Answers are grounded in this paper and cite their sources.
Precise identification of rice seedling nurseries from remote sensing images is highly dependent on semantic segmentation. Accurate semantic segmentation sample labeling requires pixel-level precision, a time-consuming and labor-intensive task. This study designs a detection network based on YOLOv8 as the prompter network for a “Segment Anything” Model (SAM). Compared to pixel-level labeling, using detection boxes as training labels effectively reduces the complexity of data annotation. YOLOv8's detection network is modified by incorporating the lightweight MobileNetV3 as the backbone network. By employing dynamic snake-like convolutions to enhance feature extraction capabilities specifically for the small-sized, elongated and predominantly rectangular structures of rice seedling nurseries, it aims to improve the accuracy and efficiency of rice nursery identification, better supporting SAM with prompt information. The proposed method was evaluated using high-resolution unmanned aerial vehicle (UAV) imagery that was resampled to a spatial resolution of 1 meter after preprocessing. Experimental results show that the optimized YOLODSC-SAM network achieved Precision and mIoU of 95.9% and 94.9%, respectively, marking improvements of 4.7% and 3.4% over the original YOLO-SAM model. The proposed method enables precise segmentation of rice seedling nurseries, particularly for nursery distribution mapping and pre-transplantation planning.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: