摘要
针对旋耕机器人作业过程中存在遮挡及遮挡识别率不高的问题,提出一种基于改进YOLOv5网络的旋耕机器人遮挡识别算法。首先以YOLOv5网络作为基础目标检测算法;然后对YOLOv5网络的颈部网络进行非邻层信息交流特征增强,并添加SENet注意力机制;最后将改进的YOLOv5网络用于旋耕机器人作业过程中的遮挡识别并进行实验验证。结果表明,所提基于改进YOLOv5网络的旋耕机器人遮挡识别算法的识别精确率、召回率和F1值分别取值为98.42%、96.17%和0.9753,均高于传统的YOLOv3算法、Faster R-CNN检测算法和SSD算法,且所提算法的遮挡识别时长仅为16.96 s,比另外3种算法分别降低了11.81 s、9.64 s和7.06 s。综合分析可知,所提算法能够提升旋耕机器人在作业过程中的遮挡识别精度和效率,对实现农业现代化和智能化具有一定应用价值。
To address the issue of occlusion and low occlusion recognition rates during the operation of rotary tiller robots,an occlusion recognition algorithm based on an improved YOLOv5 network is proposed.First,the YOLOv5 network is used as the base object detection algorithm.Then,non-adjacent layer information exchange feature enhancement is applied to the neck network of YOLOv5,and the SENet attention mechanism is incorporated.Finally,the improved YOLOv5 network is applied to occlusion recognition during the operation of the rotary tiller robot and experimentally validated.The results show that the proposed algorithm achieves a precision of 98.42%,a recall of 96.17%,and an F1-score of 0.9753,all higher than those of traditional YOLOv3,Faster R-CNN,and SSD algorithms.Moreover,the occlusion recognition time of the proposed algorithm is only 16.96 s,which is 11.81 s,9.64 s,and 7.06 s shorter than the other three algorithms,respectively.Comprehensive analysis indicates that the proposed algorithm improves the accuracy and efficiency of occlusion recognition during the operation of rotary tiller robots,demonstrating certain application value for advancing agricultural modernization and intelligence.
作者
曹国辉
程丛喜
Cao Guohui;Cheng Congxi(School of Management,Wuhan Polytechnic University,Hubei Wuhan,430048,China)
出处
《机械设计与制造工程》
2026年第1期63-68,共6页
Machine Design and Manufacturing Engineering
基金
武汉市社会科学界联合会2024年度课题(WHSKL2024051)。
