Research on Forest Fire Detection Technology Based on Deep Learning

In order to make early warning on forest fire, this paper proposes a new method that combines classification model and objection model for forest fire recognition, and builds a fire warning system based on this method. Firstly, the algorithm optimized on VGG network, and introduced transfer learning method to train smoke recognition model and flame recognition model. Secondly, consider the difficulty of small flame recognition in the early stage of fire, introduced YOLO network for fire detection, and optimization on it to improve the ability of feature extraction and multi-scale feature fusion. Finally, made use of decision tree method for joint decision of fire warning by results of classification and detection. Experiments show that the mAP of this method is 96.5%, and the detection speed is 30.9FPS, which meets the real-time requirements of fire detection.