基于改进区域全卷积网络的高压引线接头红外图像特征分析的在线故障诊断方法

With the continuous development of smart grid construction, massive infrared images have increased dramatically, while traditional infrared fault detection relies on manual inspection or manual extraction of features, low detection efficiency and great dependence on personnel experience. In order to realize the efficient and intelligent detection of infrared images and ensure the safe operation of the grid, this paper constructs an online fault diagnosis system based on infrared feature analysis, and proposes to improve the recognition performance of small targets by improving the feature extraction network of high-voltage lead connectors infrared images. Then, the region-based fully convolutional networks (R-FCN) is used to identify the location and operational status of the faulty area, and the operating state of the faulty area is secondarily diagnosed using Open CV to further reduce the false alarm rate. Finally, through testing and analysis, the average accuracy of the improved R-FCN network for high-voltage lead connectors infrared image fault diagnosis reached 80.76%, which is 8.43% higher than the original R-FCN network.