiments of the present invention to describe the preset range, these preset ranges should not be limited to these terms. These terms are only used to distinguish the preset ranges from one another. For example, without departing from the scope of the embodiments of the present invention, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.

[0045] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0046] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0047] To address the issues of missed detections due to low image quality in complex foggy environments and loss of small target features under high-resolution conditions in existing technologies, this invention provides an adaptive dual-path UAV detection method for foggy environments. First, the fog density coefficient of the input image is calculated and compared with a preset threshold (0.52) to select the processing path. For images in moderate to heavy fog, an ASC image enhancement network based on edge priors and attention mechanisms is used for texture enhancement. In the detection stage, a YOLOP2 network with a P2 high-resolution feature layer is introduced for target recognition. While ensuring near real-time performance, this significantly improves the detection accuracy of UAV targets under complex weather conditions, effectively solving the problem of missed detections in foggy environments.

[0048] Please see Figure 1 This invention discloses an adaptive dual-path UAV detection method for foggy environments. It constructs a collaborative framework of environmental perception, adaptive branching, and task optimization processing. This framework abandons a fixed process, dynamically selects the optimal processing path based on real-time fog concentration assessment, and performs targeted collaborative design for image enhancement and detection modules. The method includes the following steps: S1, Lightweight Environmental Perception and Intelligent Decision-Making The decision-making center of the adaptive framework is responsible for quickly assessing the fog condition of the input image and triggering the corresponding processing path.

[0049] S101, Fast Quantization of Fog Concentration: A lightweight statistical algorithm based on dark channel prior is employed. For the input image... Its secret passage Defined as:

[0050] in, It is based on pixels The local area centered on the fog. Global fog concentration coefficient. Obtained by calculating the mean of the dark channel:

[0051] This coefficient It can effectively and quickly reflect the overall fog concentration level of an image.

[0052] S102, Adaptive Path Decision: Set an optimal threshold that has been validated through extensive experiments. (In this embodiment) =0.52). This threshold was set on a foggy dataset, and the system analyzed different... The global performance balance point is determined after considering the detection accuracy gain and computational delay introduced by the dehazing operation. The decision function is defined as follows:

[0053] This mechanism enables "on-demand allocation" of computing resources and is key to ensuring the overall efficiency of the system.

[0054] S2, ASC Image Enhancement Network Optimized for Detection Tasks When the decision is At this point, the image enters the ASC image enhancement network. This network does not aim for general visual enhancement, but is specifically designed to improve the performance of subsequent detectors. Its innovation is reflected in three aspects: 1) Edge-prior-guided input enhancement: To address the edge blurring problem caused by fog, the network enhances the original fog image at the input. Rather than Edge feature map extracted by the operator Perform channel splicing, i.e. This move explicitly injects high-frequency structural information into the network, enhancing its ability to perceive the target's outline.

[0055] 2) Dual Attention Feature Modulation Mechanism: Convolutional Block Attention (CBAM) modules are embedded in the network feature extraction layer to suppress background noise and focus on the target region. Given intermediate feature maps... : (1) Channel attention: Generate channel weight vector , used to emphasize information-rich feature channels.

[0056] (2) Spatial attention: Features after channel weighting Generate the spatial weight matrix. It is used to focus on key areas in space.

[0057] The final output features are This mechanism can effectively suppress interference from a uniform sky background in foggy weather and sharpen regional features that include targets such as drones.

[0058] 3) Loss function for multi-objective joint optimization: The dehazing enhancement network training uses a composite loss function to ensure that the restored image conforms to physical laws and has texture characteristics that are conducive to detection.

[0059] (1) Reconstruction losses:

[0060] in, The loss, which provides the network with a clear image, transmittance, and atmospheric light, constrains the restoration process to an atmospheric scattering model.

[0061] (3) Perceived loss:

[0062] in, This is a feature extractor for a pre-trained VGG network. The loss constrains the reconstructed image at a high-level semantic feature level. Approaching real-life clear images This is the key to solving the problems of "grayish images" and texture smoothing after dehazing using traditional methods.

