遗传神经网络在稻米垩白度检测中的应用研究

新的优质稻谷国家标准中，垩白度是4个定级指标之一，被用来代表稻谷的商品品质。垩白度的检测目前仍由人工目测完成。为使检测结果更具客观性、一致性，建立了遗传神经网络对垩白像素和胚乳其它像素进行了识别，从而实现了垩白度的自动无损检测。对两种市售粳米进行了检测，计算机视觉的检测结果与人工检测结果的误差小于0.05。试验结果表明所建立的新方法是可行的，它为开发垩白度在线检测系统提供了科学依据。

Inspection of chalk degree of rice using genetic neural network

Chalk degree is one of the four important criteria for judgment of rice quality according to China National Standard of Rice. It has been determined by human inspection exclusively so far. A new method was developed to identify chalk and to grade chalk degree of rice using genetic algorithm and neural network in conjunction with computer vision. Genetic neural network was trained to identify chalk pixels and other pixels of endosperm and subsequently to evaluate chalk degree of rice. Two different kinds of rice bought on market were tested to evaluate system performance. Compared experiment results of new method using genetic neural network with that of human inspection, the error rate was less than 0.05. This method is proved to be robust and consistent. It paves the way for on-line automated judgment of chalk degree of rice.