Research on Grassland Fire Detection Based on Improved YOLO Algorithm for Unmanned Aerial Vehicle Images
Once a grassland fire occurs,it spreads rapidly and irregularly around due to the influence of wind,terrain and other factors,forming a burning strip with an expanding area.In order to improve the efficiency of grassland fire detection,combining with the image characteristics of grassland fire captured by unmanned aerial vehicle(UAV),we study the grassland fire detection method based on the improved YOLO algorithm.Firstly,for the characteristics of long and narrow fire area and small percentage of fire area,the Neck part of YOLO algorithm is optimized,and a feature extraction network FC-FP Neck with full link structure is proposed,so that the semantic features and localization features are fully integrated and the feature extraction ability of the network is improved.Secondly,an improved adaptive weighted loss function is proposed by combining the threshold segmentation technology to improve the model's convergence speed,and at the same time,solve the problem of insufficient sensitivity of fire detection,which is easy to cause false detection.The feasibility of the improved algorithm is tested on the public small target detection dataset AI-TOD,and the average accuracy is improved by 7.28%and the average precision is improved by 12.46%;the average precision on the self-constructed grassland fire dataset reaches 90.24%and the average accuracy reaches 87.33%.The experiment shows that the improved algorithm improves the efficiency of grassland fire detection.
grassland fireYOLO algorithmfeature pyramid networkthreshold segmentationadaptive weight loss function
国家科技期刊平台
NSTL
万方数据