[0063] (3) Total loss:

[0064] in, The structural similarity loss is used to preserve image structure. This combined loss collectively drives the network to output high-quality images that are beneficial for detection tasks.

[0065] Please see Figure 2 The ASC image enhancement network aims to address the shortcomings of traditional dehazing methods in visual inspection tasks, such as insufficient texture restoration and color distortion. The network primarily uses an encoder-decoder structure and incorporates the following key improvements: Multi-source input fusion: The network input is not a single path. In addition to the original foggy image... In addition, parallel use Operator extracts its edge feature map Subsequently, in the channel dimension, and The data is stitched together to create an enhanced input. This is intended to explicitly provide the network with high-frequency contour information, guiding it to better recover object boundaries.

[0066] Attention Feature Enhancement: Convolutional Block Attention (CBAM) modules are embedded after key layers of the encoder. CBAM performs channel attention and spatial attention sequentially. Channel attention recalibrates the importance of each feature channel, while spatial attention focuses on key spatial regions of the feature map. This mechanism enables the network to adaptively suppress activation responses in uniform, meaningless background regions in foggy conditions, while enhancing features in regions that may contain targets such as drones.

[0067] Parameter estimation and physical reconstruction: Decoder end output transmittance map and global atmospheric light value The estimate. Finally, based on the atmospheric scattering model. The restored clear image was obtained through calculation. .

[0068] During network training, a composite loss function is used for supervision. Ensure pixel-level accuracy; Maintain the overall structure of the image; The feature distance extracted by the VGG19 network pre-trained on ImageNet is used as a constraint to force the restored image to be close to the real clear image in the high-level semantic feature space, thereby effectively improving color naturalness and texture details and avoiding the result being "grayish".

[0069] S3, High-resolution feature-preserving YOLO P2 object detection network Regardless of the path taken, the final image is detected by the improved YOLO P2 object detection network. Its core improvement is the construction of an enhanced feature pyramid containing ultra-high resolution feature layers to address the problem of feature loss for small objects.

[0070] Please see Figure 3 This invention improves upon the YOLOX-S model, which is based on it. The original YOLOX-S feature pyramid is typically constructed from deep features of a backbone network, comprising three layers: P3 (downsampled 8 times), P4 (downsampled 16 times), and P5 (downsampled 32 times). To capture more subtle UAV target features, this invention makes the following key modifications: 1) Introduction of P2 high-resolution feature layer: Using YOLOX-S as the baseline, in the backbone network, in addition to utilizing the original... In addition to the feature layer, shallow features are extracted. (Step size 4). After the transition layer, the P2 feature layer is formed, with a spatial size of... It retains a wealth of detailed information.

[0071] Additional features are extracted from the shallower Stage 2 of the backbone network. The downsampling stride of this feature map is only 4, preserving rich spatial detail. After necessary convolutional adjustments, this feature layer is used as the P2 layer in the Feature Pyramid Network (FPN) and Path Aggregation Network (PAN) structure, where it is fused with P3, P4, and P5 layers in a top-down and bottom-up multi-scale feature fusion manner.

[0072] 2) Enhanced Feature Pyramid Fusion Structure: A four-layer feature pyramid consisting of P2, P3, P4, and P5 is constructed. Deep fusion is performed through the top-down path of the Feature Pyramid Network (FPN) and the bottom-up path of the Path Aggregation Network (PAN), which is simplified as follows:

[0073] This structure ensures an effective combination of high-level semantic information and low-level high-resolution details.

[0074] 3) Detection head designed specifically for small targets: The detection head corresponds to four levels from P2 to P5, enabling multi-scale prediction.

[0075] Specifically, the detection head corresponding to the P2 layer is specifically responsible for detecting targets with extremely small pixel areas (such as...). to (pixels), directly targeting small, high-altitude drones.

[0076] Corresponding to the expanded feature pyramid, the detection head is also expanded to output at four scales. The newly added scale is specifically responsible for predicting targets detected by the high-resolution P2 layer, and is particularly suitable for identifying extremely small UAV targets in images with a pixel area of ​​less than 32×32.

