| 摘要: |
| 为克服机场飞鸟检测中人工观测准确率低、速度慢、雷达探测造价高的缺点，保障民航安全运行，采用深度学习目标检测算法实现对机场附近飞鸟的精确感知。为提高YOLOv8对重要特征的关注度，在颈部添加高效通道注意力(Efficient Channel Attention,ECA)，使网络在增加少量参数的情况下获得较明显的精度提升。提出多分支C3(Multi-branch C3,MBC3)模块，通过引入具有不同感受野的卷积分支结构以增强模块的表达能力。探究了不同网络宽度及深度对模型性能的影响并为模型选择合适的宽度与深度因子。为减少小飞鸟特征丢失问题，提出了浅层特征-路径聚合网络(Shallow Feature-Path Aggregation Network,SF-PAN)。在机场飞鸟数据集上测试，结果表明，改进YOLOv8的mAP@50达到82.9%，相比原YOLOv8提升了2.4%；其速度达到31 frame/s。改进YOLOv8满足机场飞鸟检测实时性和精确性的要求，为复杂环境下机场鸟类检测提供了一种新思路。 |
| 关键词: 机场飞鸟检测 鸟击防范 注意力机制 多分支卷积 特征融合 |
| DOI：10.20079/j.issn.1001-893x.240110002 |
|
| 基金项目: |
|
| An Airport Bird Detection Method Based on YOLOv8 Network |
| KONG Jianguo,ZHANG Xiangwei,ZHAO Zhiwei,LIANG Haijun |
| (College of Ar Traffic Management,Civil Aviation Flight University of China,Guanghan 618307,China) |
| Abstract: |
| To overcome the drawbacks of low accuracy and slow speed in manual bird detection at airports,as well as the high cost associated with radar detection,and ensure the safe operation of civil aviation,deep learning object detection algorithms are used to achieve accurate perception of birds near airports.To enhance the network搒 focus on crucial features,an Efficient Channel Attention(ECA) attention mechanism is incorporated into the Neck,resulting in a significant improvement in accuracy while adding a small number of parameters.The Muti-branch C3(MBC3) module is designed to strengthen the model搒 expressive capability by introducing branches with different receptive fields.The impact of different network widths and depths on model performance is explored,and appropriate width and depth factors for the model are selected.The Shallow Feature - Path Aggregation Network(SF-PAN) structure is proposed to address the issue of feature loss in detecting small bird targets.Testing on an airport bird dataset demonstrates that the modified YOLOv8 achieves a mAP@50 of 82.9%,showcasing a 2.4% improvement over the original YOLOv8,while maintaining a speed of 31 frames per second.The improved YOLOv8 meets the requirements for real-time and accurate detection of birds at airports and offeres a new approach to bird detection in complex airport environments. |
| Key words: airport bird detection bird strike prevention attention mechanism multi-branch convolution feature fusion |
