基于改进Cascade R-CNN网络模型的防振锤缺陷识别  

Defect recognition of vibration dampers based on improved Cascade R-CNN network model

作　　者：程汪刘 任仰勋 倪修峰 曹成功 张可 CHENG Wangliu;REN Yangxun;NI Xiufeng;CAO Chenggong;ZHANG Ke(State Grid Tongling Electric Power Supply Company,Tongling 244099,China;School of Electronics and Information Engineering,Anhui University,Hefei 230601,China;Research and Development Center,Anhui Nari Jiyuan Power Grid Technology Co.,Ltd.,Hefei 230088,China)

机构地区：[1]国网安徽省电力有限公司铜陵供电公司,安徽铜陵244099 [2]安徽大学电子信息工程学院,安徽合肥230601 [3]安徽南瑞继远电网技术有限公司研发中心,安徽合肥230088

出　　处：《安徽大学学报（自然科学版）》2022年第5期64-70,共7页Journal of Anhui University(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61672032)。

摘　　要：针对高压输电线路中防振锤的背景复杂、缺陷目标小及类别数量不均衡问题,提出一种改进的Cascade R-CNN(cascade region convolutional neural networks)网络模型,用于防振锤的缺陷识别.将SE(squeeze and excitation)模块嵌入ResNet-101(residual network-101),以增强网络学习能力.引入FPN(feature pyramid networks)模块提取多尺度的缺陷特征.利用Focal Loss函数降低Cascade R-CNN候选区域提取模块的分类损失.实验结果表明:相对于其他4种模型,该文模型有相对高的识别准确率;识别防振锤缺陷的效果良好.因此,该文模型具有有效性.Aiming at the problems of complex background,small defect targets and unbalanced number of classes of vibration dampers in high-voltage transmission lines,an improved Cascade R-CNN(cascade region convolutional neural networks)network model was proposed for defect recognition of vibration dampers.The SE(squeeze and excitation)module was embedded into ResNet-101(residual network-101)to improve the learning capability of the network.The FPN(feature pyramid networks)module was introduced to extract multi-scale defect features.The Focal Loss function was used to reduce the classification loss of Cascade R-CNN candidate area extraction module.The experimental results showed that compared with the other four models,the proposed model had the relatively high recognition accuracy and good effect on the recognition of vibration damper defects.Therefore,the model in this paper had validity.

关 键 词：电力巡检 深度学习 缺陷识别 防振锤 Cascade R-CNN 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]