[0077] Experimental results show that, on the same test set, after introducing the P2 layer, the recall rate of the model for small-sized UAV targets is significantly improved from 0.9320 in the baseline model (without the P2 layer) to 0.9760, demonstrating the necessity of high-resolution features for improving the detection performance of small targets.

[0078] Example 1 An adaptive dual-path UAV detection system for foggy environments is provided. This system can be used to implement the aforementioned adaptive dual-path UAV detection method for foggy environments. Specifically, the adaptive dual-path UAV detection system for foggy environments includes a fog concentration calculation module, an adaptive decision module, an image enhancement module, and a target detection module.

[0079] The fog concentration calculation module is used to receive the input image and calculate the global fog concentration coefficient of the input image. An adaptive decision-making module is used to compare the global fog concentration coefficient with a preset threshold and select a processing path based on the comparison result. The image enhancement module is used to perform dehazing enhancement processing on the input image using an ASC image enhancement network when the adaptive decision module determines that the global fog concentration coefficient is greater than or equal to the preset threshold, so as to obtain an enhanced image. The target detection module is used to input the input image or the enhanced image into an improved YOLO P2 target detection network for UAV target detection, and output the detection results including the bounding box of the UAV target location and the category confidence score.

[0080] This invention provides a terminal device comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, graphics processing units (GPUs), tensor processing units (TPUs), digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or function. The processor described in this embodiment can be used for the operation of an adaptive dual-path UAV detection method for foggy environments, including: The system receives an input image and calculates its global fog density coefficient. It then compares the global fog density coefficient with a preset threshold to perform adaptive path decision-making. If the global fog density coefficient is greater than or equal to the preset threshold, it uses an ASC image enhancement network optimized for detection tasks to perform defogging enhancement on the input image, obtaining an enhanced image. If the global fog density coefficient is less than the preset threshold, it inputs either the input image or the enhanced image into an improved YOLO P2 target detection network for UAV target detection, outputting the detection result, which includes the UAV target's bounding box and category confidence.

[0081] Please see Figure 4 The terminal device is a computer device. In this embodiment, the computer device 60 includes a processor 61, a memory 62, and a computer program 63 stored in the memory 62 and executable on the processor 61. When executed by the processor 61, the computer program 63 implements the adaptive dual-path UAV detection method for foggy environments described in this embodiment. To avoid repetition, details are omitted here. Alternatively, when executed by the processor 61, the computer program 63 implements the functions of each model / unit in the adaptive dual-path UAV detection system for foggy environments described in this embodiment. To avoid repetition, details are omitted here.

[0082] Computer device 60 can be a desktop computer, laptop, handheld computer, cloud server, or other computing device. Computer device 60 may include, but is not limited to, a processor 61 and a memory 62. Those skilled in the art will understand that... Figure 4 This is merely an example of computer device 60 and does not constitute a limitation on computer device 60. It may include more or fewer components than shown, or combine certain components, or different components. For example, computer device may also include input / output devices, network access devices, buses, etc.

[0083] The processor 61 may be a Central Processing Unit (CPU), or other general-purpose processors, graphics processing units (GPUs), tensor processing units (TPUs), digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0084] The memory 62 can be an internal storage unit of the computer device 60, such as a hard disk or memory of the computer device 60. The memory 62 can also be an external storage device of the computer device 60, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on the computer device 60.

[0085] Furthermore, the memory 62 may include both internal storage units of the computer device 60 and external storage devices. The memory 62 is used to store computer programs and other programs and data required by the computer device. The memory 62 can also be used to temporarily store data that has been output or will be output.

[0086] Please see Figure 5 The terminal device is an electronic device 600, which is manifested in the form of a general-purpose computing device. The components of the electronic device may include, but are not limited to: at least one processing unit 610, at least one storage unit 620, a bus 630 connecting different platform components (including storage unit 620 and processing unit 610), a display unit 640, etc.

[0087] The storage unit stores program code, which can be executed by the processing unit 610 to perform the steps described in the method section of this specification according to various exemplary embodiments of the present invention. For example, the processing unit 610 can perform actions such as... Figure 1 The steps are shown in the figure.

