摘要
随着巡检机器人的应用,泵站、变电站、实验室对指针自动检测识别的需求增加。YOLOv3方法是使用最广泛的基于深度学习的对象检测方法之一。它使用K-means(K均值聚类算法)聚类方法来估计预测边界框的初始宽度和高度。使用该方法估计的宽度和高度对初始聚类中心敏感且耗时。为了解决这些问题,文章从YOLOv3电表检测算法出发,提出了解决特定电表检测过程中涉及的目标检测性能不足的问题,取得了改进原有算法主干网络大、参数多、计算量大的弊端的成果。
With the application of inspection robot,the demand for automatic detection and identification of pointer in pumping stations,substations and laboratories increases.YOLOv3(You Only Look Once v3 only needs one browse to identify the category and location of the objects in the graph) method is one of the most widely used deep learning-based object detection methods.It uses the K-means(K-mean clustering algorithm) clustering method to estimate the initial width and height of the predicted bounding box.Using this method,the estimated width and height are sensitive and time-consuming to the initial clustering center.In order to solve these problems,this paper starts from the YOLO 3 meter detection algorithm,and proposes to solve the problem of insufficient target detection performance involved in the specific meter detection process.The results of improving the original algorithm backbone network,many parameters and large calculation are obtained.
作者
安建平
刘晓群
An Jianping;Liu Xiaoqun(Hebei Institute of Civil Engineering and Architecture,Zhangjiakou 075000,China)
出处
《无线互联科技》
2023年第17期143-146,159,共5页
Wireless Internet Technology
关键词
目标检测
YOLOv3
仪表检测算法
性能改进
target detection
YOLOv3
instrument detection algorithm
performance improvement
