SEED CLASSIFICATION USING MACHINE VISION

This paper reports the results of applying digital image analysis in conjunction with statistical pattern recognition to measure the size and shape features of various seed types and to classify them into the primary grain, small seed, and large seed categories. The seed types used in each category were: hard red spring (HRS) wheat and barley as primary grains; canola, brown mustard, yellow mustard, oriental mustard, and flaxseed as small seeds; and 'Laird' lentils,'Eston' lentils, pea beans, green peas, black beans, and buckwheat as large seeds. The objective of the study was to assess the classification success in identifying HRS wheat and barley from other small and large seeds using morphological features. Orientation of the kernels for camera viewing was random. The kernels were, however, positioned manually in a non-touching manner. Hard red spring wheat and barley were correctly identified from all other seed types with more than 99% accuracy. Small and large seed categories were successfully discriminated from each other. Within each of the small and large seed groups, however, the classification was poor with up to 54.7% misclassification in small seed group and up to 30.3% misclassification in the large seed group. Canola yielded the worst classification with only 45.3% of canola seeds correctly discriminated from other small seeds.