A proposal of grading system for fallen rice using neural network
Fumiaki Takeda, Hideaki Uchida, T. Tsuzuki, Hideki Kadota, S. Shimanouchi · 2003
Rice sorting in high-speed is needed for the mass shipment. However, the recognition performance of the conventional sorter is not fast enough for the transaction volume concerned. In a conventional rice sorter, if the rice flow rate exceeds a few thousands [kg/h], the recognition percentage is below 90% and recognition ability is not guaranteed. Thus we propose a new system for the rice grading using the neural network. We show the effectiveness of the proposed method by simulation with real rice data, such as the normal rice and damaged rice. Furthermore, we also propose an extraction algorithm, which can sample a single grain or rice among a large quantity of rice in a single image frame. Moreover, we develop a prototype system for rice grading, and show its performance and effectiveness.