PCANet Based Digital Recognition for Electrical Equipment Infrared Images

In this paper, a digital recognition method for electrical equipment infrared images based on PCANet is proposed. The main purpose of this paper is to recognize the displayed digits which can recover the temperature matrix of the whole infrared image. We use the PCANet deep learning network to identify the printed digital quickly, and then reconstruct the temperature matrix of the thermal image. The PCANet architecture here includes two PAC stages and one output stage. We discuss the related parameters among the whole procedure, including the number of the PCA stages, the number of filters, and the block overlap ratio.Besides, we compare the proposed digital recognize method with the traditional HOG+SVM method. It can be found that the proposed one has a higher accuracy. We also define a criterion to evaluate the performance of the combined temperature value recovery. The criterion contains the recognized and groudtruth temperature range which can reflect the effect of error recognition. The experimental results show that the algorithm has high accuracy and robustness. With the recovered temperature matrix, a further fault analysis can be proceeded for electrical equipment infrared images. Furthermore, deep learning architectures can be chosen to get intelligent infrared image fault diagnosis for electrical equipment. (paper)