Water Change Detection Based On Sparse Unmixing Of Hyperspectral Imagery

Hyperspectral remote sensing is an advanced remote sensing technology. Due tohyperspectral imagery with high spectral resolution has the unique characteristic ofacquiring spectral and spatial information simultaneously, it has been widely applied invarious fields, such as water change detection, environmental monitoring, geologicalprospecting, agriculture and forest survey, marine biological and physical research fields.However, due to the low spatial resolution of hyperspectral data and the complexitydiversity of natural surface, a pixel in hyperspectral remote sensing imagery is usually ahomogeneous mixture rather than a single feature, the existence of mixed pixels affectobject recognition and classification accuracy. The spectral unmixing technique is the mosteffective method to solve the mixed pixel problem, which estimates the number ofendmembers and corresponding fractional abundances in each mixed pixel. In order toachieve water change detection, the unmixing accuracy improvement becomes very critical.In recent years, many new unmixing methods have been proposed, sparse based approachhas recently received much attention in hyperspectral unmixing area and obtained verygood result, but it also has some shortcomings. In order to achieve water change detection,focusing on the application of sparse method in hyperspectral unmixing, this thesis hasmade a lot of research, and the main works and innovations are as follows:1. A novel smoothed0regularizer sparse unmixing method has been proposed,namely SL0SU algorithm. Despite the success of sparse unmixing based on the0or1regularizer, the limitation of this approach on its computational complexity or sparsityaffects the efficiency or accuracy. As the smoothed0regularizer is much easier to besolved than the0regularizer and has stronger sparsity than the1regularizer. We thenuse the variable splitting augmented Lagrangian algorithm to solve it. Experimental resultson both simulated and real hyperspectral data demonstrate that the proposed SL0SU ismuch more effective and accurate on hyperspectral unmixing than the state-of-the-artSUnSAL method.2. A novel local collaborative sparse unmixing algorithm has been proposed, namelyLCSU algorithm. The collaborative sparse unmixing globally assumes that all pixels in ahyperspectral image share the same active set of endmembers. However, this globalassumption rarely holds as in reality, one endmember is likely to appear in a local regioninstead of the whole scene. Based on this observation, in this paper, we introduce a localassumption to preserve local collaborativity for sparse unmixing. The LCSU algorithm bytaking advantage from the neighborhood information, also holds the global assumption,which subjects to the local one. The experimental results with both simulated and real hyperspectral data sets demonstrate that the proposed LCSU is much more effectivespectral unmixing algorithm for hyperspectral remote sensing imagery in comparison toSUnSAL, CLSUnSAL, and SUnSAL-TV.3. A novel water change detection based on sparse unmixing has been proposed. Dueto traditional water change detection methods have many limitations, it is difficult to detectthe changes of water long time. Hyperspectral remote sensing as an advanced technologyhas many advantages for the water change detection in comparison to traditional methods.In this paper, we use the above proposed sparse unmixing algorithms to solve the mixedpixels of hyperspectral image. Then we get the unmixing results of water body applied tothe water change detection. The experimental results demonstrate that the proposed methodis effective for water change detection...