Early Image Recognition of Forest Fire Based on Improved YOLOv5 Neck Network

Forest fires can lead to environmental degradation and loss of biodiversity. In order to detect forest fires more efficiently in the early stage, an improved YOLOv5 algorithm is proposed in this paper. Adding a small target detection layer to YOLOv5 neck network to enhance the Identification ability of small target images at the onset of forest fire, because the increase of channel length of feature fusion network after adding a small target layer will lead to information loss, therefore, BIFPN feature fusion network is fused. Experimental results indicate that improved YOLOv5 outperforms the original YOLOv5 in small target recognition of fire image, especially when the flame is occluded.