AI detection method for substation transformer temperature anomaly defects based on infrared image semantic segmentation

Due to the influence of temperature feature decomposition in substation transformers, defect detection accuracy is poor. Therefore, an AI-based detection method for abnormal temperature defects in substation transformers based on infrared image semantic segmentation is proposed. Integrating edge computing, an effective AI-based intelligent box for detecting abnormal temperature defects in substation transformers is designed. This intelligent box collects transformerrelated temperature defect signals and deploys lightweight defect detection instructions. Edge computing technology is used to analyze vibration signals, decompose wavelet packets, and construct feature vectors using the IEC three-ratio method. An infrared image semantic segmentation algorithm is designed. A single-stage regression-based partitioning backbone feature extraction network is used to expand detection channels. Reg confidence scores are obtained through quadratic convolution classification and regression prediction. Gain processing is performed to consider detection computation time and Sigmoid smoothing characteristics, completing AI-based detection of abnormal temperature defects in substation transformers. Experimental results show that the proposed method achieves consistently high defect detection accuracy, approaching 100%, at different learning rates, demonstrating robust detection.