Early warning of photovoltaic power plant fire under abnormal target recognition in binocular thermal imaging images

The large scale and complex environment of photovoltaic power plants result in significant noise in the collected thermal imaging images, making the accuracy of existing methods for early fire alarms unsatisfactory. In response to this, a method for early warning of photovoltaic power plant fires under abnormal target recognition in binocular thermal imaging images is proposed. Using binocular thermal imaging cameras, obtain a large number of thermal imaging images of photovoltaic power plants, and reconstruct and correct them. Extract color features, texture features, and shape features from the image to identify anomalous targets. Calculate alarm indicators, establish an alarm mechanism, and complete the initial alarm of photovoltaic power plant fires. The experimental results show that after the application of this method, the average IoU between the alarm area and the actual fire area reaches above 0.94, and the false alarm rate is less than 0.56%, which improves the alarm accuracy.