基于YOLOv5-PNCM的飞鸟目标检测算法研究

Research on target detection algorithm of flying birds based on YOLOv5-PNCM

针对小目标飞鸟检测存在的检测精度低、漏检率高等问题,提出了基于YOLOv5的小目标飞鸟的实时检测算法。首先,在YOLOv5原有的检测层上添加了一层小目标检测头;其次,采用CARAFE上采样算子改进了上采样方法,引用NWD度量代替IoU,有效降低了小目标位置偏差的敏感性;最后,使用M-CBAM注意力模块。改进后的算法在自制鸟类数据集上平均精度为77.3%,检测速度达到78FPS,与改进前相比,检测精度提升了9.1%,检测速度提升了23.8%。

Aiming at the problems of low detection accuracy and high leakage rate of small target bird detection,a real-time detection algorithm for small target bird detection based on YOLOv5 is proposed.Firstly,a layer of small target detection head is added to the original detection layer of YOLOv5,secondly,the CARAFE up-sampling operator is used to improve the up-sampling method;the NWD metric is quoted instead of the IoU,which effectively reduces the sensitivity of the positional deviation of small targets.Finally,the M-CBAM attention module is used.The improved algorithm achieves an average accuracy of 77.3% and a detection speed of 78 FPS on the homemade bird dataset,which is a 13.3% increase in detection accuracy and a 23.8% increase in detection speed compared to the pre-improvement period.