Forest Fire Detection Based on Lightweight Yolo

Early fire detection is quite important to fighting of forest fire. A lightweight YOLO forest fire detection model is proposed in this study. First, the YOLOv4 object detection model is selected as the overall framework for fire detection. In order to reduce the number of parameters and computation, this study replaced YOLOv4's backbone network with the lightweight MobileNetV3. Finally, the data set with flame and smoke is used for training model, and fire-like images are used as negative samples to improve the robustness of the model. The experimental results demonstrate that compared with YOLOv4, the number of parameters of the proposed architecture is reduced by 62.78%, and inference speed increases by 3.04 times,. Moreover, the detection accuracy of the proposed algorithm achieves 0.666 mAP. The proposed method has advantages of less number of parameters, fast inference speed and low memory occupation. The method presented in this paper has referential value to realize real-time forest fire detection and object detection on embedded devices.