基于Faster R-CNN和U-net改进的混合模型绝缘子故障检测

Hybrid Model Insulator Fault Detection Based on Faster R-CNN and U-net

在航拍影像中定位绝缘子爆裂的位置是一项艰巨的任务。针对绝缘子和绝缘子爆破位置在图像中占比过小、背景复杂以及拍摄图像角度和大小不一等问题,采用一种注意力机制与Faster R-CNN和U-net相结合的绝缘子识别模型,对某电力科学研究院提供的绝缘子航拍图像进行测试和对比试验,结果表明,该模型识别绝缘子的平均精度(Average Precision,AP)为92.1%,识别绝缘子爆裂故障的平均精度(AP)为91.9%。所提出的绝缘子自爆故障检测模型在绝缘子定位、爆裂位置判定等应用方面的效果优于部分经典方法。

It is an arduous task to locate the position of the insulator burst in the aerial image.The insulator and the blasting position of the insulator are too small in the image,the background is complicated,and the angle and size of the captured image are different.The accuracy of the detection of a certain model is low,and the effect is not good.Therefore,this paper adopts a kind of insulator recognition model with attention mechanism,which is combined with Faster R-CNN and U-net.The model is used to test and compare the aerial images of insulators provided by a certain electric power research institute to verify the effect of the proposed method.The results show that the Average Precision(AP)of the model for identifying insulators is 92.1%,and for identifying insulator bursts the Average Precision(AP)is 91.9%.The proposed insulator self-explosion fault detection model can accurately and efficiently locate the insulator and identify its burst location.