Power System Equipment Detection in Thermal Images by Deep Learning Approach

The non-contact accuracy of infrared thermography has made it indispensable for electrical equipment inspection. In order to recognize electrical equipment in thermal pictures, this study presents a revolutionary neural network-based learning model. Based on the neural network algorithm, the proposed work suggests a real-time object detection model YOLO-V8, that can predict the coordinate points, orientation angle, and type of class of certain equipment. To improve accuracy, particularly for small items, the model incorporates an orientation consistency prior. In the proposed work, an extensive thermal imaging dataset of a distribution transformer; covered a range of settings and equipment kinds. According to experimental results, the suggested model achieves a mean average precision of 93.7%, demonstrating its robustness to noise and ability to function at 20 frames per second resulting in accurate detection of different segments of a distribution transformer in a thermal image.