基于改进YOLOv8的机场飞鸟实时目标检测方法

Real Time Object Detection Method for Airport Flying Birds Based on Improved YOLOv8

飞鸟对航空器的安全运行有着严重的威胁。常用探鸟方法中,人工探测精度较低,雷达探测成本较高。为解决上述问题,对中小型机场而言,提出基于改进YOLOv8的机场飞鸟实时检测方法。利用视频监控设备来检测飞鸟能以较低的成本实现高效的检测精度及速度。首先,设计特征融合C3(feature fusion C3,FFC3)模块,该模块在更细粒度的层次上实现多尺度特征融合,然后为模型选择合适的通道数,实现了检测速度和精度的平衡;其次,设计CSPPF(CBAM-spatial pyramid pooling fast)模块,在SPPF中引入CBAM(convolutional block attention module)模块,实现检测精度和速度的进一步提升;最后,发现了原AirBirds数据集的两点不足之处,对此改进了机场飞鸟数据集,同时利用了数据集增强技术。结果表明:改进YOLOv8的mAP@50达到0.820,相比原YOLOv8提升了0.015;改进YOLOv8的速度达到32帧/s。改进YOLOv8满足机场鸟类检测实时性和精确性的要求,为复杂环境下中小机场飞鸟检测提供了一种新思路。

Birds pose a serious threat to the safe operation of aircraft.Among commonly used bird detection methods,manual detection exhibits lower accuracy while radar detection incurs higher costs.In order to solve the above mentioned problems,for small and medium-sized airports,an airport bird detection method based on an improved YOLOv8 model was proposed.Employing video surveillance equipment was used to detect birds allows for efficient detection accuracy and speed at a lower cost.Firstly,a feature fusion C3(FFC3)module was designed to achieve multi-scale feature fusion at a finer level,followed by the selection of an appropriate channel number for the model to balance detection speed and accuracy.Secondly,a CBAM-spatial pyramid pooling fast(CSPPF)module was devised to enhance mAP@50 and detection speed by introducing a convolutional block attention module(CBAM)to SPPF module.Finally,two shortcomings in the original AirBirds dataset were identified and rectified,and the data augmentation technology was utilized.The results indicate that the improved YOLOv8 achieves a mAP@50 of 0.820,marking an improvement of 0.015 compared to the original YOLOv8,with a processing speed of 32 frames per second.The enhanced YOLOv8 meets the requirements of real-time and accurate bird detection at airports,offering a novel approach for bird detection in complex airport environments.