Abstract
Recently, a growing number of advanced hyperspectral remote sensing image classification techniques have been proposed and reported superiority in accuracy on the public available urban datasets, e.g., the Washington DC, and the Pavia Centre and University. Since the task of hyperspectral image classification is basically a special case of pattern recognition, many of these dominate techniques are machine learning based methods, such as the manifold learning, the sparse representation, and the newly raised deep learning. However, according to the literature review, most of them are the supervised learning methods, which require a certain number of high quality training samples to obtain an optimal model, and there is a huge imbalance between the researches of the unsupervised approaches compare to the supervised ones. In fact, the high performance of hyperspectral image unsupervised classification (i.e., clustering) without any prior information still leaves as a huge challenge. In this work, we present a method based on spectral clustering algorithm for hyperspectral urban image unsupervised classification, which tries to overcome the following two issues of the kmeans clustering in practice: (1) the course of dimensionality of the hyperspectral data, and (2) the unstable results caused by the initialization. In detail, the principal component analysis (PCA) is employed to reduce the feature dimensionality of the spectral domain, and then a spectral clustering based algorithm with adaptive neighbors is introduced for data clustering in the projected feature space. Experimental results on the standard hyperspectral urban image show that the proposed method not only outperforms the kmeans clustering in accuracy, but also provides reproducible performance.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have