Fault Recognition Method for Isolating Switch Infrared Images Based on HOG Feature

In order to achieve fault recognition for isolating switch infrared images, this paper utilizes an improved SLIC algorithm to segment and label the fault areas of the isolating switch based on color space conversion, effectively improving the accuracy of image segmentation. Leveraging the histogram of oriented gradient (HOG) feature extraction and support vector machine (SVM) algorithm, the paper classifies and categorizes the infrared images of isolating switches to distinguish whether they are operating in normal states. For isolating switches in normal states, the relative temperature difference method is employed to judge their fault states, where a higher relative temperature difference indicates a more severe fault. Experimental results demonstrate that under optimized HOG feature parameters, the highest accuracy in image device recognition can be achieved. By diagnosing faults using infrared images, the severity of defects in isolating switches can be determined, and maintenance recommendations can be provided. The proposed model exhibits high accuracy and reliability.