Research on Suspicious Person Detection in Forest Fire based on commercial technology

Tingting Sun
Shandong Ship Control Engineering And Intelligent System Engineering Technology Research Center, Weihai, 264300, China

Abstract:

In order to  detect suspicious persons effectively and prevent fire in the forest, and realize the intelligent inspection of drones, A detection algorithm of SE-AM-YOLOV3 based on multi-attention mechanism is proposed. Firstly, Attention mechanism was  introduced into Darknet-53 backbone network to replace the original residual module, enhance useful features, suppress useless features, and improve the accuracy of model detection of occlusion targets. Secondly, the attention mechanism is used to assign weights to features of various scales, and learn the correlation of features autonomously between channels. Finally, the camouflage data set is used for the algorithm. The experimental results show that the algorithm can quickly locate suspicious persons with an average accuracy of 94%, and the detection speed of a single image can   reach 0.8s.