基于边缘计算与深度学习的输电设备异物检测方法

在输电设备上经常会出现各种异物，如鸟巢、塑料袋，如果不能及时发现并清理将会对输电系统造成很大的安全隐患。因此，及时对输电设备是否有异物进行检测非常必要。针对该问题，提出了一种基于边缘计算和深度学习的异物检测方法。该方法与现有利用无人机拍摄传回云端服务器计算方法不同，通过将检测计算下沉到边缘设备，使用Mobilenet加上优化后SSD的目标检测方法在边缘设备直接处理计算，将检测出异物的图像发回云端。该方法在CPU上的运行速度是基于VGG（目视图像生成器）的SSD方法的5倍左右，是Faster-RCNN的58倍左右；在模型大小上是基于VGG的SSD方法的2/9左右，是Faster-RCNN的2/49左右，精确度为89%；与直接将数据传回云端服务器再进行处理的方式相比，数据传输量减少约90%。该方法不仅满足实时性，还具有可靠的效果，基于该方法的系统已经得到实际部署。

Foreign Body Detection Method for Transmission Equipment Based on Edge Computing and Deep Learning

Various foreign bodies, such as bird's nests and plastic bags, often appear on transmission equipment. Failure to detect and clean them up in time will cause great potential safety hazards to the transmission system. Therefore, it is necessary to timely detect the presence of foreign bodies on transmission equipment. To solve this problem, a foreign body detection method is proposed based on edge computing and deep learning. Different from the existing method that sends UAV pictures back to the cloud server for processing, this method, by sinking the detection calculation to the edge device, uses the target detection method of Mobilenet and optimized SSD to directly make process calculation in the edge device, and sends the pictures of detected foreign bodies back to the cloud server. The proposed method is about 5 times faster than the VGG-based SSD method and 58 times faster than the Faster-RCNN method in CPU running speed, and 2/9 times of the VGG-based SSD method and 2/29 times of the Faster-RCNN method in model size, with an accuracy of 89%. Compared with the method that sends original data back to the cloud server for processing, the proposed method can reduce the data transmission amount by about 90%. It is concluded that the proposed method can reliably detect foreign bodies on transmission equipment in real time. The detecting system based on this method has been deployed in practice.