Two-Level Model for Detecting Substation Defects from Infrared Images

Abstract

Training a deep convolutional neural network (DCNN) to detect defects in substation equipment often requires many defect datasets. However, this dataset is not easily acquired, and the complex background of the infrared images makes defect detection even more difficult. To alleviate this issue, this article presents a two-level defect detection model (TDDM). First, to extract the target equipment in the image, an instance segmentation module is constructed by training from the instance segmentation dataset. Then, the target equipment is segmented by the superpixel segmentation algorithm into superpixels according to obtain more details information. Next, a temperature probability density distribution is constructed with the superpixels, and the defect determination strategy is used to recognize the defect. Finally, experiments verify the effectiveness of the TDDM according to the defect detection dataset.