摘要
针对电力巡检对照片快速批量目标识别的业务需求,该文以YOLOv4为技术手段实现对电力巡检照片的目标检测过程。文中首先对电力巡检的训练数据集进行精确标注,后经过Darknet深度学习框架训练,试验检测达到了良好的效果。试验结果显示,该次试验检测的准确度为0.875,召回率为0.840。此目标检测效果满足部分电力巡检对图片目标检测的需求,但仍存在训练集图片数据量少以及图片中物体特征不显著等问题。
Aiming at the business needs of power inspection for rapid batch target recognition of photos,this paper uses YOLOv4 as a technical means to realize the target detection process of power inspection photos.In this paper,the training data set of power inspection is accurately labeled first,and then trained by the Darknet deep learning framework,the test and detection have achieved good results.The test results show that the accuracy of this test is 0.875,and the recall rate is 0.840.This target detection effect meets the needs of some power inspections for image target detection.However,there are still some problems,such as the small amount of image data in training set and the insignificant features of objects in the pictures.
作者
孙兴达
郝赫
刘远
赵园园
王一梦
SUN Xingda;HAO He;LIU Yuan;ZHAO Yuanyuan;WANG Yimeng(Beijing Guodiantong Network Technology Co.,Ltd.,Beijing 100085,China)
出处
《现代信息科技》
2020年第20期115-117,共3页
Modern Information Technology