[0088] Storage unit 620 may include a readable medium in the form of a volatile storage unit, such as random access memory (RAM) 6201 and / or cache memory 6202, and may further include a read-only memory (ROM) 6203.

[0089] Storage unit 620 may also include a program / utility 6204 having a set (at least one) program module 6205, such program module 6205 including but not limited to: operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0090] Bus 630 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the multiple bus structures.

[0091] Electronic device 600 can also communicate with one or more external devices 700 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with electronic device 600, and / or with any device that enables electronic device 600 to communicate with one or more other computing devices (e.g., router, modem). This communication can be performed via input / output interface 650. Furthermore, electronic device 600 can also communicate with one or more networks (e.g., local area network, wide area network, and / or public network, such as the Internet) via network adapter 660. Network adapter 660 can communicate with other modules of electronic device 600 via bus 630. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage platforms.

[0092] Example 2 This invention also provides a storage medium, specifically a computer-readable storage medium, which is a memory device in a terminal device for storing programs and data. It is understood that the computer-readable storage medium here can include both built-in storage media in the terminal device and extended storage media supported by the terminal device; it can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor, which can be one or more computer programs (including program code). More specific examples of the computer-readable storage medium include: an electrical connection with one or more wires, a portable disk, a hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical fiber, portable compact disk read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.

[0093] Computer-readable storage media also include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable storage medium can also be any readable medium other than a readable storage medium that can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium can be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, radio frequency, etc., or any suitable combination thereof.

[0094] Program code for performing the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0095] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the adaptive dual-path UAV detection method for foggy environments in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by the processor in the following steps: The system receives an input image and calculates its global fog density coefficient. It then compares the global fog density coefficient with a preset threshold to perform adaptive path decision-making. If the global fog density coefficient is greater than or equal to the preset threshold, it uses an ASC image enhancement network optimized for detection tasks to perform defogging enhancement on the input image, obtaining an enhanced image. If the global fog density coefficient is less than the preset threshold, it inputs either the input image or the enhanced image into an improved YOLO P2 target detection network for UAV target detection, outputting the detection result, which includes the UAV target's bounding box and category confidence.

[0096] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0097] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0098] Example 3 An adaptive dual-path UAV detection method for foggy environments, the specific steps of which are as follows: (1) Fog Concentration Detection: The system receives real-time video streams from surveillance cameras and processes them frame by frame. For the current frame image, a lightweight fog concentration detection module (e.g., a fast statistical method based on dark channel priors) is invoked to calculate a coefficient characterizing the global fog concentration. .

[0099] (2) Adaptive judgment: The calculated fog concentration coefficient With the preset threshold (In this embodiment) The threshold was compared with that of standard datasets containing foggy days of varying concentrations. This threshold was determined through extensive testing on these datasets, weighing the accuracy improvement (mAP) and introduced latency (FPS) from defogging. Experiments show that when... In this case, the positive benefits of defogging to the final detection accuracy significantly outweigh its computational cost.

[0100] (3) Path selection and processing: like The image was determined to be moderate to heavy fog. The system automatically selected a dehazing enhancement path, inputting the current frame image into a pre-trained ASC dehazing network to obtain a clear image with enhanced texture and contrast. .

[0101] like The system determines the weather to be either light fog or clear. It automatically selects a fast detection path, skipping the fog removal step and directly displaying the original image. As the image to be detected.

[0102] (4) Target detection: The images output from the above paths are uniformly input into a pre-trained improved YOLO P2 target detection network. The network outputs the bounding boxes of all suspected UAV targets in the images and their corresponding class confidence scores.

[0103] Among them, the ASC image enhancement network provides clearer images that are more favorable for detection, while the YOLO P2 object detection network enhances the ability to capture features of small targets. Working together within an adaptive framework, the two networks improve the average detection precision from 0.8880 to 0.9225 and the recall from 0.9320 to 0.9760 on a heavily foggy test set, significantly reducing the risk of missed detections.

