基于改进Faster R-CNN的绝缘子缺陷检测识别与定位

Insulator Defect Detection,Identification and Location Based on Modified Faster R-CNN

针对现有算法对绝缘子检测精度不高的问题,在Faster R-CNN算法的基础上进行改进,利用检测效果更好、性能更优的ResNet50代替原始VGG网络进行缺陷识别。实验结果表明,改进算法在数据集上的mAP达到77.29%,召回率达到87.55%,与其他经典算法相比具有更好的准确性与较强的实时性。

In view of the problem of unsatisfactory accuracy of existing algorithm in insulator detection,the present work made an improvement by using Faster R-CNN algorithm,and introduced Resnet50,which has better detection utility and excellent performance,to replace the original VGG network for defect identification.The improved algorithm,according to experimental results,can achieve a mAP reaching 77.29%on the data set in this paper,and a recall rate of 87.55%,exhibiting better accuracy and stronger real-time performance compared with other typical algorithms.