Low-altitude unmanned aerial vehicle target detection method and device based on improved YOLOv8n
By improving the YOLOv8n method, using the Gold-YOLO model and multi-scale attention mechanism, the information attenuation problem of YOLOv8n in complex scenarios is solved, the accuracy and efficiency of drone target detection is improved, and the model resource consumption is optimized.
Patent Information
- Application Number
- CN202411282802.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-13
- Publication Date
- 2024-11-12
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing YOLOv8n algorithm has information attenuation problems in drone target detection, resulting in slow processing speed and low recognition accuracy in complex scenarios, and the model size is too large, which limits the effectiveness of the drone monitoring system.
Using the improved YOLOv8n method, the feature information fusion process is performed through the Gold-YOLO model, and a parallel multi-scale subnet is set up to build a parallel multi-scale subnet, capture the short-range and long-range dependencies, and adjust the competition between the mass anchor boxes through intelligent gradient gain allocation strategy. The degree, combined with the efficient multi-scale attention mechanism, maintain the integrity of information between channels and reduce the computing burden.
It significantly improves the model's perception range and detail processing capabilities, improves the accuracy and recall rate, and improves the average accuracy by 2.7%, while reducing the number of model parameters and calculation complexity, optimizing detection performance and resource consumption.
Smart Images

Figure CN118941997A_ABST
Abstract
Description
Technical Field
[0001] The invention discloses a low-altitude unmanned aerial vehicle target detection method and device based on improved YOLOv8n, and relates to the technical field of image data processing. Background Art
[0002] With the rapid advancement of drone technology, its applications in areas such as aerial photography, precision agriculture, emergency services and cargo transportation have become increasingly widespread, greatly promoting efficiency and innovation. However, the popularity of drones has also brought many security concerns. In view of this, it has become crucial to develop an efficient and real-time drone monitoring system to maintain social order and protect the rights of citizens.
[0003] Deep learning has become a core tool in the technical solution of real-time monitoring of low-altitude drones. Deep learning relies on multi-layer artificial neural networks, automatically extracts features through learning from massive data, and can then analyze complex visual scenes. As a leader in modern target detection technology, the YOLOv8n algorithm has shown great potential in the field of drone monitoring with its excellent image analysis and positioning capabilities.
[0004] There is an essential difference between traditional and modern methods of visual object detection. Traditional methods rely on hand-crafted features and sliding window strategies to pre-select areas that may contain objects and then use simpler classifiers for discrimination. In contrast, deep learning-based methods implement an end-to-end automated detection process, which can be divided into two main categories: one is two-stage detectors, such as the R-CNN series, which first generate candidate regions and then classify and fine-tune the position of each region; the other is single-stage detectors, such as the YOLO series, which directly predict the target category and position from the input image without the additional candidate region generation step, so the processing speed is faster and more suitable for real-time monitoring scenarios. Both methods have their own advantages, but the latter is more popular in drone monitoring due to its real-time performance and efficiency.
[0005] In YOLOv8n, although the PANet structure (Path Aggregation Network) enhances the cross-layer fusion of feature information by adding a bottom-up path based on the Feature Pyramid Network (FPN), this recursive information transmission method still has limitations, because the information between layers that are not directly connected needs to be conveyed through the intermediate layers, which may cause information attenuation. Summary of the invention
[0006] In view of the problems of the prior art, the present invention provides a low-altitude UAV target detection method and device based on improved YOLOv8n, and the technical solution adopted is: In the first aspect, a low-altitude UAV target detection method based on improved YOLOv8n is provided, the method comprising: S1, obtaining a set of low-altitude drone images as initial data, and labeling the initial data by category through Roboflow; S2, performing feature information fusion processing through the Gold-YOLO model according to the category labeling; S3, by setting up a spatial learning framework and constructing parallel multi-scale sub-networks, captures short-range and long-range dependencies; S4, according to the intelligent gradient gain distribution strategy, is adjusted by the degree of competition between quality anchor boxes.
[0007] In some implementations, the S2 specifically includes: S21, set up a collection and distribution mechanism in the Gold-YOLO model, and use the comprehensive module to obtain feature data through various levels of the network; S22, redistribute the feature data evenly to each level.
