Abstract:

Most of the transmission lines are in the mountain forests with complex environment, and early mountain fires often appear in the form of smoke. But the study of smoke detection methods for mountain fires in foggy conditions is rare. Aiming at the detection of wildfires in foggy weather conditions, this paper proposes a de- fogging image enhancement method. First, perform the localized equalization of the image; then conduct enhanced process by the global single- scale Retinex method, and detect smoke generated during early wildfires using CNNbased wildfires smoke detection network. Experimental analysis proves that the local and global image enhancement methods can significantly improve the accuracy of wildfire smoke detection, and the accuracy of smoke detection through convolutional neural network reaches 97.2%. 

Key words: transmission line, image de- foggy, smoke detection, convolutional neural network