Research And Implementation Of Hyperspectral Remote Sensing Image Compression Algorithm Based On Redundant Dictionary

Hyperspectral remote sensing technology can obtain the image of high spatial resolution and high spectral resolution by using imaging spectrometer, which is playing an important role in national development. However, the huge amount of data of hyperspectral image has brought tremendous pressure to the data storage and transmission, which restrict the application and development of hyperspectral imaging technology seriously. It will pay a high price via improving the corresponding hardware conditions to storage and process data. Therefore, the method of mathematics and digital signal processing combining is an efficient compression method for hyperspectral image to enhance the hyperspectral remote sensing image compression and storage efficiency significantly.In this context, after the study of the characteristics of hyperspectral image and the practical demand of the compression application, the research of hyperspectral image compression coding based on redundant dictionary was carried out., the main work as follows:(1) We designed and researched the theory of sparse representation, and realized the hyperspectral image sparse decomposition and the training of redundant dictionary.By designing an optimal redundant dictionary in order to achieve the purpose, which represented the hyperspectral images using the typical elements in the dictionary linearly. In this paper, the dictionary training method is K-SVD algorithm, we trained and obtained a redundant dictionary through this algorithm for hyperspectral images. The experiments’ results showed that hyperspectral images can be represented by the redundant dictionary effectively.(2) A hyperspectral image compression algorithm which is based on redundant dictionary was designed and implemented.According to the principle of sparse decomposition and data feature, a new compression algorithm was designed and the experimental results and analysis were given. Hyperspectral image after sparse decomposition would get a redundant dictionary and corresponding sparse decomposition coefficient. If we compress the coefficients and the dictionary, we can compress the hyperspectral image. In this paper, the coefficients are compressed after the method of quantization and JPEG-LS, which can be used to restore the image information well and get a larger compression ratio.(3) A new hyperspectral image compression algorithm which is based on PCA and redundant dictionary was proposed.After the experimental results and analysis of the algorithm, we use the method of principal component analysis(PCA) and remove the noise in the image and secondary information before the sparse decomposition preprocessing of hyperspectral images, retained the image information only and then sparse decomposition and compression images. Compared with the experimental results of JPEG2000, in the case of the same compression ratio, the PSNR value of the proposed algorithm is higher about 1.2dB.