Superpixels Segmentation and Low-rank Matrix Recovery for Thermal Fault Diagnosis of Power Equipment

Abstract

Traditional thermal fault diagnosis methods usually focus on an entry image or square patches. These methods may fail to fully spatial structure information for thermal fault diagnosis of power equipment. To overcome this issue, in this paper, we propose a novel thermal fault diagnosis method. This method integrates superpixel segmentation (SS) and low-rank matrix recovery (LRMR) for diagnosis. Specifically, the proposed method has two main steps. First, an input infrared image is transformed by using a principal component analysis (PCA) algorithm, and a superpixel segmentation method is employed to the first principal component, to divide the infrared image into non-overlapping homogeneous superpixel regions. Then, the thermal fault region is detected by employing the low-rank matrix recovery (LRMR) in a superpixel-by-superpixel manner. Since we jointly excavate the spatial structure and thermal information by combining SS with LRMR, it is better than simply analyzing an entry image or square patches. Experimental results that the proposed diagnosis method has a better diagnostic performance than the current state-of-the-art detectors.