Feature extraction and analysis of power equipment images using Grad-CAM heatmap

To accurately extract image features of power equipment, improve the diagnostic efficiency of power equipment faults, and ensure the stable operation of the power system, this study optimizes the convolutional neural network algorithm using gradient weighted class activation mapping heatmap. Based on the optimized algorithm, a method for extracting image features of power equipment has been developed. The experiment analyzed the feature extraction performance of the optimization algorithm on the MS COCO, ImageNet, LFW, and ORB datasets. The results demonstrated that the algorithm had an accuracy rate of over 96% in extracting different features from images in different datasets, and its feature extraction accuracy rate was also over 94% under the influence of interference. In the analysis of the actual effectiveness of optimization methods, the extraction error rate of features from images of different power equipment was less than 3%. Therefore, the proposed method for extracting image features of power equipment can accurately extract features from different power equipment images, thereby enabling accurate diagnosis of the operating status of power equipment and improving the stability of power system operation.