Dockage identification in wheat using machine vision

Abstract

Algorithms were developed to classify dockage components from Canadian Western Red Spring (CWRS) wheat and other cereal grains like durum wheat, barley, rye, and oats based on morphological and color features. The dockage classes used were: wheat heads, chaff, wildoats, canola, wild buckwheat, flax, and broken-wheat pieces. The wheat head dockage class was subdivided into single and multiple wheat heads to improve the classification accuracy. The developed algorithms were tested on images taken with an area scan camera. Training and test data sets were established to evaluate the classification accuracies based on the extracted features. Morphology-color, morphology, and color models were evaluated for classifying the dockage components. Morphology-color model gave 90.9 and 99.0% mean accuracies when tested on the test and on the training data sets, respectively. The mean accuracies of 90.5 and 98.7% were obtained when the first 15 features from the morphology-color model were used on the test and on the training data sets, respectively. The mean accuracies of 89.4 and 96.3% for the morphology model and 71.4 and 75.6% for the color model were achieved when tested on the test and on the training data sets, respectively.