YOLO-RSTS: a precise segmentation model for detecting preservative and stimulant spraying regions on rubber trees

The application of preservatives and ethylene stimulants is critical for improving latex yield and extending the lifespan of rubber trees; however, traditional manual spraying methods are inefficient and unsuitable for large-scale plantation management. Moreover, existing segmentation models are challenged by complex bark textures and varying illumination conditions, resulting in blurred spraying boundaries and reduced recognition accuracy. To address these issues, this study proposes an improved segmentation model based on the YOLOv12n-Seg framework, termed YOLO-RSTS (YOLO for Rubber Spraying Target Segmentation), for accurately distinguishing preservative and stimulant spraying regions on rubber trees. The proposed model introduces three novel modules: CrossScaleDSC, CG-Attention, and C2f-DSC, which enhance long-range dependency modeling, suppress background noise through combined spatial-channel attention, and enable fine-grained multi-scale feature extraction with low computational complexity. In addition, RFCAConv and DWConv are incorporated into the backbone and head to strengthen spatial diversity and contextual representation. Experiments conducted on a self-constructed dataset demonstrate that YOLO-RSTS significantly outperforms the baseline YOLOv12n, achieving improvements of 6.3% in Precision (from 0.819 to 0.882), 6.3% in mAP0.50 (from 0.788 to 0.851), and 8.1% in Recall (from 0.740 to 0.821), while reducing the parameter count by 14.5% (from 2.72M to 2.33M). Meanwhile, compared with the latest YOLOv13n, YOLO-RSTS also achieves superior performance, with increases of 7.5% in mAP0.50 and 9.2% in F1 score. These results indicate that the proposed method provides an effective and efficient solution for vision-based autonomous spraying and holds significant potential for advancing intelligent rubber plantation management.