Chen Shan-Jing, Hu Yi-Hua, Shi Liang, Wang Lei, Sun Du-Juan, Xu Shi-Long
Electronic Engineering Institute, State Key Laboratory of Pulsed Power Laser Technology , Hefei 230037, China.
Guang Pu Xue Yu Guang Pu Fen Xi. 2013 Aug;33(8):2192-7.
A novel classification algorithm of hyperspectral imagery based on ant colony compositely optimizing support vector machine in spatial and spectral features was proposed. Two types of virtual ants searched for the bands combination with the maximum class separation distance and heterogeneous samples in spatial and spectral features alternately. The optimal characteristic bands were extracted, and bands redundancy of hyperspectral imagery decreased. The heterogeneous samples were eliminated form the training samples, and the distribution of samples was optimized in feature space. The hyperspectral imagery and training samples which had been optimized were used in classification algorithm of support vector machine, so that the class separation distance was extended and the accuracy of classification was improved. Experimental results demonstrate that the proposed algorithm, which acquires an overall accuracy 95.45% and Kappa coefficient 0.925 2, can obtain greater accuracy than traditional hyperspectral image classification algorithms.
提出了一种基于蚁群在空间和光谱特征上复合优化支持向量机的高光谱图像分类新算法。两类虚拟蚂蚁交替搜索具有最大类间分离距离的波段组合以及空间和光谱特征中的异类样本。提取了最优特征波段，降低了高光谱图像的波段冗余。从训练样本中剔除了异类样本，优化了样本在特征空间中的分布。将经过优化的高光谱图像和训练样本用于支持向量机分类算法，从而扩展了类间分离距离，提高了分类精度。实验结果表明，该算法的总体精度为95.45%，Kappa系数为0.925 2，比传统高光谱图像分类算法具有更高的精度。
