High Spectral Band Selection And Application Inquiry For The Observation Of The Multi-mode

Compared with the traditional multi-spectral remote sensing technology, hyperspectral remote sensing can obtain rich information from a continuous narrow band, which greatly enhancing the ability to identify and classify surface features fine. However, due to the high number of hyperspectral data, there is a big-band redundancy, resulting in the classification accuracy increases rapidly saturation as the characteristic dimension expand, also along with the characteristic dimension expand there is Hughes phenomenon which decrease after the first increase, namely the emergence of the phenomenon of the curse of dimensionality. On the one hand, to solve this problem hyperspectral, it is very necessary to retain the original information in hyperspectral data, while reducing the amount of data, reduce data dimensionality. Explore a powerful band preferred algorithm is a major issue of this paper. On the other hand, after the existing satellite payload developed and launched its spectral and spatial observation mode is fixed and the mode cannot be optimization and adjustment based on the diverse needs of complex surface, and the current band remote sensor setting is not perfect yet exist to optimize space. In this paper, preferred bands from hyperspectral data which is different type surface features in different area, By analyzing the spectrum of preferred bands combinations,proposed band setting scheme for Optimization of earth observation satellite sensors.This model makes the changes that satellite sensors observed pattern toward application-oriented, real-time flexible shift pattern. Also make the earth observation model "what give to" change to the "for what you want," thereby increasing the efficiency of remote sensing imagery and data utilization efficiency.The main contents and results are as follows:Firstly, researching and analysing several traditional existing hyperspectral band selection algorithm, including entropy and joint entropy, the the covariance matrix determinant of band combination, Optimum Index Factor(OIF), etc.Found that traditional optimal bands method have specific problems in programming languages and have limitations for our study.The same time, an improved algorithm based on the traditional is proposed which is band combination covariance matrix determinant based on subspace division. Applying the improved algorithm preferred high-band spectrum, from the running speed and the preferred result both have been greatly improved, thus proving that the algorithm is an excellent improvement algorithms.Secondly, the paper introduced ACO(Ant colony optimization, ACO) algorithm to carry out band preferred research for CASI + SASI hyperspectral data, While do the same study introduce band combination covariance matrix determinant based on subspace division algorithm in the same area. And evaluate classification result obtained using two algorithms preferred band combination, Obtained ACO have more advantages for band selection hyperspectral data. Preferred band combination can fully express most of the information of the data, and classification accuracy is improved, fully demonstrates the advantages of ant colony optimization algorithm in terms of processing hyperspectral data.Third,using ant colony optimization algorithm to carry out optimal band combination of different surface types, and found the best band combination for each type of surface there are some differences. For each type of surface observations have specific band combination program, the results of this research provide important support of a priori knowledge for band optimization of intelligent observations.Finally, to carry out band set optimization study the typical multi-spectral satellite load,the results show some marginal band, such as the red edge band, dark blue bandand so on for earth observation plays an important role,in the earth observation sensors should increase these spectral settings.And get the set optimization for typical multispectral satellite, evaluate existing satellite sensors, such as TM, OLI, SPOT through optimization program,the results show that the fixed sensor mode has some limitations for widely observed object,there is big space to optimal adjustment for existing sensors.