Insight on the Application of Deep Learning-Based Thermal Image Processing Methods in Electrical System Anomaly Detection and their Comparative Analysis

Ensuring the reliability and safety of electrical systems necessitates constant inspections. However, manual inspections pose risks, are time-consuming, and are impractical for real-time monitoring. This paper presents a novel, noninvasive, and efficient approach for automated electrical system anomaly detection using deep learning and thermal image processing. We have proposed a Convolutional Neural Network (CNN) based framework utilizing the well-established GoogLeNet and other deep learning-based architecture to classify thermal images of electrical systems as "normal" or "abnormal." This framework achieves a high accuracy of 99% in anomaly detection, surpassing traditional methods and paving the way for real-time monitoring and early fault identification.