A novel skeletonization algorithm combined with hierarchical segmentation for phenotyping siliques of oilseed rape

Abstract. Silique morphology phenotypes are important yield-related traits in winter oilseed rape (Brassica napus L.). It is challenging to quantify the morphological traits of siliques due to the complex structures of oilseed rapes at the reproductive stage. This study aims to develop an accurate skeletonization algorithm combined with the hierarchical segmentation method to separate siliques from the whole plant using three-dimensional (3D) point clouds. Firstly, outlier noise points are removed from the raw 3D point clouds using a density outlier filter. And pot points are removed based on least-squares fitting circles. After pre-processing, the L1-median skeleton combined with the random sample consensus (Ransac) constraining extracts skeleton points iteratively to achieve skeletonization. Furthermore, skeleton points are connected by the distance, angle, and direction to neighboring points. The hierarchical segmentation including the coarse and fine segmentation steps is then performed using the density-based spatial clustering of applications with noise (DBSCAN) and weighted unidirectional graph (WUG), respectively. The number, length, and volume of siliques are automatically calculated. Experimental verification, tested using three cultivars of oilseed rape, 10 pots totally, demonstrates that the proposed method performs well in skeleton extraction, silique segmentation, and phenotypic traits extraction. Compared with the ground truth, the accuracy of the proposed method was proved, according to results that the R2 of number and total length of siliques are 0.962 and 0.942 respectively, and the average recall of silique segmentation is 93.33%. In addition, the correlation coefficient R of number, total length, and total volume of siliques related to yield per plant are 0.910, 0.905, 0.929, respectively, which indicates the potential our method in estimating yield. The proposed method could achieve accurate phenotyping of siliques, and it is promising for the selection of high yield and quality rapeseed in the breeding programs.