A Blurry Image Recognition Method for Straw Burning Detection

Abstract

Straw burning strongly affects the atmospheric environment quality. Smoke detection can monitor the straw burning because smoke is a common feature of straw burning. Foggy weather and camera shake will blur the image, and blurry images may reduce the accuracy of smoke detection. To reduce the false alarm rate of smoke detection, the captured images should be filtered first before executing the smoke detection program. We propose a novel smoke detection method with the blurry image recognition. The gradient information is a critical parameter to evaluate whether the image is blurry. Then the Faster R-CNN processes smoke detection after eliminating the blurry images. In our confirmatory experiments, the method excludes the majority of interference images, and it improves the total average accuracy by 26%. At the same time, it does not significantly affect the detection accuracy of clear images.