Identification the appearance quality of rice kernels by vision technology and neural network classifier

In this study, the appearance quality of Hashemi variety of rice grains was evaluated using image processing and artificial neural network (ANN) classifier. Non-touching kernel images of different classes in a Hashemi rice sample were acquired using a flatbed scanner. Then preprocessing, segmentation, feature extraction and effective feature selection process were done on each objects of image. To categorized grains, various structures of ANN consisting network with one and two hidden layer with different hidden nodes, different training and transfer functions were considered. Results of validation stage showed ANN with 13-18-18-5 topology and LM training and tansig transfer functions had highest mean of classification accuracy (97.33%) and the lowest value of RMSE (0.08361). It’s concluded that the suggested method uses low cost equipment to identify quality of rice with acceptable accuracy. Results of this research can be used for fast and accurate grading and developing an efficient rice sorting system.