航拍巡线低照度彩色图像增强与电力小部件检测  

Low-illumination color image enhancement for aerial photography line patrol and detection of small power line components

作　　者：高晓娟 梅秀庄 白福忠[2,3] 李萍 GAO Xiaojuan;MEI Xiuzhuang;BAI Fuzhong;LI Ping(School of Astronautics,Inner Mongolia University of Technology,Hohhot 010051,China;Inner Mongolia Key Laboratory of Special Service Intelligent Robotics,Hohhot 010051 China;College of Mechanical Engineering,Inner Mongolia University of Technology,Hohhot 010051,China;Beijing Polytechnic College,Beijing 100042,China)

机构地区：[1]内蒙古工业大学航空学院,呼和浩特010051 [2]内蒙古自治区特殊服役智能机器人重点实验室,呼和浩特010051 [3]内蒙古工业大学机械工程学院,呼和浩特010051 [4]北京工业职业技术学院,北京100042

出　　处：《内蒙古工业大学学报（自然科学版）》2021年第6期461-467,共7页Journal of Inner Mongolia University of Technology：Natural Science Edition

基　　金：内蒙古自然科学基金项目(2020LH06002);内蒙古工业大学科学研究项目(X201703);北京工业职业技术学院科研重点项目(BGY2020KY-44Z)。

摘　　要：由于照明、逆光拍摄和复杂背景等因素导致航拍巡线图像经常出现低照度、照明不均匀、细节丢失等问题,严重影响目标检测可靠性,同时对于电力小部件及其缺陷识别影响尤为严重.基于航拍巡线图像特征,引入同态滤波算法增强亮度分量,保持色调和饱和度分量不变,从而实现彩色图像增强.进而利用YOLOv3算法对增强前后图像进行小部件(线夹)检测,识别率由71.1%提高至98.8%,识别精度指标也有较大提升.Due to factors such as lighting,backlighting,and complex backgrounds,aerial line patrol images often have problems such as low illumination,uneven lighting,and loss of details,which seriously affect the reliability of visual inspection,and at the same time,it has a particularly serious impact on the identification of small electrical components and their defects.Based on the characteristics of the aerial photography line patrol image,the homomorphic filtering algorithm is introduced to enhance the brightness component,while keeping the hue and saturation components unchanged,so as to achieve color image enhancement.Furthermore,by using the YOLOv3 algorithm to detect small parts(line clips)on the images before and after the enhancement,the recognition rate has been increased from 71.1%to 98.8%,and the recognition accuracy has also been greatly improved.

关 键 词：航拍巡线图像 低照度彩色图像 电力小部件检测 同态滤波 YOLOv3算法 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置] TP391.4[自动化与计算机技术—控制科学与工程]