Hyperspectral data feature extraction using spatial coherence based neighborhood preserving embedding

Manifold learning methods, such as locally linear embedding (LLE) and neighborhood preserving embedding (NPE), extract the main structure feature of hyperspectral images for better understanding and processing of the data. However, these methods ignore the correlation between adjacent pixels in the image. In order to solve this problem, a spatial coherence based neighborhood preserving embedding (SC-NPE) feature extraction algorithm was proposed in this paper. In the proposed algorithm, local neighborhood structure of data was constructed through an optimal local linear embedding in high dimensional space. Meanwhile, the spatial context of pixels was considered by adopting the difference between the surrounding patch of pixels. Then local neighborhood structure of data was projected to the low-dimensional space to perform feature extraction by finding an optimal transformation matrix. Compared with LLE and NPE algorithm, this algorithm took into account not only the manifold structure but also the spatial information of hyperspectral image. As a result, the local neighborhood structure in the proposed algorithm is especially useful in low-dimensional space. The effectiveness of the proposed algorithm is demonstrated in hyperspectral data classification experiments.