Straw Burning Detection Method Based on Improved Frame Difference Method and Deep Learning

Straw burning has serious pollution to the air. Only by finding the location of straw burning can we stop the pollution caused by straw burning. The detection of straw burning can start from two aspects: flame and smoke. Because straw burning is usually accompanied by strong smoke, we decide to determine whether there is straw burning through smoke. The existing smoke detection methods all has various shortcomings, such as not using the dynamic characteristics of smoke, and inefficient and complex processing. Therefore, this paper proposes a smoke detection method based on improved frame difference method and Faster R-CNN. For smoke detection, first uses the improved frame difference method to extracts candidate regions, and then uses the Faster R-CNN model for smoke detection. For the extracted candidate areas, this paper proposes a variety of schemes to expands the candidate areas to ensure that the complete smoke information could be obtained to the maximum extent. Through the experiment, we get the best expansion scheme. Experiments shows that the improved frame difference method has obvious effects, compared to Faster R-CNN model method, the maximum accuracy rate has improved by 10.6%.