[0008] In some implementations, the S3 specifically includes: S31, reconstructing the channel dimension into a batch dimension, and performing local interaction across channels through the batch dimension; S32, fusing the output feature maps of the two parallel multi-scale self-networks through a cross-space learning strategy.
[0009] In a second aspect, an embodiment of the present invention provides a low-altitude UAV target detection device based on improved YOLOv8n, the device comprising: A processing module, used to obtain a set of low-altitude drone images as initial data, and to perform category labeling on the initial data through Roboflow; A fusion module, used for performing feature information fusion processing through a Gold-YOLO model according to the category annotation; The network module is used to capture short-range and long-range dependencies by setting up a spatial learning framework and building parallel multi-scale sub-networks; S4, according to the intelligent gradient gain distribution strategy, is adjusted by the degree of competition between quality anchor boxes.
[0010] In some implementations, the fusion module specifically includes: The feature unit is used to set up a collection and distribution mechanism in the Gold-YOLO model, using the comprehensive module to obtain feature data through various layers of the network; The allocating unit is used to evenly redistribute the feature data to each level.
[0011] In some implementations, the network module specifically includes: An interaction unit, used to reconstruct the channel dimension into a batch dimension, and perform local interaction across channels through the batch dimension; The strategy unit is used to fuse the output feature maps of the two parallel multi-scale self-networks through a cross-space learning strategy.
[0012] In a third aspect, an embodiment of the present invention provides an electronic device, comprising a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein when the one or more computer instructions are executed by the processor, the method described in the first aspect above is implemented.
[0013] In a fourth aspect, an embodiment of the present invention provides a computer storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it is used to implement the method described in the first aspect.
[0014] One or more embodiments of the present invention can bring at least the following beneficial effects: The improved YOLO model of the present invention is implemented through the Gold-YOLO model, which incorporates an efficient multi-scale attention mechanism to maintain the integrity of information between channels while reducing the computational burden. This mechanism is based on the concept of grouping, uses a spatial learning framework, and is constructed through parallel multi-scale sub-networks to capture short-range and long-range dependencies. In this process, some channel dimensions are reconstructed into batch dimensions, which can avoid dimensionality reduction in conventional convolution operations and ensure that local interactions across channels are achieved without sacrificing dimensions. Finally, through a cross-space learning strategy, the output feature maps of the two parallel sub-networks are fused, thereby enhancing the model's perception range and detail processing capabilities. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0016] Figure 1 It is a YOLOv8 network structure diagram provided by an embodiment of the present invention; Figure 2 It is a Gold-YOLO structure diagram provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0017] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.
[0018] First, if Figure 1 The YOLOv8 network structure diagram shown in the figure explains the YOLOv8 involved in the present invention. As the latest member of the YOLO family, YOLOv8 represents the cutting-edge progress in the field of object detection. It inherits the advantages of previous YOLO versions, and innovates and optimizes them to provide better performance and higher operational flexibility. YOLOv8 is capable of a variety of computer vision tasks, including but not limited to target detection, image classification, and instance segmentation, and this article focuses on the application of target detection.
[0019] The YOLOv8 series includes five variants of different sizes: YOLOv8n, YOLOv8s, YOLOv8m, YOLOv8l and YOLOv8x. The number of parameters and computational complexity of these models increase in turn. Considering the dual requirements of application scenarios for real-time performance and accuracy, YOLOv8n was selected as the basic architecture for improvement because it has a small model size, easy deployment and high recognition accuracy, and is especially suitable for tracking fast-moving and low-altitude drone targets with blurred features.
[0020] The architecture of the YOLOv8 model consists of four key parts: the input layer, the backbone network, the neck network, and the head. Each part works together to achieve efficient and accurate object detection.
[0021] The input part uses mosaic data enhancement technology (Mosaic), which generates more diverse training samples by randomly scaling, cropping and reorganizing four images, thereby enhancing the generalization ability of the model. In addition, YOLOv8 uses an anchor-free mechanism to directly predict the center point of the target, eliminating the reliance on preset anchor frames, which not only simplifies calculations, but also improves detection speed and accuracy.
