SCS-YOLO11: a robust detection framework for pileus of deer antler mushrooms in greenhouse environments

In the intelligent cultivation of mushrooms within greenhouses, monitoring during the blooming period is crucial. This stage involves the formation and differentiation of young fruiting bodies, where timely detection of mushroom pileus is essential for automated environmental control. However, accurately detecting and counting immature caps remains challenging due to their small size, similar morphology, dense clustering, and complex background interference in greenhouse environments. To address these issues, this paper proposes an improved detection system, named SCS-YOLO11, based on the YOLOv11s architecture. To address the challenges of small-target detection in mushroom pileus recognition, we propose a coupled multi-scale attention (CMCA) module that effectively integrates global context and multi-scale spatial features. Additionally, a lightweight SPConv module is introduced to reduce computational cost while maintaining feature expressiveness, and a compact spatial-channel attention module (SCAM) further enhances feature discrimination in the detection head. It jointly models spatial and channel attention to guide the model to focus on key mushroom cap regions across multi-scale feature maps. Compared with the baseline YOLO11s model, SCS-YOLO11s shows remarkable improvements. Its precision increases from 79% to 84%, and mAP rises from 74.6% to 79%, with only 2.13M parameters and 3.6G FLOPs, demonstrating high efficiency. When applied to mushroom datasets, experiments show that its performance surpasses other YOLO-series models. SCS-YOLO11 strikes a balance between detection accuracy and computational efficiency, making it a promising solution for real-time monitoring of small mushroom pileus in the complex and dynamic settings of greenhouse mushroom cultivation.