Purity discrimination of maize seeds based on hyperspectral imaging combined with large margin projected transductive support vector machine

Abstract. Hyperspectral imaging is a promising technique for identifying seed variety, in which the identification accuracy relies on the selection of training set. All the traditional methods assume that training set and test set are drawn from same distribution. However, the test accuracy of model will decline when training set and test set have different distributions. This study develops the large margin projected transductive support vector machine (LMPROJ) for purity discrimination of maize seeds with same variety from different harvested years. Based on LMPROJ, the distribution between training set and test set is not required to be the same, which expands the application scene of the algorithm. The LMPROJ model obtained 90% – 94% purity of test set, which were higher than 71% – 92% of SVM model. Experiments show that the method based on LMPROJ has better adaptability than the traditional SVM method.