Anomaly detection in thermal images using deep neural networks

Infrared thermography has become an effective tool in electrical preventive maintenance program due to its high precision and the capability of performing non-contact diagnostic. Anomalies in a thermal image is typically detected by comparing the temperatures of the equipment with reference temperatures. Manual detection is time-consuming and unreliable, making it unable to meet the excessive demand for condition monitoring in industrial applications. In this paper, we propose an automatic method to detect thermal anomalies based on deep neural networks (DNNs). The DNN model is trained to learn the statistical regularities of normal thermal images, and anomalies are detected based on pixel-wise comparison between the learned reference temperatures and the actual temperatures. We test our method on a variety of electrical equipment and the experimental results demonstrated the effectiveness of the proposed method.