Authors

Abstract

To overcome the inefficiency and lack of system coordination caused by the separation of plug seedling sorting and replanting processes, this study designed and developed an intelligent integrated operation system that combines both sorting and replanting functions. The system aims to enhance the overall operational synergy and automation level of plug seedling management. It integrates modules for seedling grading, grasping parameter generation, and transplanting execution, thereby achieving autonomous identification, intelligent grasping, precise replanting, and efficient collection and reuse of weak seedlings. In the grading module, a comparative analysis of YOLO series models was performed, and YOLOv11 was selected for accurate identification of robust and weak seedlings. For the grasping strategy, a lightweight grasping pose parameter prediction network (LRGN) was introduced to generate optimal grasping angles and widths, effectively minimizing physical damage to the seedlings. Experimental results indicated that, for trays with a 4×8 cavity configuration, the recognition accuracy reached 96.0%, sorting success rate 96.67%, replanting success rate 96.0%, and leaf damage rate 2.15%. For trays with a 5×10 configuration, the recognition accuracy was 96.33%, sorting success rate 95.83%, replanting success rate 94.67%, and leaf damage rate 2.88%. The proposed system provides reliable technical support and a practical reference for advancing the intelligent and precise operation of plug seedling cultivation equipment.

Abstract in Chinese

针对穴盘苗分选与补栽功能分离导致的效率低、系统协同性差问题，本研究设计并开发了一种集分选与补栽功能于一体的智能作业系统，以提升整体作业的协同效率与自动化水平。该系统集成等级识别、夹取参数生成及栽植执行模块，实现了穴盘苗的自主识别、智能夹取、精准补栽以及弱苗的收集与再利用。在等级识别模块中，通过对YOLO系列模型的对比分析，选用YOLOv11实现壮苗与弱苗的准确识别；在夹取策略方面，引入轻量化夹取姿态参数预测网络LRGN，生成最优夹取角度与宽度，以降低对穴盘苗的物理损伤。实验结果显示，在4×8穴盘规格下，识别准确率为96.0%，分选成功率为96.67%，补栽成功率为96.0%，叶片损伤率为2.15%；在5×10穴盘规格下，识别准确率为96.33%，分选成功率为95.83%，补栽成功率为94.67%，叶片损伤率为2.88%。本研究为育苗装备的智能化与精细化作业提供了可靠的技术支持与实践依据。