[0022] Back-bone is the core of feature extraction. The backbone network integrates multiple modules, including convolutional layer (Conv), cross-stage local layer (C2f) and fast spatial pyramid pooling (SPPF). The convolution module performs basic feature extraction, while the C2f module draws on the design ideas of C3 in YOLOv5 and achieves lightweight model and performance improvement by adding gradient propagation paths. The SPPF module can convert feature maps of different scales into feature vectors of uniform size, enhancing the model's adaptability to multi-scale targets.
[0023] Neck focuses on feature fusion and adopts the PANet structure, which combines the feature pyramid network (FPN) and the path aggregation network (PAN). Through top-down feature fusion and bottom-up information transmission, it not only retains the high-level semantic information, but also strengthens the position sensitivity of the bottom layer, thus achieving full utilization of cross-scale features.
[0024] As the output layer of the model, the head makes the final prediction based on the aforementioned multi-scale feature maps, including the category and location information of the object, ensuring that the model can maintain good detection effects on objects of different sizes.
[0025] When dealing with such targets in complex scenarios, existing detection algorithms face the challenges of slow processing speed, low recognition accuracy, and large model size, which limit the effectiveness of drone monitoring systems. In order to overcome these limitations, this embodiment provides a low-altitude drone target detection method and device based on improved YOLOv8n, and the specific implementation methods are as follows.
[0026] Embodiment 1:
[0027] This embodiment provides a low-altitude UAV target detection method based on improved YOLOv8n, the method comprising: S1, obtaining a set of low-altitude drone images as initial data, and labeling the initial data by category through Roboflow; S2, performing feature information fusion processing through the Gold-YOLO model according to the category labeling; S3, by setting up a spatial learning framework and constructing parallel multi-scale sub-networks, captures short-range and long-range dependencies; S4, according to the intelligent gradient gain distribution strategy, is adjusted by the degree of competition between quality anchor boxes.
[0028] First, according to S1, a set of dedicated experimental datasets was constructed to verify the effectiveness of the improved algorithm. The initial data came from a public collection of low-altitude drone images. By implementing data augmentation techniques on some samples, a variety of adverse weather conditions, such as rainy and foggy days, as well as common noise interference, were simulated to enhance the robustness of the model. Image blurring was also incorporated, using Gaussian blur to simulate the loss of clarity that drone cameras may encounter in real environments. In addition, the number of small target images was increased through random splicing, cropping, and embedding techniques, ultimately forming a dataset containing 9,000 640×640 pixel images. All images were annotated using Roboflow software and divided into four categories: drones, infrared drones, undetected objects, and blurred drones.
[0029] The next step, according to S2, specifically includes: S21, set up a collection and distribution mechanism in the Gold-YOLO model, and use the comprehensive module to obtain feature data through various levels of the network; S22, redistribute the feature data evenly to each level.
[0030] In YOLOv8n, although the PANet structure (Path Aggregation Network) enhances the cross-layer fusion of feature information by adding a bottom-up path based on the Feature Pyramid Network (FPN), this recursive information transmission method still has limitations because the information between layers that are not directly connected needs to be conveyed through the intermediate layer, which may cause information attenuation. To solve this problem, Figure 2 As shown in the Gold-YOLO structure diagram, Gold-YOLO introduces an innovative collection and distribution mechanism to comprehensively optimize the collection and distribution of information. This mechanism actively absorbs features from all levels of the network through a comprehensive module and evenly redistributes them to each layer, which not only overcomes the shortcomings of traditional recursive fusion, but also strengthens the interaction of deep information while controlling the additional time cost.
[0031] Next, S3 specifically includes: S31, reconstructing the channel dimension into a batch dimension, and performing local interaction across channels through the batch dimension; S32, fusing the output feature maps of the two parallel multi-scale self-networks through a cross-space learning strategy.
