The Pattern Recognition Of Producing Areas Of The Rice

In the rice trade and production, a fast and accurate evaluation of rice producing areas is imperative. Currently, the advanced recognition method is the near infrared spectroscopy technique, which extract the composition and structure of rice, and then analyze the producing areas of the rice by multivariate statistical method. It is very difficult to recognize where the rice grows, because there are more than 8000 near infrared spectrum characteristics. And we have also noticed that if producing areas of collected samples are neighboring and rice varieties are similar, the rice class will overlap. This means that the information characteristics are very similar and the process of rice recognition can be complicated. As a consequence, An attempt is made to find a feasible way to recognize the rice whose producing areas are neighboring.The near infrared data of the rice collected in four neighboring areas of southeastern Harbin is regarded as sample information. Fisher discriminant analysis, unexplained variance of step-wise discriminant analysis, Mahalanobis distance algorithm, F minimum value, Rao’s V and Wilks’ lambda are separately applied in the discrimination two groups awaiting recognition, among which unexplained variance is better, but only with a higher accuracy rate of approximate 80%. This means that a single method not works well. As a result, two combination methods are given.The first method is step-wise discriminant analysis. All the rice is firstly classified by Fuzzy C-means Clustering to establish different sample database. Then the classes of samples-awaiting recognition are identified based on the largest closeness degree and the highest similar degree of near infrared spectrum image. At last subclass sample databases are classified step by step by using the K-nearest neighbor classification of denial decision.The second method is a combination of fuzzy pattern recognition and hypothesis testing. The highest closeness degree and the lowest discrepancy ratio are regarded as discriminant conditions of producing areas. It has a very ideal recognition result, because samples can be particularly identified through external representation to the internal organizational structure. Therefore, this method can be applied in groups recognition.In conclusion, combination methods works well with a good recognition result, which can be popularized in recognizing the producing areas of the rice, while the a single discriminant analysis method, which is not ideal and suitable to recognize the rice whose producing areas are neighboring.