International Journal of Artificial Intelligence and Neural Networks
Author(s) : PHUVIN KONGSAWAT, NAWAPAT JAMROENRAK, TUANJAI ARCHEVAPANICH, BOONCHANA PURAHONG, ATTASIT LASAKUL
This article presents an inspection system to detect red kernel defect, normally contaminating in white rice product. This contamination causes a reduction in the price rice of 6% approximately. To detect Red defect successfully, a method proposed in this paper was build up on Machine Vision techniques. The method contains three processing steps as follows. Firstly, noise elimination and localization were executed through image processing techniques. After that, RGB image would be transformed to HSV in order to obtain discriminative features. Finally, the pre-processed data was then passed into model training by using both linear and non-linear Support Vector Machines. Apart from that, Logistic regression was then employed to challenge margin maximization ability of the SVMs. The experimental result shows that linear-SVM still yields the highest performance at 86.3% of classification accuracy.