Fig. 1

Scheme of the procedure for seed sorting and classification using supervised and unsupervised classification methods. The supervised method requires training of a classification algorithm, for which manual separation of seed samples under a (dissecting) microscope and dust classification via clustering is required. In the unsupervised method, the clustering algorithm is applied directly without prior training (and thus no manually-separated samples are required). Coarse separation involves sieving, sedimentation and drying prior to subsequent sorting performed by the large-particle flow cytometer. Feature extraction calculates indices summarizing the shape of each particle from the high-resolution axial profiles. All the steps after sorting are implemented in the SeedSorter R package