基于R-FCN的航拍巡检图像目标检测方法

航拍巡检是输电线路巡检的主要方式之一，目前的航拍巡检方式效率较低，受巡检员主观因素影响大，亟需一种智能检测算法自动定位并识别输电线路巡检图片中的故障。基于深度学习的航拍巡检图像目标检测技术作为一种可能的解决方案，得到了广泛关注。提出了一种利用基于区域的全卷积网络(R-FCN)的航拍巡检图像目标检测方法，并利用在线困难样本挖掘(OHEM)、样本优化、软性非极大值抑制(Soft-NMS)等改进方法进行优化。实验证明，所提方法具有目标定位准确、平均准确率高、单模型可同时检测目标种类多等特点。

Object Detection Method for Aerial Inspection Image Based on Region-based Fully Convolutional Network

Aerial inspection is one of the main methods of transmission line inspection. In consideration of inefficient of aerial inspection mode and subjective factors of inspectors, there is an urgent need for intelligent detection algorithm to locate and identify the faults in inspection pictures of transmission line. As a possible solution, object detection technology of aerial inspection imagge based on deep learning has attracted extensive attention. An object detection method of aerial inspection image utilizing region-based fully convolutional network(R-FCN)is proposed. Online hard example mining(OHEM), sample adjusting and soft non-maximum suppression(Soft-NMS)are adopted to improve the performance of the proposed algorithm. The experiment results show that the proposed method has obvious advantages on accurate target location, high average precision, and simultaneous detection of target species by single-model.