结合不确定性准则和差分进化算法的高光谱数据分类方法

he cost of obtaining the feature labels for Hyperspectral image data is high，which has a great impact on the classification accuracy. The semi-supervised classification method can achieve better classification performance by using a small mount of supervisory information with given labels and a large amount of unsupervised information. After combining with the active learning method, a good classification effect is achieved, but there is still room to improve the accuracy and computational efficiency. Therefore, this paper proposes a semi-supervised classification method combining multi-level uncertainty sampling strategy and adaptive differential evolution algorithm. First, samples with higher information content and samples with determinable categories are selected from unlabeled samples through the multi-level uncertainty criterion, the latter and known labeled samples are used to label the former together, and then the adaptive differential evolution algorithm are used to execute optimization, expand the training samples, and finally use the new training sample set to train the support vector machine(SVM) classifier to classify the test samples. Experimental results show that this method can make full use of sample information and effectively improve the classification accuracy of hyperspectral images in the case of small samples.