Abstract: A new classification method of hyperspectral remote sensing image based on SLIC and active learning is proposed. First, it extracts image texture and merges with spectral feature, gets new data, uses PCA to reduce dimension of new feature data, generates false color image by the top three principal components, then uses SLIC to obtain superpixels; second, randomly selecting superpixels to create initial training samples, spectral information mean value of all pixels in superpixels as spectral information of samples, samples’ label is the largest category of number of pixels. Then, it uses active learning to obtain SVM classifier; final, it classifies superpixels by classifier, class of each superpixel has been given to pixels in corresponding superpixel, so as to achieve the purpose of hyperspectral remote sensing image classification. Experimental results show that this method can evidently reduce the time of active learning, improve the classification results, and its OA, AA and Kappa value are significantly better than the active learning methods without SLIC.

Key words: active learning, superpixels, Principal Component Analysis（PCA）, Simple Linear Iterative Clustering（SLIC）, Support Vector Machine（SVM） classifier

摘要： 在主动学习的基础上，提出一种基于SLIC的高光谱遥感图像主动分类方法。首先提取图像纹理特征并与光谱特征融合，使用PCA对新数据进行降维，取前三个主成分构成假彩色图像，然后使用SLIC处理该图像获得超像素；接着随机抽取定量超像素作为初始训练样本，样本光谱信息为超像素样本中所有像素点的光谱信息均值，样本标签为超像素中出现次数最多的类别；然后通过主动学习得到SVM分类器；最后使用分类器对超像素分类得到其类别，并将超像素类别赋予其包含的像素点，从而达到高光谱遥感图像分类的目的。实验表明：该方法明显降低了主动学习过程的时间消耗，有效地提高了分类效果，其OA，AA和Kappa值显著优于未使用SLIC的主动学习方法。

关键词: 主动学习, 超像素, 主成分分析（PCA）, 简单线性迭代聚类（SLIC）, 支持向量机（SVM）分类器