SEARCH

SEARCH BY CITATION

Abstract

While being developed, plant somatic embryos change shape and increase size. An effective kinetic description of growth and development of somatic embryos is important for process scale-up and optimization. An essential component of such a kinetic description is the developmental characterization of the individual embryos present in culture. Embryo morphological data obtained by image processing techniques were transformed into sizeand size-independent morphological descriptors. Qualitative relations between the descriptors and geometric properties of the embryos were established to interpret the results. For training, a branch-and-bound search technique was used to search for optimal subsets of descriptors, as determined by member clustering and class separability properties evaluated from within-class and between-class scatter matrices. In the classification mode, individuals were identified using a voting nearest neighbor classifier. This nonparametric nearest-neighbor classifier was trained on optimal projections of the feature space established from developmental stage discrimination (branch-and-bound algorithm). Using a test population, normal and abnormal embryos and callus were assigned to six morphological classes. The image-analysis-based classification was in 80–90% agreement compared to the results obtained through visual classification by an experienced operator.