摘要
由于鸟类监测设备需在野外环境下工作,因此最好采用轻量级网络并兼具检测精度和实时性的特点。文中根据EfficientNet-lite轻量级网络,提出一种适用于野外高压输电线路上检测鸟类的实时检测网络,即轻量级目标检测网络(EFYOLO)。网络特征提取部分借鉴EfficientNet-lite轻量级模型,预测输出部分则使用YOLO算法,采用Ciou损失函数和Diounms非极大值抑制策略。实验结果表明:EF-YOLO检测精度达87.60%,平均检测速度为138 f/s,在检测速度方面,文中提出的EF-YOLO优于目前主要的三种网络模型;且模型权重大小为4.01 MB,适合在输电线路边缘检测平台上进行部署,辅助驱鸟器工作。
As the bird monitoring equipment has to work in the field environment,it is best to use a lightweight network that has the character of accurate detection and real-time performance. A real-time monitoring network(lightweight target detection network EF-YOLO) is proposed based on lightweight network EfficientNet-lite,which is suitable for bird detection on high-voltage transmission lines in the field environment. The lightweight model EfficientNet-lite is referred to the network feature extraction part. The YOLO algorithm is used at the prediction output part,and Ciou loss function and Diou-nms non-maximum suppression strategy are applied. The experimental results show that the detection accuracy of EF-YOLO can reach 87.60%,its average detection speed is 138 f/s and is better than that of three mainstream network models,and its model weight is 4.01 MB,which is suitable for deployment on the edge detection platform of the transmission line to assist the bird repellent in working.
作者
何俊
蒋昌辉
李倡洪
刘鹏
聂勇
HE Jun;JIANG Changhui;LI Changhong;LIU Peng;NIE Yong(College of Information Engineering,Nanchang University,Nanchang 330031,China)
出处
《现代电子技术》
2022年第10期94-98,共5页
Modern Electronics Technique
基金
国家自然科学基金项目(62066025)。