[0104] Based on fog concentration threshold The dynamic decision-making mechanism enables the system to skip the computationally intensive defogging steps under light fog conditions, achieving the best balance between accuracy and efficiency under complex and variable weather conditions.

[0105] The ASC image enhancement network employs an edge prior and perceptual loss design, ensuring that the restored image, while improving target-background contrast, can maintain the target's edge structure and realistic texture details to the greatest extent, avoiding color distortion and providing high-quality input for high-precision detection from the source.

[0106] The effectiveness and inventiveness of the method of this invention were verified through ablation experiments. The following schemes were compared on a self-built foggy drone dataset: Option A (benchmark): AOD-Net dehazing + standard YOLOX detection, mAP is 0.8880.

[0107] Solution B (Dehazing Improvement Only): ASC dehazing + standard YOLOX detection, mAP improved to 0.9047, indicating that the ASC network provides image quality that is more conducive to detection.

[0108] Solution C (Detection Improvement Only): AOD-Net Dehazing + YOLOP2 Detection, mAP improved to 0.9101, indicating that the P2 layer directly improves the detection capability of small targets.

[0109] Solution D (this invention): ASC defogging + YOLOP2 detection, mAP reaches the highest of 0.9225.

[0110] The experimental data clearly show that: Improving either the defogging module (B) or the detection module (C) individually can lead to performance gains; however, this invention generates a positive synergistic effect through dual-path adaptive decision-making and collaborative optimization of the two modules, achieving optimal performance in detection accuracy (especially recall rate for small targets). At the same time, the low latency in light fog / sunny weather scenarios meets the requirements for real-time monitoring, fully demonstrating the technical ingenuity and practical value of this invention.

[0111] In summary, this invention presents an adaptive dual-path UAV detection method and system for foggy environments. By introducing a lightweight fog concentration rapid assessment and adaptive path decision-making mechanism based on dark channel priors, the system can intelligently judge weather conditions and only activate the computationally intensive defogging enhancement module when necessary (moderate to heavy fog). This significantly saves computational resources under light fog or clear weather conditions, ensuring the overall operational efficiency and real-time performance of the system, and achieving "on-demand allocation" and optimized utilization of computational resources. Secondly, the designed ASC image enhancement network innovatively integrates edge prior guidance and dual attention mechanisms, and is trained using a composite loss function with multi-objective joint optimization. This results in enhanced images that are not only visually clear and naturally colored, but more importantly, highly compatible with the requirements of target detection tasks in terms of texture and structure, improving the quality of input images from the source. Finally, the improved YOLO P2 target detection network, by introducing a P2 high-resolution feature layer and constructing an enhanced four-layer feature pyramid and a dedicated detection head, greatly enhances the feature capture capability for small UAV targets, effectively curbing missed detections caused by feature loss. Experiments show that the proposed solution significantly improves the mean detection accuracy (mAP) and recall to 0.9225 and 0.9760 respectively on a heavily foggy test set, which is a substantial improvement over the benchmark solution. This fully demonstrates the effectiveness and superiority of the proposed solution in achieving high-precision and high-efficiency UAV detection in complex foggy environments.

[0112] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0113] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0114] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0115] In the embodiments provided by this invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0116] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0117] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0118] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random-access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0119] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus, and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0120] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0121] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0122] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. An adaptive dual-path UAV detection method for foggy environments, characterized in that, Includes the following steps: S1. Receive the input image and calculate the global fog concentration coefficient of the input image; S2. Compare the global fog concentration coefficient with a preset threshold and perform adaptive path decision-making: if the global fog concentration coefficient is greater than or equal to the preset threshold, proceed to step S3; if the global fog concentration coefficient is less than the preset threshold, proceed to step S4. S3. The input image is dehazing and enhanced using an ASC image enhancement network optimized for detection tasks to obtain an enhanced image; S4. Input the input image or the enhanced image into the improved YOLO P2 target detection network to perform UAV target detection, and output the detection results, which include the location bounding box and category confidence of the UAV target.

