LFS-YOLO: A Lightweight Forest Smoke Detection Network Based on YOLOv8-n

Forest smoke detection has consistently been a challenging research topic. Due to environmental constraints, forest smoke detection typically relies on edge devices such as unmanned aerial vehicles (UAVs). In the early stages of forest fires, fire sources and smoke targets are often small, and traditional detection methods are limited by insufficient detection accuracy, large model sizes, and slow processing speeds, making them unsuitable for practical applications. In recent years, advancements in computer vision have led to research focusing on small object detection and real-time performance improvements. However, existing models still face significant limitations in complex forest environments and edge device deployments such as UAVs. To address these challenges, this study construct a forest smoke dataset comprising 7969 images from various forest scenes, which is used to train and validate the experimental models. Based on this dataset, we propose a lightweight forest smoke detection network—LFS-YOLO, built upon the YOLOv8-n architecture. Specifically, the backbone network integrates multi-scale feature extraction and efficient feature fusion modules to enhance the network's feature representation capabilities. The neck employs a depth-efficient convolution module, combining depthwise and pointwise convolutions to significantly reduce model size. Additionally, the detection head incorporates a lightweight shared detail-enhanced convolution structure, further reducing parameter count and computational cost. Experimental results demonstrate that the proposed model outperforms current mainstream target detection methods in terms of computational complexity and number of parameters, achieving a GFLOPs reduction to 2.9 and a parameter count of only 1.4M. Furthermore, it maintains competitive performance in both accuracy and FPS, highlighting its strong potential for practical applications in forest fire monitoring.