[0032] Attention mechanisms are crucial to improving the focus and efficiency of models, especially when dealing with complex scenes. Therefore, Gold-YOLO incorporates an efficient multi-scale attention mechanism to maintain the integrity of inter-channel information while reducing the computational burden. This mechanism is based on the concept of grouping and proposes a novel spatial learning framework that is constructed through parallel multi-scale sub-networks to capture short-range and long-range dependencies. In this process, part of the channel dimension is reconstructed into a batch dimension, avoiding the dimensionality reduction in conventional convolution operations and ensuring that local interactions across channels are achieved without sacrificing dimensionality. Finally, through a cross-space learning strategy, the output feature maps of the two parallel sub-networks are fused to enhance the model's perception range and detail processing capabilities.
[0033] Next, according to S4, the attention mechanism is crucial to improve the model's focus and efficiency, especially when dealing with complex scenes. Therefore, Gold-YOLO incorporates an efficient multi-scale attention mechanism, which aims to maintain the integrity of inter-channel information while reducing the computational burden. This mechanism is based on the concept of grouping and proposes a novel spatial learning framework, which is constructed through parallel multi-scale sub-networks to capture short-range and long-range dependencies. In this process, part of the channel dimension is reconstructed into the batch dimension, avoiding the dimensionality reduction in conventional convolution operations and ensuring that local interactions across channels are achieved without sacrificing dimensionality. Finally, through a cross-space learning strategy, the output feature maps of the two parallel sub-networks are fused to enhance the model's perception range and detail processing capabilities.
[0034] A series of key indicators were used to evaluate the performance of the improved YOLOv8n network, including Average Precision (AP), Model Parameters (Params), Floating Point Operations per Second (FLOPS), Recall and Precision. Among them, maP50 and maP50:95 reflect the average accuracy of the model between 50% and 50%-95%, respectively, and Params and FLOPS represent the model complexity and computational complexity, respectively. The recall rate shows the ability of the model to correctly identify positive samples, while the precision rate reflects the proportion of samples predicted to be positive that are actually positive. These two indicators jointly evaluate the accuracy and reliability of the model in real-world applications. The precision rate and recall rate are:
[0035]
[0036] Among them, TP is the number of positive samples correctly identified; FP is the number of false positive negative samples; FN is the number of false positive positive samples.
[0037] Taking YOLOv8 as the benchmark algorithm, we sequentially and combined improvement measures to conduct a series of fusion experiments. The experimental results are shown in Table 1: Table 1 Ablation experiment results method Improve the trunk Improved neck Improved loss P% R% Map50% YOLOv8 92.4 78.6 85.7 A √ 91.7 78.2 85.7 B √ 92.2 79.8 87.6 C √ 91.9 79.2 85.9 D √ √ 92.5 79.6 87.4 Ours √ √ √ 92.7 80.3 88.4 The improved YOLOv8 network shows significant performance improvements compared to the baseline model. Specifically, the precision rate increased by 0.3%, the recall rate increased by 1.7%, and the average accuracy Map50 increased by 2.7%. At the same time, the improved model has reduced the number of parameters by 0.2M and the total floating point operations (FLOPS) by 0.2G, which means that while maintaining or improving the detection performance, the model's computational efficiency and resource consumption have been optimized.
[0038] Embodiment 2:
[0039] The embodiment of the present invention provides a low-altitude UAV target detection device based on improved YOLOv8n, the device comprising: A processing module, used to obtain a set of low-altitude drone images as initial data, and to perform category labeling on the initial data through Roboflow; A fusion module, used for performing feature information fusion processing through a Gold-YOLO model according to the category annotation; The network module is used to capture short-range and long-range dependencies by setting up a spatial learning framework and building parallel multi-scale sub-networks; S4, according to the intelligent gradient gain distribution strategy, is adjusted by the degree of competition between quality anchor boxes.
[0040] Furthermore, the fusion module specifically includes: The feature unit is used to set up a collection and distribution mechanism in the Gold-YOLO model, using the comprehensive module to obtain feature data through various layers of the network; The allocating unit is used to evenly redistribute the feature data to each level.
[0041] Furthermore, the network module specifically includes: An interaction unit, used to reconstruct the channel dimension into a batch dimension, and perform local interaction across channels through the batch dimension; The strategy unit is used to fuse the output feature maps of the two parallel multi-scale self-networks through a cross-space learning strategy.
