Hyperspectral Image Of Wetland Unmixing Based On Sparse Regression Algorithm

Wetland,As one of the major ecosystems in the world,wetland enjoys the reputation of "kidney of the earth",which is the trend of hyperspectral research in recent years.Hyperspectral image has a lot of continuous and narrow spectral bands,and each pixel can display continuous spectral curve,so it has been developed rapidly.Compared with traditional remote sensing technology,hyperspectral image can detect more ground objects from the perspective of ground object spectrum.However,due to its high spectral resolution and limited spatial resolution,a pixel often contains many types of ground objects,so the existence of mixed pixels is the main factor affecting area measurement and target detection.Spectral unmixing algorithm based on sparse regression is a new method in the field of hyperspectral mixed pixel decomposition.Compared with the endmember classes with a large number of spectral libraries,the endmember classes involved in hyperspectral mixed pixels are usually smaller,which involves linear regression based on sparsity.The optimal subset of spectral features can be found in the spectral library which is most likely to contain endmembers,and the spectral library can be used for each image Mixed pixels are used for optimal modeling.The main work and innovation of this paper include the following aspects.1.Combine the digital spectrum database published by USGS with the end member spectrum obtained from Dagu River Wetland image survey to construct the end member spectrum database.Vertex component analysis(VCA),maximum volume method(N-FINDR)and minimum volume model(SISAL)were used to extract pure pixels from wetland images.Through field survey and spectral analysis,the endmember species in Dagu River wetland were determined.2.An improved hyperspectral mixed pixel decomposition algorithm based on pseudo endmember constraint(Endmember Choice of Collaborative Sparse Unmixing via variable Splitting and Augmented Lagrangian,EC-SUNSAL)is proposed.As an improved algorithm based on Cooperative sparse unmixing(CLSUNSAL),the purpose of this algorithm is to improve the accuracy of abundance inversion after unmixing.Firstly,signal subspace recognition algorithm is used to obtain the number of endmembers in wetland image.Traditional clsunsal is used for preprocessing,and the constraints of real endmember number and pseudo endmember abundance are added.In the obtained high-dimensional abundance matrix,threshold analysis is used to select endmembers,and the pseudo endmembers are removed to construct a new spectral library A1.Finally,the endmembers are unmixed based on clsunsal framework to improve the abundance The accuracy of the inversion.The visual quality evaluation and signal reconstruction error(SRE)are used as the evaluation indexes to verify the simulation data,and the abundance sparsity(So A)value and visual quality evaluation are used as the evaluation indexes of the real hyperspectral data set to verify the accuracy of the algorithm.The results show that the ec-sunsal model can get the visual effect image and quantitative index data after abundance inversion,which is better than the traditional sparse model The algorithm has been greatly improved.3.The collaborative sparse unmixing algorithm based on pseudo endmember constraint is used to unmixing the hyperspectral Dagu River Wetland image.Seven types of surface features are obtained,and their abundance maps are obtained.The distribution characteristics and abundance information of each wetland endmember are analyzed,and the Dagu River image is mapped.The traditional sparse regression unmixing algorithm and the collaborative sparse unmixing algorithm based on pseudo endmember constraint are compared in the wetland image data,and the accuracy is verified by the So A index.The results show that compared with the classification results of the traditional sparse regression unmixing algorithm,the collaborative sparse unmixing algorithm based on pseudo endmember constraint has higher correlation with the wetland feature types obtained from field survey To verify the effectiveness of ec-clsunsal algorithm in wetland abundance inversion.