Assessment of machine vision algorithm for quantification of foreign matter in wheat

Canadian Grain Commission has stringent regulations on the cleanliness and uniformity of wheat grain for both domestic and export grades. Research at laboratory levels has demonstrated that machine vision is an effective method for classification of cereal grains. Robust machine vision algorithms have been developed and tested to extract morphological, color and textural features of cereal grains and dockage content. The objective of this study was to assess the ability of machine vision in classifying foreign matter (barley) in wheat using a machine vision algorithm. The samples used in this study were bulk images of Canada Western Red Spring (CWRS) wheat mixed with known quantities of barley (0.6 to 5%). Back propagation neural network (BPNN) and statistical classifiers were used for classification. Results of the study indicate that classification was reduced from about 94% for clean wheat to about 77% for 1.2% barley admixture and then increased again to about 97% for 3% and 5% barley admixture using neural network classifiers. This reflects that the machine vision algorithm was unable to classify 1.2% barley admixture correctly and requires some modification before it can be used for practical purposes.