基于改进YOLOv5的森林烟雾检测算法

Forest Smoke Detection AlgorithmBased On Improved YOLOv5

在森林火灾烟雾检测任务中,最重要的是能在火灾发生的初期及时发现小而稀薄的烟雾,目前的烟雾检测算法普遍存在检测精度较差,错检、漏检较多的问题。为解决上述问题,文章提出一种基于改进YOLOv5的烟雾检测算法。为了使提取到的多尺度特征更好的融合,强化对小尺度烟雾的检测,引入了BiFusion Neck融合结构,并对本森林火灾烟雾检测算法中的损失函数进行了改进,使用SIOU作为损失函数,使网络考虑真实框与预测框之间的方向,加快收敛速度。使用本文所提出的方法在自制烟雾数据集上实验,结果表明本文方法的整体性能要优于其他的烟雾检测算法。

In the forest fire smoke detection task,the most important thing is to find small and thin smoke in time at the early stage of the fire.Current smoke detection algorithmgenerally has the problems of poor detection accuracy,more false detection and missing detection.To solve the above problems,this paper proposes a smoke detection algorithm based on improved YOLOv5.In order to fuse the extracted multi-scale features better,the BiFusion Neck fusion structure is introduced,and the loss function in the forest fire smoke detection algorithm is improved.SIOU is used as the loss function,which makes the network consider the direction between the true bounding box and the prediction bounding box,and accelerates the convergence speed.Using the method proposed in this paper to test on public smoke data sets,the results show that the overall performance of this method is better than other smoke detection algorithms.