Development of machine for evaluating quality of rice by using real time image processing

Abstract

Quality assessment of rice prior to marketing is very important. Up to now, the rice quality inspection is conducted visually by trained examiners who have expertise and experience, but the method used has disadvantages such as: (1) the subjectivity factor that causes rice quality testing results to be biased between the observer, (2) the physical exhaustion of observer causes the observation result is inconsistent, and (3) the time required relatively much longer. The purpose of this study was to develop a method of determining the physical quality of rice by image processing techniques in real-time. The method used was the technology of image processing and artificial neural networks. Architecture of ANN to predict the physical quality of rice was built with 13 input layers, 20 hidden layers and 8 output layers, while the ANN to predict the degree of rice milling was built with 9 input layers, 20 hidden layers and 5 output layers. The results showed that the machine for testing rice physical quality in real time had worked functionally as expected, both hardware and software. The training process of the 5 rice varieties showed good results, especially in the estimation of head rice which was above the average 90 %, whereas the accuracy of validation decreased due to predictive accuracy of foreign matter dropped drastically due to the irregular shape and varied colors, making it difficult for the system to recognize it. The test results show rice milling training and validation results are not consistent, so it can not be resumed in the application process.