Research On Forest Fire Detection Method Based On Ensemble Learning

Forest fire is a serious natural disaster,and its occurrence is often due to the interaction of natural and human factors,such as high temperature,lightning strikes,indiscriminate logging,etc.Once a fire occurs,the fire often spreads rapidly,causing great harm to the ecological environment,human life and property safety,etc.Therefore,timely and accurate detection and early warning of forest fires to prevent fires from spreading in the early stage is crucial to protect forest resources and reduce fire hazards.Currently,a large number of researchers and scholars have explored the application of deep learning in forest fire detection,but there are still shortcomings in detecting early forest fires and small target flames at long distances.In addition,the use of a single model to perform target recognition tasks is prone to serious under-detection problems.To solve the above problems,this paper proposes an integrated learning-based forest fire detection method with the following studies:(1)To solve the problem of scarcity of datasets in the field of forest fire identification,this paper selects large-scale datasets of good quality containing different kinds of forest fires from publicly available flame datasets,covering surface fires,canopy fires,small fires,medium fires,and large fires.In order to meet the needs of monitoring and early warning of early forest fires and small target flames at a long distance,this paper makes small target datasets in a targeted manner.In addition,in order to improve the efficiency of dataset labeling,this paper proposes a semi-automatic labeling method,which can effectively complete the labeling of the dataset and greatly improve the labeling efficiency.(2)To address the difficult problem of small target flame detection for early forest fires,the U＿YOLOv5(YOLOv5 Upgrade)target detection algorithm is proposed,which makes four improvements on the basis of the YOLOv5 algorithm: adopting the Mosaic9 data enhancement method at the data input side to improve the feature extraction capability of the network for the input images;adding a new small target detection layer at the detection side to increase the The network perceptual field is increased by adding a small target detection layer at the detection side;the CBAM attention mechanism module is introduced in the backbone network to improve the model’s ability to extract key features from the template;and the original CIo U loss calculation is replaced with EIo U loss in the prediction stage to speed up the network convergence.The experimental results show that compared with the YOLOv5 algorithm,the U＿YOLOv5 algorithm has a significant improvement in the detection of flames in small target areas,and its m AP value is improved by 3.55%.(3)A forest fire detection algorithm based on integrated learning is proposed to address the problem of missing detection caused by different sensitivity of individual models to different types of fires.The algorithm combines two "good but different" individual models,U＿YOLOv5and Efficient Det,into a pair of "partners" to achieve a "complementary" effect.The algorithm forms a comprehensive and strongly supervised learning,which effectively improves the current system’s leakage detection problem.By comparing the algorithm with current mainstream target detection algorithms,the superiority of the proposed integrated algorithm in multi-class fire and multi-scale flame detection is verified.The experimental results show that the detection accuracy of the integrated algorithm proposed in this paper is improved by 2.5% and 3.5%,respectively,compared to the individual models U＿YOLOv5 and Efficient Det.(4)In this paper,an easy-to-operate forest fire detection system is designed and implemented based on integrated learning forest fire detection algorithms.By testing the functionality of the system,it is verified that the system can quickly and accurately detect flame areas in images or videos in complex forest environments,and provide information on the location and category of fires for forest fire early warning.