Grain classifier with computer vision using adaptive neuro-fuzzy inference system.

BACKGROUND: A computer vision-based classifier using an adaptive neuro-fuzzy inference system (ANFIS) is designed for classifying wheat grains into bread or durum. To train and test the classifier, images of 200 wheat grains (100 for bread and 100 for durum) are taken by a high-resolution camera. Visual feature data of the grains related to dimension (#4), color (#3) and texture (#5) as inputs of the classifier are mainly acquired for each grain using image processing techniques (IPTs). In addition to these main data, nine features are reproduced from the main features to ensure a varied population. Thus four sub-sets including categorized features of reproduced data are constituted to examine their effects on the classification. In order to simplify the classifier, the most effective visual features on the results are investigated.

RESULTS: The data sets are compared with each other regarding classification accuracy. A simplified classifier having seven selected features is achieved with the best results. In the testing process, the simplified classifier computes the output with 99.46% accuracy and assorts the wheat grains with 100% accuracy.

CONCLUSION: A system which classifies wheat grains with higher accuracy is designed. The proposed classifier integrated to industrial applications can automatically classify a variety of wheat grains. © 2017 Society of Chemical Industry.