基于YOLOv5的小动物目标检测算法研究

Research on Small Animal Target Detection Algorithm Based on YOLOv5

针对现有变电站入侵检测算法误报率高、对小目标检测精度低等问题,提出了一种基于改进YOLOv5的变电站入侵小动物目标检测算法。将SENet通道注意力模块和卷积注意力模块(CBAM)中的激活函数改进为HardSwish函数,并在主干网络和颈部网络中分别引入改进后的SENet_H模块和CBAM_H模块;采用空洞空间池化金字塔(ASPP)对空间金字塔池化进行优化,并在检测端增加一个小目标检测层,以提高对小动物的检测精度。此外,还构建了小动物数据集,并采用9-Mosaic数据增强方式,丰富了样本目标。实验结果表明:改进后的小动物目标检测算法相较于原YOLOv5算法精确率提升了11.6%,召回率提升了10.2%,平均精度均值提升了8.1%。

Aiming at the problems of high false alarm rate and low detection accuracy of small targets in existing substation intrusion detection algorithms,a substation intrusion small animal target detection algorithm based on improved YOLOv5 is proposed.The activation function in the SENet channel attention module and Convolutional Block Attention Module(CBAM) is improved to HardSwish function,and the improved SENet_H module and CBAM_H module is separately introduced into the backbone network and the neck network.Experimental results show that the improved small animal target detection algorithm outperforms the original YOLOv5 algorithm by 11.6% in precision,10.2% in recall,and 8.1% in mean average precision.