基于迁移学习和R-FCN的电力设备红外图像识别算法

Recognition algorithm for infrared image of power equipment based on transfer learning and R-FCN

电力设备类型的准确识别是实现红外图像智能诊断的前提。针对电力设备红外图像数量多、识别任务重等问题,采用迁移学习和基于区域的全卷积网络(R-FCN)算法实现电力设备红外图像的智能识别。首先,利用Labelimg软件制作电力设备红外图像数据集;然后,基于迁移学习的原理,选择在VOC数据集上表现优秀的识别算法R-FCN;最后,将R-FCN检测模型分别与2种不同大小的卷积神经网络结合,并利用在线难例挖掘(OHEM)改进算法。实验结果表明:在相同样本条件下,以上提出的算法能够快速有效地识别电力设备红外图像,精度达到0.9143,具有较好的准确性和鲁棒性。该算法为电力设备红外图像热故障诊断提供诊断基础。

Accurate identification of power equipment is a prerequisite for intelligent diagnosis of infrared images.Aiming at the problems of large number of infrared images of power equipment and heavy recognition tasks,intelligent recognition of infrared image of power equipment is realized by transfer learning and region-based fully convolutional networks(R-FCN)algorithm.Firstly,the infrared image dataset of power equipment is made by Labelimg software.Then,based on the principle of transfer learning,the identification model R-FCN which performs excellent on VOC dataset is selected.Finally,the R-FCN detection model is combined with two different convolutional neural network models separately to improve the algorithm,using online example mining(OHEM)method at the same time.The experimental results show that under the same sample conditions,the algorithm can quickly and effectively identify the infrared image of power equipment,and the precision reaches 0.9143,which has better performance on accuracy and robustness.The algorithm provides a basis for thermal fault diagnosis of power equipment based on infrared images.