基于改进YOLOv8网络的工厂火灾烟雾检测OA

Factory fires can lead to severe property damage and pose significant threats to human safety.Therefore,fast and accurate fire and smoke detection is of critical importance.To address the challenges of high computational complexity,low real-time performance,and insufficient accu-racy in existing object detection algorithms for factory fire and smoke scenarios,this paper propo-ses an improved YOLOv8-based detection algorithm.First,StarNet_s050 was introduced as the backbone network,which utilized depthwise separable convolutions and multi-scale feature extrac-tion to reduce complexity and enhance feature representation capabilities.Second,a novel C2f-StarBlock module was proposed,in which the traditional Bottleneck was replaced by StarBlock with a channel-wise cross-fusion mechanism to improve feature fusion efficiency.Finally,a light-weight shared convolutional(LWSC)detection head was designed to reduce the number of param-eters and computational complexity.Experimental results show that compared to the original YOLOv8 model,the improved algorithm reduces the number of parameters and computational complexity by approximately 55.8%and 48.3%,increases the detection speed by 23 frame per second,and improves accuracy by approximately 2.1%.Moreover,compared to other YOLO vari-ants and mainstream detection models,the proposed method achieves a better balance between de-tection accuracy and computational efficiency.This makes it more suitable for complex factory en-vironments,enabling more effective and precise fire and smoke detection.