MSA-YOLO: A Lightweight Detection Model for Wheat Spikelet Fusarium Head Blight Based on YOLO11

Fusarium Head Blight (FHB) is one of the most destructive fungal diseases in global wheat production. Traditional methods for FHB detection face limitations such as high technical expertise requirements, limited coverage scope, and insufficient timeliness, making them inadequate for modern precision agriculture management demands. To address this challenge, this study proposes a lightweight MSA-YOLO detection model based on the YOLO11 deep learning framework. The proposed model achieves a favorable balance between performance and efficiency through three innovative design aspects: first, it replaces the original backbone network with the MobileOne network, establishing a foundation for model lightweight design; second, it substitutes the multi-head attention mechanism in the C2PSA module's PSABlock with a more computationally efficient SE module, further reducing model complexity while maintaining detection performance; finally, it introduces an Adaptive Threshold Focal Loss (ATFL) function to address class imbalance issues, enhancing the model's recognition capability for minority classes. The experimental data comprise 629 photographs of wheat spikelets covering various growth and development stages. Results demonstrate that the improved MSA-YOLO model reduces parameter count from 2.58M to 1.65M and computational complexity from 6.4 GFLOPs to 3.9 GFLOPs. Furthermore, comparative analysis with YOLOv10, YOLOv9, YOLOv8, and YOLOv5 models shows that MSA-YOLO exhibits an exceptional balance between speed and accuracy, making it well suited for practical applications in precision agriculture monitoring systems.