Tree seeds vary in size and shape, which influences seed processing efficiency and impacts on seed lot quality. In some species, seed lots contain high proportions of empty seeds and cannot be single sown in container nurseries. Nowadays single-seed processing is possible with seed phenotyping, which generates metrics from images produced across the electromagnetic spectrum. Seed phenotyping is well-suited to automation, which potentially enables real-time assessments of seed lot quality. This research aims to develop software that can use image metrics to automatically differentiate between seed categories (filled, empty or insect-infested) from x-ray images.
The research objective is to develop software that can use image metrics to automatically differentiate between seed categories. The main aims are to:
Using R-software, three functions have been developed to automatically predict seed category. The first function processes raw images by removing background objects, re-aligning seeds, and standardizing image intensity with a fixed calibration reference. The second function extracts relevant image metrics including line profiles and texture for each seed. The third function tests the attributes of loaded seeds against a library of pre-identified seeds. Using hard validation, our models had a high degree of accuracy and precision, which shows potential for further development with a wider range of species.
Seed Technician