基于改进Mask R-CNN强风沙环境绝缘子识别

Insulator Recognition for Transmission Lines in Strong Wind and Sand Environment Based on Improved Mask R-CNN

由于许多输电线路处于强风沙区域,其对绝缘子的破坏尤为严重,因此对输电线路绝缘子进行检测无疑是重中之重。笔者针对现阶段强风沙环境下基于深度学习的球机远程检测精确度不高的问题进行了研究。为提高强风沙环境下绝缘子检测的精确度,提出一种基于卷积神经网络的Ms-Net绝缘子目标识别算法。Ms-Net网络为提高目标检测精确度以及缩短训练时间,将Mask R-CNN的101层卷积层改进为56层,并且在特征提取的第2至第5阶段的每个残差块前引入基于注意力机制的挤压与激励网络(SENet)结构,SENet结构对绝缘子进行特征提取时可以增强有用的特征,抑制无用的特征,提高了算法鲁棒性。对比几种主流的检测算法,经实验验证,文中Ms-Net算法在强风沙环境下检测精确度与时间得到了改善。

Many transmission lines are located in strong wind and sand area,the damage to insulators is particularly serious,the detection of transmission line insulators is undoubtedly the most important.The problem of low accuracy of ball machine remote detection based on deep learning in strong wind and sand environment is studied.In order to improve the accuracy of insulator detection in strong wind and sand environment,an MS-Net insulator target recognition algorithm based on convolution neural network is proposed.In order to improve the accuracy of target detection and shorten the training time,the 101 layer convolution layer of Mask R-CNN is improved to 56 layers,and the attention mechanism based squeeze and excitation network(SENet)structure is introduced before each residual block in the second to fifth stages of feature extraction.The SENet structure can enhance the useful features and suppress the useless features.The robustness of the algorithm is improved.Compared with several mainstream detection algorithms,the detection accuracy and time of Ms-Net algorithm are improved under strong wind and sand environment.