CHG-YOLO: A Lightweight Forest Fire Detection Framework

Abstract: To address the challenges of substantial scale variations of flame and smoke targets, insufficient discriminability of weak-feature targets, and the susceptibility of fine-grained information to loss in complex forest environments, this study proposes a lightweight forest fire detection framework, termed CHG-YOLO. The framework collaboratively optimizes the network architecture along three dimensions, namely global-local feature perception, weak-target feature enhancement, and fine-grained information preservation. This study proposes the C3k2_RBC module in the global-local feature perception layer. The module leverages large-kernel reparameterization to expand the effective receptive field and enhance global contextual perception and local detail extraction for flames and smoke. The HyperACE_Fire module is proposed in the weak-target feature enhancement layer. It enhances the feature discriminability of weak flame and smoke targets and suppresses complex background noise interference in forest environments through attention enhancement, multi-scale spatial context modeling, and a feature-adaptive fusion mechanism. The GSConvE module is proposed in the fine-grained information preservation layer. It reduces the loss of fine-grained information during downsampling and enhances the representation of flame textures and smoke edges through dual-path feature extraction and channel reorganization. The ADown module is also incorporated to reduce the number of model parameters. On the self-built forest fire dataset, CHG-YOLO achieves a precision of 83.2%, a recall of 77.1%, an mAP50 of 84.2%, and an mAP50-95 of 64.8%. The model has 2.2M parameters, approximately 8.3% fewer than the YOLOv13 baseline, and achieves an inference speed of 375.7 FPS. Cross-dataset testing further validates the generalization stability of the framework across different scenarios. The experimental results demonstrate that the framework achieves a favorable balance among detection accuracy, lightweight design, and inference efficiency.