Least Square Support Vector Machine Analysis for the Classification of Paddy Seeds by Harvest Year

Least square support vector machines (LSSVM) were used to classify paddy seeds of different harvest years based on visible/near-infrared (vis/NIR) spectroscopy. The binary LSSVM classifier combined with a coding technique was extended for multi-class classification. To eliminate the noise and effectively extract the features of the spectral data, wavelet transform was implemented to decompose the spectral data. To evaluate the performance of wavelet coefficients, low-frequency coefficients and high-frequency coefficients were used as inputs of LSSVM classifiers. In addition to a linear kernel LSSVM classifier, a Gaussian radial basis function (RBF) kernel LSSVM classifier and a radial basis function neural network (RBF-NN) classifier were trained and tested. As a result, the RBF kernel LSSVM classifier outperformed the other classifiers with the best classification accuracy of 98% for samples in the prediction set. The results indicated that vis/NIR spectroscopy could be used to classify paddy seeds of different harvest years nondestructively, and the proposed method of integrating LSSVM with wavelet transform showed the potential for multi-class classification.