Smoke and Fire Detection Based on YOLOv7 With Convolutional Structure Reparameterization and Lightweighting
Junjie Hu, Yun He, Ming Zeng, Yingjing Qian, Renmin Zhang
Abstract
This letter presents a YOLO-based detection model called Diverse branch block and slimneck-YOLO (DS-YOLO) [Diverse Branch Block (DBB) and SlimNeck], which enables fast and accurate identification of smoke and fire. First, Diverse branch block and partial convolution- efficient layer aggregation network (DP-ELAN) module in the backbone is constructed based on DBB and partial convolution. It is possible to enhance detection capability while reducing the number of parameters. Second, utilizing the SlimNeck based on ghost shuffle convolution to reduce model complexity. Finally, we use normalized Gaussian Wasserstein distance instead of the standard intersection over union to improve the detection capability for small-sized objects.The experimental results on our custom smoke and fire dataset indicate that the proposed DS-YOLO achieves a mean average precision of 70.1%, while reducing computational complexity. This represents an 1.3% improvement over the baseline model.