基于深度学习的变电站设备油液渗漏检测识别

Object Detection of Oil Leakage in Substation Equipment Based on Deep Learning

为解决变电站中变压器、互感器等充油设备油液渗漏的机器人智能识别定位的问题,提出一种基于深度学习的变电站设备油液漏油可见光图像信息检测方法。首先建立包含3528张变电站充油设备油液渗漏的图像数据库,并对图像进行人工标注;然后使用Mobilenet-SSD深度网络模型对数据库中油液渗漏的基本图谱特征进行有监督的学习训练;最后将训练所得的模型参数部署到Jetson TX2边缘计算设备上,并搭载于变电站巡检机器人中。训练所得模型参数召回率大于85%,在嵌入式设备上的处理速度达到4帧/s,在满足准确性的前提下实现了油液渗漏识别定位的实时边缘计算,具有较高的实用性,为深度学习在变电站智能巡检中的应用提供了新的思路。

To solve the problem of robotic intelligent identification and localization of oil leakage from oil-filling equipment such as transformers and transformers in the substations,this paper proposes a deep learning based method for detecting oil leakage from visible image information of the substation equipment.Firstly,the image database of oil leakage of substation oil-filling equipment including 3528 images was established and manually labeled,and then the Mobilenet-SSD deep network model was used to perform supervised learning training on the basic mapping features of oil leakage in the database.Finally,the trained model parameters were deployed to the Jetson TX2 edge computing device and equipped in the substation inspection robot.The results show that the recall rate of the trained model parameters is more than 85%,and the processing speed of the embedded device reaches 4 frames per second,which realizes real-time edge computing for oil leakage identification and localization on the premise of meeting the accuracy,and provides new ideas for the application of deep learning in substation intelligent inspection.