基于深度学习的架空线路绝缘子掉串识别研究

Research on string breakage recognition of overhead line insulators based on deep learning

针对无人机巡检拍摄的架空线路绝缘子设备的照片进行处理,旨在建立一个基于Faster R-CNN目标检测算法的架空线路绝缘子设备识别与掉串诊断的模型。首先通过TensorFlow建立训练框架,将收集到的绝缘子数据集训练Faster R-CNN网络识别绝缘子,其次利用小波变换去噪增强图像特征信息,再对经过二值化处理的图像进行霍夫变换直线检测以及垂直投影确定有无缺陷。该模型绝缘子识别率为85.6%,掉串检测正确率为96%,有较强的鲁棒性。通过这样一个检测模型可以及时发现绝缘子设备存在的绝缘隐患,降低出现绝缘故障的风险,并且可以配合无人机巡检,大大减少人力劳动,更有效地分配人力资源及减少运维的成本。

In order to process the photos of the overhead line insulator equipment taken by the UVA routing inspection,a model of the overhead line insulator equipment recognition and string breakage diagnosis based on the Faster R-CNN object detection algorithm is established.The training framework is established by means of the TensorFlow,and the collected insulator data set is trained to identify the insulators by the Faster R-CNN network.Then the wavelet transform is used to denoise and enhance the image feature information,and then Hough transform line detection and vertical projection of the binarized images are conducted to determine whether there are defects.The insulator recognition rate of this model is 85.6%,and the correct rate of string breakage detection is 96%,which has strong robustness.By means of the detection model,it is possible to timely discover the insulation hidden dangers of the insulator equipments,reduce the risk of insulation failure,cooperate with the UVA routing inspection to greatly reduce human labor,allocate human resources and reduce the cost of operation and maintenance more effectively.