Processing And Application Of Dimensionality Reduction Of Spectral Data Of Potato

Potato is the main economic crops and food crops in the west of Inner Mongolia. Because of the large market demand of potatoes and its processed products, it has high economic value. At the same time, the potato has rich nutritional value. As a grain and a vegetable, it is one of the sources of dietary fiber, and also provides nutrients, such as protein, phosphorus, iron and so on. So It is the fourth most important food crop in the world. The role of potatoes in different industries is not the same. In industry, the potato starch can be used as calcimine. In the textile, thesis and other fields, it can be used as an additive. In medicine, the potato can be used as a raw material for producing yeast, a variety of enzymes, vitamins, and artificial blood. The different effects of potato is mainly caused by the difference of starch content. Different kinds of potato has different starch content, so classified potato accurate and detected starch content rapidly are necessary.The main researching content is as follows:1) Acquired the hyperspectral image of potato, and used The ENVI software to extract the reflectance spectrum data in the region of interest and then pre-process it.2) the Principal Component Analysis (PCA), Stepwise Discriminant Analysis and Successive Projections Algorithm (SPA) method were used to reduce the dimensionality of the spectral data of potato. Taked the data as input in a support vector machine model, and classified the three varieties of potato (Atlantic, Feiwuruita, Kexin).3) BP artificial neural network and Bias discriminant analysis model were established respectively. Take the Low-dimensional data by using the successive projection Algorithm as input data to classify the Atlantic, Feiwuruita and Kexin.4) This thesis established partial least squares model, BP artificial neural network model and successive projection model respectively. Favorita data that treated with Successive Projections Algorithm was used as input data, to predict the starch content of potato.SPA method is considered to be a good method to reduce the dimension,because the number of bands selected by SPA method is small and free. The identification accuracy rate on potato species reached 100%.In classification, because classification SVM classification model not only accurate, but fast, it is good classification model. Compared to the other models of starch content prediction, the mean square variance (RMSEP)of SPA model is equal to 1.3781, smaller than the other two models, correlation coefficient (R)is equal to 0.82976, larger than the other two models.so SPA model is better.