基于内卷算子的YOLOv5野生动物检测  

YOLOv5 Wildlife Detection Based on Involution Operator

作　　者：贺鹏飞[1] 王菲菲 孙彩惠 聂荣[3] 刘志航 HE Pengfei;WANG Feifei;SUN Caihui;NIE Rong;LIU Zhihang(School of Physics and Electronic Information,Yantai University,Yantai 264005;College of International Education,Ludong University,Yantai 264005;College of Intelligent Engineering,Zhengzhou University of Aeronautics,Zhengzhou 450000)

机构地区：[1]烟台大学物理与电子信息学院,烟台264005 [2]鲁东大学国际教育学院,烟台264005 [3]郑州航空工业管理学院智能工程学院,郑州450000

出　　处：《计算机与数字工程》2025年第3期701-707,共7页Computer & Digital Engineering

基　　金：烟台市2021年校地融合发展项目(编号:1521001-WL21JY01);2022年河南省科技攻关项目(编号:222102220048)资助。

摘　　要：野生动物是自然环境的重要组成部分,保护野生动物对人类发展具有重要意义。利用红外相机与深度学习算法监测野生动物,为生物保护提供了有效途径。论文设计了一种基于YOLOv5的红外野生动物图像检测算法。在YOLOv5的颈部网络部分引入了内卷算子与特征拼接操作。改进了颈部网络的原始concat拼接操作。根据特征的重要程度对不同特征层进行加权操作,为重要特征层赋予更高的权重,使网络更加关注关键信息。改进的算法相比原始算法在各个指标方面都有大幅度提升,为野生动物检测提供了更有效的方法。Wildlife is an important part of the natural environment,and its conservation is of great importance to human development.Nowadays,using infrared cameras with deep learning algorithms to monitor wildlife provides an effective way for biological conservation.In this paper,an infrared wildlife image detection algorithm based on YOLOv5 is designed.This paper introduces the involution operator with feature splicing operation in the neck network part of YOLOv5.The original concat splicing operation of the neck network is improved.Its weighting operation is applied to different feature layers according to the importance of the features,giving higher weights to the important feature layers and making the network focus more on the key information.The improved algorithm in this paper has substantial improvement in all metrics compared with the original algorithm,providing a more effective method for wildlife detection.

关 键 词：野生动物检测 YOLOv5 内卷算子 特征融合 特征层加权 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]