摘要
机场净空域内和跑道一边、五边范围内的无人机和鸟群严重威胁民航的安全,图像中的目标占比小,机场的背景复杂。现有机场净空域的雷达探测手段存在无法识别相距较近的目标类型与数量的问题,因此,提出了注意力特征融合结构并应用在YOLOv5s模型,通过注意力特征融合结构可融合更多的小目标信息至特征图,并使用更浅特征层检测目标,进一步提升原模型针对小目标检测的性能。在无人机和鸟群数据集上实现了93.5%的mAP50,相较于原模型的检测修正率为13.3%,处理视频的速度为28.1帧/s。提出的方法为机场现有监管手段提供了有力的技术支撑,与现有机场探测手段共同保障机场安全。
Unknown UAVs and birds flocking in the clear airspace of airports are a serious threat to the safety of civil aviation,especially on one and five sides of the runway.The percentage of targets in long-range captured images is small and the background of the airport is complex.The existing radar detection means of airport clear airspace has the limitation of not being able to identify the type and number of closely spaced objects.Therefore,an Attention-Path Aggregation Network(A-PAN)feature fusion structure is proposed and applied to the YOLOv5s model based on the disadvantage that object information is lost in the convolution process.The method first focuses on both the flux channel and spatial information of the feature map through the Convolutional Block Attention Module model,which further enhances the model’s ability to aggregate the information of small target birds and UAVs.Then,by combining the attention module with the PAN structure,the channel information and spatial information of different feature maps are fused to obtain a deeper feature map containing more information about small targets,and the target is detected using a shallower feature layer,further improving the performance of the original model for small target detection.A 93.5%mAP50 is achieved on the UAVs and birds dataset,with a detection correction rate of 13.3%compared to the original model,with a processing speed of 28.1 fps for the video.The method has good performance in detecting small UAVs and birds in complex airport environments,providing strong technical support for existing airport supervision means and safeguarding airport safety together with existing airport object detection tools.The results of this project can be applied to all civil airports in China and have a promising application.In addition,the results of this project can also be further applied to military airports to achieve the security of key airspace.
作者
刘闪亮
吴仁彪
屈景怡
李云龙
LIU Shanliang;WU Renbiao;QU Jingyi;LI Yunlong(Tianjin Key Lab for Advanced Signal Processing,Civil Aviation University of China,Tianjin 300300,China)
出处
《安全与环境学报》
CAS
CSCD
北大核心
2023年第8期2742-2749,共8页
Journal of Safety and Environment
基金
中央高校基本科研业务费项目(3122019185)
国家自然科学基金项目(U2133204)。
关键词
安全工程
机场安全
目标检测
注意力机制
特征融合
safety engineering
airport safety
object detection
attention mechanism
feature fusion