摘要
针对接触网绝缘子破损识别,传统的特征匹配和神经网络分类识别率较低,同时因其需要人工提取和训练等问题,识别速率也较慢。相比传统卷积神经网络(CNN),胶囊网络(CapsNet)首次采用矢量作为输入,可以很好的保留目标的方向,角度等特征信息,更适合于识别复杂背景下的绝缘子。因此提出一种基于改进胶囊网络和CV模型结合的绝缘子破损识别算法,通过1×1归约层和3×3卷积层简化传统9×9胶囊网络的卷积层,并采用优化算法进行参数寻优,缩短训练权重时间,同时输出量保留方向角度,能更准确对棒形、针式和蝶式绝缘子破损情况进行分类。最后与AlexNet、YOLO、局部特征分析等方法进行了对比。通过对绝缘子巡检图像应用本文方法可得,绝缘子识别率提高到95%,实时速率达到32帧/s,所提出的绝缘子破损识别方法可以准确、迅速的从复杂背景识别出绝缘子,并准确的找到绝缘子破损的位置,大大提高了输电线路智能巡检的效率。
For the damage identification of contact network insulators,using the traditional feature matching and neural network classification methods to identify the damage,the recognition rate is low. At the same time,and the recognition speed is also slow because of the manual extraction and training. Compared with the traditional convolutional neural network( CNN),the capsule network( CapsNet) uses the vector as the input for the first time,which can well retain the target information such as direction and angle,and is more suitable for identifying insulators under complex backgrounds. Therefore,the author proposes an insulator damage identification algorithm based on the improved capsule network and CV model. The 1 × 1 reduction layer and 3 × 3 convolution layer are used to simplify the convolution layer of the traditional 9 × 9 capsule network,and the optimization algorithm is adopted. The parameters are optimized,the training weight time is shortened,and the output retains the direction angle,which can more accurately classify the damage of the rod,the pin and the butterfly insulators. Finally,the comparison has been made with AlexNet,YOLO,local feature analysis and other methods. By applying the method to the insulator inspection image,the insulator recognition rate is increased to 95%,and the real-time rate reaches 32 frames/s. This proposed insulator damage identification method can accurately and quickly identify the insulator from the complex background and precisely find the damage location,improving the efficiency of the intelligent inspection of the transmission line greatly.
作者
卞建鹏
郝嘉星
赵帅
滑伟静
高世闯
BIAN Jianpeng;HAO Jiaxing;ZHAO Shuai;HUA Weijing;GAO Shichuang(School of Electrical and Electronics Engineering,Shijiazhuang Tiedao University,Shijiazhuang 050043,China)
出处
《电瓷避雷器》
CAS
北大核心
2021年第1期194-200,共7页
Insulators and Surge Arresters
关键词
绝缘子破损识别
胶囊网络
CV模型
智能巡检
insulator damage identification
capsule network
CV model
intelligent inspection