2. The adaptive dual-path UAV detection method for foggy environments according to claim 1, characterized in that, In step S1, a lightweight statistical algorithm based on dark channel priors is used to calculate the global fog concentration coefficient. The specific process is as follows: Define the dark channel of the input image as: in, It is based on pixels The local area centered on y represents the pixel value in the red, green, and blue channels; The mean value of the dark channel is calculated to obtain the global fog concentration coefficient.

3. The adaptive dual-path UAV detection method for foggy environments according to claim 1, characterized in that, In step S2, the preset threshold is 0.

52.

4. The adaptive dual-path UAV detection method for foggy environments according to claim 1, characterized in that, In step S3, the ASC image enhancement network adopts an encoder-decoder structure, and its input processing is as follows: Original foggy image use Operator extracts edge feature map The original foggy image With the edge feature map The enhanced input is obtained by concatenating along the channel dimension. .

5. The adaptive dual-path UAV detection method for foggy environments according to claim 4, characterized in that, The ASC image enhancement network embeds a convolutional block attention module after the encoder key layer. The processing procedure of the convolutional block attention module is as follows: First, channel attention processing is performed on the intermediate feature map to generate channel weight vectors; Then, spatial attention processing is applied to the channel-weighted features to generate a spatial weight matrix; Final output features for: in, The characteristics are those of the channel after weighting. This is the spatial weight matrix.

6. The adaptive dual-path UAV detection method for foggy environments according to claim 4, characterized in that, The ASC image enhancement network is trained using a composite loss function, which is: in, For structural similarity loss, To rebuild the losses, In order to perceive loss, , , These are the weighting coefficients.

7. The adaptive dual-path UAV detection method for foggy environments according to claim 1, characterized in that, In step S4, the improved YOLO P2 target detection network uses the YOLOX-S model as the baseline and extracts shallow high-resolution features with a downsampling step size of 4 in Stage 2 of the CSPDarknet backbone network. After processing by the transition layer, the P2 feature layer is formed. At the same time, the P3 feature layer with a downsampling step size of 8, the P4 feature layer with a downsampling step size of 16, and the P5 feature layer with a downsampling step size of 32 are extracted to construct a four-layer feature pyramid containing P2, P3, P4, and P5.

8. The adaptive dual-path UAV detection method for foggy environments according to claim 7, characterized in that, The four-layer feature pyramid performs deep feature fusion through the top-down path of the feature pyramid network and the bottom-up path of the path aggregation network. The fusion process satisfies the following: in, For convolution operations, The output features of Stage 5 of the CSPDarknet backbone network are shown. The output features of Stage 4 of the CSPDarknet backbone network are shown. The output features of Stage 3 of the CSPDarknet backbone network. The output features of Stage 2 of the CSPDarknet backbone network are shown. To perform upsampling operations on the P5 feature layer. This is to perform upsampling operations on the P4 feature layer. This is to perform upsampling operations on the P3 feature layer.

9. The adaptive dual-path UAV detection method for foggy environments according to claim 7, characterized in that, The improved YOLO P2 target detection network has four detection heads corresponding to the four-layer feature pyramid: P2, P3, P4, and P5. The P2 detection head is used to detect UAV targets with a pixel area of ​​10×10 to 32×32, the P3 detection head is used to detect small UAV targets, the P4 detection head is used to detect medium-sized UAV targets, and the P5 detection head is used to detect large UAV targets.

10. An adaptive dual-path UAV detection system for foggy environments, characterized in that, include: A fog concentration calculation module is used to receive an input image and calculate the global fog concentration coefficient of the input image; An adaptive decision-making module is used to compare the global fog concentration coefficient with a preset threshold and select a processing path based on the comparison result. The image enhancement module is used to perform dehazing enhancement processing on the input image using an ASC image enhancement network when the adaptive decision module determines that the global fog concentration coefficient is greater than or equal to the preset threshold, so as to obtain an enhanced image. The target detection module is used to input the input image or the enhanced image into the improved YOLO P2 target detection network to perform UAV target detection, and output the detection result including the bounding box of the UAV target location and the category confidence.

Images