Research On UAV Forest Fire Image Detection Algorithm Based On Deep Learning

Forest fire is a disaster that occurs in the natural environment of the forest,causing adverse effects on the environment and ecology.In order to monitor the changes of forest fires and the movement of personnel in real-time,forest fire departments and firefighting departments widely use unmanned aerial vehicles for aerial tracking.Due to the complexity of the natural environment of the forest,the detection accuracy and speed of ignition points and personnel in the forest become the key to the efficiency of firefighting and rescue operations in the fire scene.In this paper,aiming at the shortcomings of traditional forest fire detection methods such as single target and low accuracy,based on deep convolutional neural network,a high-efficiency and multi-target unmanned aerial vehicle forest fire image detection method(YOLO-TF)is proposed,and the main work is as follows:Firstly,the application of drones in forest fire detection and the YOLO series fire image analysis network based on drone aerial images are introduced.According to the target detection requirements in firefighting and rescue,this paper proposes to replace the traditional single target detection with multiple target detection.Specifically,the detection algorithm is built using the key elements "Smoke","Fire",and "Person" in the forest fire detection task as the detection targets to address the problem of single detection target in the existing neural network detection methods for UAV forest fire images.Secondly,in order to address the characteristics of the ignition point,smoke,and personnel targets in forest fire detection,this paper introduces a hybrid attention mechanism into the YOLO-TF algorithm to improve the weight allocation of the deep learning network to each target object,thereby enhancing the network’s ability to detect multiple targets.Then,in order to address the problem of insufficient training image data for new targets in UAV forest fire detection research,a hybrid network,an open FLAME dataset,and domestic UAV forest fire aerial image data were used to construct an experimental dataset.The model was tested on the constructed dataset and compared with the latest fire detection algorithm based on YOLOv5.The results showed that the proposed YOLO-TF algorithm had higher detection accuracy and speed than the YOLOv5 algorithm,effectively demonstrating the feasibility and effectiveness of the proposed detection method in forest fire monitoring tasks.Finally,in order to further verify the effectiveness of the attention mechanism in the YOLO-TF framework,as well as its adaptability to multiple targets and scenarios in forest fire detection,the dataset constructed in this paper was further divided into different perspectives and scenes,and ablation experiments were introduced.The results show that the introduction of the attention mechanism improves the performance of the original framework and makes it more adaptable to multiple scenarios in forest fires.The deep learning network proposed in this article has the advantages of fast response,high detection accuracy,and strong adaptability.It can effectively improve the monitoring ability of forest fires by drones and provide powerful technical support for assisting forest fire rescue.