基于主动学习和空间约束的高光谱影像分类

Hyperspectral Image Classification Based on Active Learning and Contextual Constraints

高光谱影像具有数据量大、波段数多和信息冗余等问题,其分类一直是目前的一项研究热点。针对高光谱影像分类存在的问题,本文提出了一种利用主动学习和空间约束的高光谱影像分类方法。首先利用样本的先验分布状态建立样本的置信度模型,迭代选择最有"价值"的样本扩充训练样本库,以此训练最优的支持向量机分类器对高光谱影像进行分类,然后利用马尔科夫随机场(Markov Random Fields,MRF)引入空间信息,优化分类结果。文中在Indian Pines数据集上验证提出方法的有效性。实验结果表明,本文提出的方法通过样本的先验信息训练最优的SVM模型,能够有效地分类不同地物,总体分类正确率达到88%以上。

Due to the large amount of data,the number of bands and the redundancy of information,hyperspectral image classification has always been a hot topic. Aiming at the problems of hyperspectral image classification,this paper presents a hyperspectral image classification method based on active learning and contextual constraint. Firstly,using the prior distribution of the sample to establish the confidence model of the sample,iteratively selects the optimal＂value＂sample expansion training sample database to train the optimal SVM classifier to classify the hyperspectral images. Then,the spatial information is introduced using Markov Random Fields（MRF） to optimize the initial classification results. We validate the effectiveness and robustness of the proposed method using Indian Pines datasets. The experimental results indicate that the proposed method trains the optimal SVM classifier by the prior knowledge of samples,which can effectively classify different objects with the overall accuracy of over 88%.