基于光谱库优化学习的光谱超分与分类精度提升

The spectral library can contain the spectral information on the whole types of ground surface objects in the observation area of hyperspectral images. Thus, the optimized dictionary learning via spectral library refers to the process of constructing optimized spectral dictionary under strict theoretical derivation, in which the spectra in the spectral library are used as training samples. The abovementioned process enables the spectra in the hyperspectral image to be sparsely represented under the learned spectral dictionary. To this end, a new spectral super-resolution method using optimized dictionary learning via spectral library is proposed in this study. This method uses only one high spatial multispectral image to reconstruct high spatial hyperspectral image. The aforementioned problem is formulated in the framework of sparse representation, as an estimation of the band matching matrix, the optimized spectral dictionary, and the corresponding sparse coefficients. Specifically, a band matching method is proposed to map the common spectral library to a specific spectral library corresponding to the reconstructed high spatial hyperspectral image. Then, an optimization of spectral dictionary and its corresponding sparse coefficients is derived theoretically using the alternating direction method of multipliers (ADMM) algorithm and by utilizing the abovementioned specific spectral library and the high spatial multispectral image. Comparison results with the relative methods demonstrate that our method not only can achieve a high-quality reconstruction of the high spatial hyperspectral image but also can significantly improve the classification accuracy of multispectral images by even only using one high spatial multispectral image. We aim to reconstruct high spatial hyperspectral image only from one high spatial multispectral image with high quality. Three steps of our proposed method are discussed in detail. First, the band matching matrix is estimated using the band wavelength information. Second, the matched spectral dictionary is optimized using the matched spectral library and the high spatial multispectral image. Third, the equivalent sparse coefficient matrix with respect to the matched spectral dictionary is derived theoretically and estimated iteratively. Extensive experiments and comparative analyses of the proposed method are conducted on various datasets to demonstrate the performance and practical application value of our proposed method. The improvement in classification accuracy on the reconstructed high spatial hyperspectral images is also evaluated using some typical classification methods. A spectral super-resolution method is proposed, and it uses only one high spatial multispectral image to reconstruct high spatial hyperspectral image. A band matching matrix, which is used to map the common spectral library to a specific spectral library, is obtained by solving the minimum distance problem. A spectral dictionary and its corresponding sparse coefficient matrix are optimized from the matched spectral library and the high spatial hyperspectral image by minimizing augmented Lagrangian function using ADMM iteratively. Experiments on simulated and real datasets demonstrate that our proposed method can produce comparable results for the spectral superresolution to the other relative state-of-the-art reconstruction or fusion-based methods using additional low spatial hyperspectral image. It can also provide higher reconstruction quality than the HIRSL method without optimization. Our proposed SODL method that uses only one multispectral image may help develop new light and small high spatial hyperspectral imaging equipment.