基于深度学习的电力设备红外与可见光图像配准

针对现有电力设备红外与可见光图像配准难度大、配准时间长等问题，提出一种基于深度学习的电力设备红外与可见光图像配准融合的方法。本文将特征提取与特征匹配联合在深度学习框架中，直接学习图像块对与匹配标签之间的映射关系，用于后续的配准。此外为了缓解训练时红外图像样本不足的问题，提出一种利用红外图像及其变换图像学习映射函数的自学习方法，同时采用迁移学习来减少训练时间，加速网络框架。实验结果表明:本文方法与其他4种配准算法相比性能指标均有显著提升，本文平均准确率为89.909，同其余4种算法相比分别提高了2.31%、3.36%、2.67%、0.82%，本文平均RMSE（Root Mean Square Error）为2.521，同其余4种配准算法相比分别降低了14.68%、15.24%、4.90%、1.04%，算法平均用时为5.625 s，较其余4种算法分别降低了5.57%、6.82%、2.45%、1.75%，有效提高了电力设备红外与红外可见光图像配准的效率。

Infrared and Visible Image Registration for Power Equipments Based on Deep Learning

A registration fusion method of infrared and visible images of power equipment based on deep learning is proposed that aims at problems with difficult and long registration time of infrared and visible images of existing power equipment. In this study, feature extraction and feature matching are combined in a deep learning framework to directly learn the mapping relationship between image block pairs and matching labels for subsequent registration. In addition, a self-learning method using infrared image and its transform image to learn the mapping function is proposed to alleviate the problem of insufficient infrared image samples during training Simultaneously, transfer learning is used to reduce the training time and accelerate the network framework. The experimental results show that the performance index of this method is significantly improved compared with the other four registration algorithms. The average accuracy of this method is 89.909, which is 2.31%, 3.36%, 2.67%, and 0.82% higher than that of the other four algorithms, respectively. The average RMSE of this method is 2.521. Compared with the other four registration algorithms, the algorithm is reduced by 14.68%, 15.24%, 4.90%, and 1.04%, respectively. The average time of the algorithm is 5.625 s, which is reduced by 5.57%, 6.82%, 2.45%, and 1.75% respectively. The efficiency of infrared and visible image registration of the power equipment must be effectively improved.