Research on Phenotypic Parameter Estimation and Sorting System of Sugarcane Setts for Pre-Cut Planting

Sugarcane is an important economic and energy crop cultivated globally. Pre-cut sugarcane planters have advantages such as seed saving and high efficiency, but irregularly shaped sugarcane setts can impair seed metering performance. Therefore, sugarcane setts must be screened by phenotypic parameters before planting. Manual sorting is labor-intensive, and existing studies rarely integrate parameter acquisition with sorting systems. To address this, we propose a machine vision-based method for estimating sugarcane sett phenotypic parameters. The sugarcane variety used in this study was Guitang 40. First, improved YOLOv8-obb and YOLOv8-seg algorithms are used to establish detection and segmentation models for sugarcane setts and nodes, achieving mAP@0.75 of 87.8% and 93.4%, respectively. Second, node number is calculated from rotated bounding boxes based on affiliation relationship between nodes and sugarcane setts, with 98% accuracy. Length, diameter, and curvature are estimated from sugarcane sett contours. Length is derived from the longer side of the minimum bounding rectangle, diameter from the two long edges after corner partitioning, and curvature by fitting Bezier curves. The mean errors and root mean square errors (RMSE) for length and diameter are 3.14 mm, 0.9 mm and 3.6 mm, 1.01 mm, respectively, and the RMSE for curvature is 0.042. Third, a sorting system is established to conduct experiments. The sorting success rates for sugarcane setts unqualified in node number, length, diameter, and curvature are 96.6%, 91.7%, 85%, and 80%, respectively. After sorting, the multiple-seeding rate and miss-seeding rate of the seed metering device were 3.54% and 2.55%, respectively. Under the tested mixing conditions, sorting reduced both the multiple-seeding rate and the miss-seeding rate, indicating that the proposed method can improve metering performance when unqualified sugarcane setts are present.