Improved YOLOv11 Model for Rose Detection and Counting

This study addresses the problem of automatic detection and counting of roses for yield estimation, presenting an enhanced YOLOv11-based model to overcome the time-consuming and error-prone nature of manual counting. To improve detection accuracy for small targets in complex environments, we introduce three key modules into YOLOv11: CBAM (Convolutional Block Attention Module), BiFPN (Bidirectional Feature Pyramid Network), and LMPDIoU (Log-Mean Polynomial Distance IoU). The CBAM module refines feature selection under variable lighting by utilizing spatial and channel attention; BiFPN enables efficient, bidirectional multi-scale feature fusion for dense flower detection; LMPDIoU optimizes bounding box regression, enhancing localization accuracy in occluded and small-target scenarios. Experimental results demonstrate that the improved YOLOv11 significantly outperforms other mainstream algorithms across various lighting conditions and complex scenarios, offering high application value.