基于改进YOLO v5的森林火灾检测算法研究

Study on Forest Fire Detection Algorithm Based on Improved YOLO v5

为提高森林火灾检测识别率,同时降低漏检、误检率,缩减后续救援时间,提出一种基于改进YOLO v5的森林火灾目标检测算法。针对火灾目标漏检问题,在YOLO v5骨干网络加入卷积注意力机制(CBAM),对森林火灾的小目标在通道和空间进行特征提取;针对火灾目标重合问题,把路径聚合网络(PANet)替换为加权双向特征金字塔网络(BiFPN),增强模型识别能力;针对YOLO v5中原始采样方式无法将特征信息完全捕获的问题,所以将CARAFE上采样作为最新的上采样方式,提高采样效率。实验结果表明,与其它主流算法相比较,该算法拥有更好的检测效果,其检测性能相比于原始算法提升了2.6%,更适用于森林火灾检测。

In order to improve the detection and recognition rate of forest fires,reduce the missed detection and false detection rates,and shorten the follow-up rescue time,this paper proposes a forest fire target detection algorithm based on improved YOLO v5.Aiming at the problem of missed detection of fire targets,this paper adds a convolutional attention mechanism(CBAM)to the YOLO v5 backbone network to extract features of small targets in the channel and space of forest fires.Aiming at the problem of overlapping fire targets,the path aggregation network(PANet)is replaced with a weighted bidirectional feature pyramid network(BiFPN)to enhance the model recognition ability.In view of the problem that the original sampling method in YOLO v5 cannot fully capture the feature information,CARAFE upsampling is used as the latest upsampling method to improve the sampling efficiency.Experimental results show that compared with other mainstream algorithms,the proposed algorithm has better detection effect,and its detection performance is improved by 2.6%compared with the original algorithm,which is more suitable for forest fire detection.