Study On Manifold Learning Algorithms Of Hyperspectral Image

Hyperspectral remote sensing image(HSI) consists of hundreds of bands that contain rich space, radiation and spectral information. The high-dimensional data can also lead to the course of dimensionality problem making it difficult to be used effectively. In the paper, we introduced a manifold learning algorithm to reduce the dimensionality for HSI data. For high dimensional datasets with continuous variables, it is often the case that the data points are arranged along with low dimensional structures, named manifolds, in the high dimensional space. Manifold learning aims to identifying those special low dimensional structures for subsequent usage such as classification or regression. However, many manifold learning algorithms perform an eigenvector analysis on a data similarity matrix whose size is N×N, where N is the number of data points. The memory complexity of the analysis is at least O(N2) that is not feasible for a regular computer to compute or store for very large datasets.To solve this problem, a manifold learning dimensionality reduction framework for hyperspectral image is introduced. Firstly, statistical sampling methods were used to sample a subset of data points as landmarks. A skeleton of the manifold was then identified based on the landmarks. The remaining data points were then inserted into the skeleton by locally linear embedding(LLE). At last, original data sets and data sets reduced with different manifold learning approaches were classified by k-nearest neighbor(KNN) classifier to evaluate the performance of the proposed framework.A L1 norm based neighbor selection algorithm was introduced to the framework to deal with puzzles with neighborhood selection in isometric mapping(Isomap) and LLE because of L1 norm’s sparsity. Firstly, L1 norm was combined with original L2 norm in neighborhood selection step in Isomap approach due to possible short circuiting problem. Next, the algorithm is enhanced using the L1 norm to automatically select sparse neighbors. To fully utilize the available information, an adaptation is incorporated into the proposed general procedure to account for labeled information.The hyperspectral image manifold learning framework with local curvature variation and enhanced manifold learning approach was utilized on airborne visible infrared imaging spectrometer(AVIRIS) datasets. Different sampling methods, manifold learning approaches and key parameters were tested to evaluate the results.The results from experiments showed that the enhanced algorithms performed better than linear methods as well as using the raw datasets during dimensionality reduction. The main structure was preserved while the scale of the data was reduced. It is convenient for subsequent applications such as analysis of land use and land cover. The results showed that the classification accuracy of the improved Isomap performed 2 percent to 5 percent better than other manifold learning approaches such as PCA and original Isomap.