Application of Object Detection on Power Station Environmental Anomaly Monitoring

For the environmental abnormal conditions such as open fire, smoke and liquid leakage that may occur in the power station, the traditional recognition methods often study them separately, regarding them as different problems, and lack of unified and effective solutions. From the perspective of visual anomaly detection, we propose a universal anomaly detection technology based on deep learning object detection. We compare the precisions and speeds of several state-of-the-art object detection algorithms on four anomaly detection tasks: fire, smoke, water and oil leakage detection. The results on our anomaly database show that object detection can not only detect single anomaly accurately, but also monitor several anomalies at the same time, which shows the effectiveness and extensibility of our proposed method.