[0042] Embodiment three:
[0043] This embodiment further provides an electronic device, including a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions implement the method of embodiment 1 when executed by the processor; In practical applications, the processor can be an application specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a controller, a microcontroller unit (MCU), a microprocessor or other electronic components to execute the methods in the above embodiments.
[0044] The method implemented in this embodiment is as shown in the content of the first embodiment.
[0045] Embodiment 4:
[0046] This embodiment further provides a computer storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by one or more processors, the method of the first embodiment is implemented; Among them, the computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.
[0047] The method implemented in this embodiment is as shown in the content of embodiment 1.
[0048] In the several embodiments provided in the embodiments of the present invention, it should be understood that the disclosed system and method can also be implemented in other ways. The system and method embodiments described above are merely illustrative.
[0049] It should be noted that, in this article, the terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. The terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a..." does not exclude the presence of other identical elements in the process, method, article or device including the element.
[0050] Although the embodiments disclosed in the present invention are as above, the above contents are only embodiments adopted for facilitating the understanding of the present invention and are not intended to limit the present invention. Any technician in the technical field to which the present invention belongs can make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in the present invention, but the patent protection scope of the present invention shall still be subject to the scope defined in the attached claims.
Claims
1. A low-altitude UAV target detection method based on improved YOLOv8n, characterized in that: The method comprises: S1, obtaining a set of low-altitude drone images as initial data, and labeling the initial data by category through Roboflow; S2, performing feature information fusion processing through the Gold-YOLO model according to the category labeling; S3, by setting up a spatial learning framework and constructing parallel multi-scale sub-networks, captures short-range and long-range dependencies; S4, according to the intelligent gradient gain distribution strategy, is adjusted by the degree of competition between quality anchor boxes.
2. The method according to claim 1, characterized in that The S2 specifically includes: S21, set up a collection and distribution mechanism in the Gold-YOLO model, and use the comprehensive module to obtain feature data through various levels of the network; S22, redistribute the feature data evenly to each level.
3. The method according to claim 1, characterized in that The S3 specifically includes: S31, reconstructing the channel dimension into a batch dimension, and performing local interaction across channels through the batch dimension; S32, fusing the output feature maps of the two parallel multi-scale self-networks through a cross-space learning strategy.
4. A low-altitude UAV target detection device based on improved YOLOv8n, characterized in that: The device comprises: A processing module, used to obtain a set of low-altitude drone images as initial data, and to perform category labeling on the initial data through Roboflow; A fusion module, used for performing feature information fusion processing through a Gold-YOLO model according to the category annotation; The network module is used to capture short-range and long-range dependencies by setting up a spatial learning framework and building parallel multi-scale sub-networks; S4, according to the intelligent gradient gain distribution strategy, is adjusted by the degree of competition between quality anchor boxes.
5. The device according to claim 4, characterized in that The fusion module specifically includes: The feature unit is used to set up a collection and distribution mechanism in the Gold-YOLO model, using the comprehensive module to obtain feature data through various layers of the network; The allocating unit is used to evenly redistribute the feature data to each level.
6. The device according to claim 4, characterized in that The network module specifically includes: An interaction unit, used to reconstruct the channel dimension into a batch dimension, and perform local interaction across channels through the batch dimension; The strategy unit is used to fuse the output feature maps of the two parallel multi-scale self-networks through a cross-space learning strategy.
7. An electronic device, characterized in that: It includes a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein when the one or more computer instructions are executed by the processor, a low-altitude UAV target detection method based on improved YOLOv8n as described in any one of claims 1 to 3 is implemented.
8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, is used to implement the low-altitude UAV target detection method based on the improved YOLOv8n as described in any one of claims 1 to 3 above.
Citation Information
Patent Citations
Unmanned aerial vehicle-oriented power transmission line lightning stroke point and bolt defect detection method and system
CN118014968A
Unmanned aerial vehicle aerial photography small target detection method based on YOLOv7 neural network
CN118015496A
YOLOv8 lightweight fire identification method based on smoke enhancement algorithm
CN118379625A
Small target detection method for images acquired by unmanned aerial vehicle based on improved YOLOv8 algorithm
CN118628939A
