Fire-locating Based on Superpixel Features by Convolutional Neural Network

To recognize and locate fire accurately, we developed a wildfire-locating method based on superpixel features. Previous research detected fire at the pixel level of the image using deep learning, whereas we preprocessed and classified superpixels for fire recognition and locating. Firstly, a simple linear iterative clustering algorithm was created to segment the image into superpixel blocks. Then, color, texture, and shape features were extracted from each superpixel, and a convolutional neural network (CNN) was employed to classify the superpixels into two categories: fire superpixels and background superpixels. Finally, superpixels were refined based on the superpixel adjacency relationship. Experimental results demonstrated that the combined approach of superpixel features and CNNs performed satisfactory segmentation performance with an accuracy of 96.58%, which was effective in wildfire-locating.