YOLO-AREL: A Lightweight Algorithm for Wild Mushroom Species Recognition in Complex Environments

• Proposes YOLO-AREL, a lightweight improved model based on YOLOv11, tailored for wild mushroom recognition in complex environments. • Integrates the ADown module, RepViTBlock, and EMA attention mechanism, enhancing the model’s capability for multi-scale feature extraction, background suppression, and target-focused representation. • Designs LQEHead to optimize bounding box localization accuracy and category prediction reliability; the model achieves a 2.0 percentage point higher mAP@50 (82.5%) than the baseline YOLOv11n, with 23.9% fewer parameters and 19% lower GFLOPs, balancing detection precision and lightweight performance. • Realizes real-time inference (15 ms per frame on RTX 4060) and deployable GUI system, providing a feasible solution for forestry resource monitoring, on-site rapid wild mushroom recognition. As important forestry resources, wild mushrooms possess significant value in food consumption and ecological resource utilization. However, issues such as scale variations, morphological similarity, and cluttered backgrounds lead to poor recognition performance of existing deep learning models in complex environments (e.g., fields and markets). To address these issues, this study proposes YOLO-AREL, a lightweight improved model based on YOLOv11. First, the model incorporates the ADown downsampling module to enhance its capabilities for multi-scale feature extraction and background suppression, and integrates structural reparameterization to further reduce model complexity. Second, it integrates RepViTBlock and the EMA attention mechanism to improve the model’s feature expression ability in scenarios with morphological similarity and occlusion. Finally, an improved LQEHead detection head is adopted to enhance the bounding box regression accuracy and category prediction reliability. Experimental results on the self-constructed wild mushroom dataset show that YOLO-AREL achieves a precision of 81.5% and a mAP@50 of 82.5%, which are 1.3 percentage points and 2.0 percentage points higher than those of the baseline model YOLOv11n, respectively. Meanwhile, the model reduces the number of parameters by 23.9% and GFLOPs by 19%. When deployed and tested in the RTX 4060 environment, its average inference time per frame is 15 ms, which can meet the requirements of real-time detection. This study demonstrates that YOLO-AREL achieves a favorable balance between detection precision, inference efficiency, and lightweight performance, providing a feasible solution for rapid wild mushroom recognition and offering preliminary technical support for smart forestry and food safety monitoring applications.