Classification of bulk grains using their reflectance characteristics.

摘要

The grain handling system in Canada would benefit from research involving machine vision systems that produce a more consistent, error free, fast and reliable technique for grain grading than that is currently available. It was hypothesized that the machine vision system could be improved by using reflectance characteristics as one of the parameters in classifying grain. The reflectance characteristics of seeds from seven cereals and buckwheat, 10 pulses, three oilseeds and 25 specialty crops were recorded using a spectrophotometer (Model: Cary 5, Varian Canada Inc., Mississauga, ON). The effects of the growing region, seed moisture content, and foreign material content in bulk samples, on the reflectance characteristics of Canada Western Red Spring (CWRS) wheat were also determined.;From the reflectance curves, 465 features based on the ratios, slopes and slope-ratios of the reflectance data were extracted and tested as three models for classification. Procedure STEPDISC was used to rank the features and the top 20 features were used in Procedure DISCRIM for classification. A back Propagation Neural Network (BPNN) was used to collect the weights of the individual features and the top twenty features were used to test the classification accuracy. Ratio features and the slope-ratio features were more successful in classifying than the slope features.;BPNN and discriminant analysis performed similarly in classifying bulk grain. The top twenty features consisted of features from many regions of the electromagnetic spectrum. These classifiers were not successful in classifying the effects of the growing regions and crop-year, moisture content or foreign material content of CWRS wheat, i.e. these parameters do not affect the reflectance characteristics significantly.