改进全卷积网络优化绝缘子缺陷检测  

Insulator Defect Detection Optimized by Improved Full Convolution Network

作　　者：杨永娇 唐亮亮 郑勇 YANG Yong-jiao;TANG Liang-liang;ZHENG Yong(Guangdong Electric Power Information Technology Co.,Ltd.,Guangdong Guangzhou510000,China;School of Mechanical and Electrical Engineering,Guangzhou University,Guangdong Guangzhou510000,China)

机构地区：[1]广东电力信息科技有限公司,广东广州510000 [2]广州大学机械与电气工程学院,广东广州510000

出　　处：《机械设计与制造》2021年第3期177-180,共4页Machinery Design & Manufacture

基　　金：国家社会科学基金项目(No.19BJY121);。

摘　　要：绝缘子各种缺陷的准确检测与定位是保障电网安全正常运行的关键,针对传统学度学习全卷积网络在进行航拍绝缘子图像缺陷识别时准确率不高问题,提出基于改进全卷积网络优化的航拍电网绝缘子缺陷自动检测算法,算法通过优化模型结构、剔除全连接层Dropout、增加多尺度池化与孔洞卷积以及采用双目标优化函数,实现FCN模型的有效改进,实验结果表明,改进全总卷只网络模型,有效提高了对绝缘子缺陷检测的性能和对背影的鲁棒性,取得了比已有算法更有优的检测结果。Accurate detection of grid insulation detection defects is the prerequisite for effective monitoring and fault diagnosis of grid operating conditions.Based on aerial grid insulator image of UAV,in order to solve the problem of false detection and local information loss in deep learning defect detection,an automatic defect detection algorithm based on improved deep learning full convolution network(FCN)is proposed.In the proposed algorithm,the FCN model is improved in the detection of insulator defects by improving the VGG structure,expanding the filter size,eliminating the full connection layer Dropout and the model depth,which realizes the effective improvement of the FCN model in the detection of insulator defects.The experimental results show that,the proposed model improves the performance of the insulator defect detection and the robustness under the less increase of running time,and obtains better detection results than the existing algorithms.

关 键 词：深度学习 电网绝缘子 缺陷检测 全连接网络模型 

分 类 号：TH16[机械工程—机械制造及自动化] TM726[电气工程—电力系统及自动化]