Due to diastrophism, tide action and human activities, the coastline is always in flux. There are lots of coral islands in the south sea of China. Remote sensing survey for the coastline not only can reassert the necessity and importance of coral protection, but also can provide basic data and scientific basis for island ecologic protection, reasonable utilization of land resources. The study area named Zhao Shu Island lies in Jintong Islands of Xisha. It is a coral island which has people inhabited. Using WorldView-2 satellite remote sensing images as data sources we carry out three phases of coastline investigation and monitoring. The satellite data phases are 2002, 2010 and 2013. Firstly, affirm the bands valuable for color composition on the basis of spectral and correlation analysis. Then extract the coastline by a series of image process, such as image correction, fusion, waterline extraction and coastline revision. Finally determine the coastline types and length by artificial interpretation. The results show that the island length is gradually smaller, which means the island area is reducing. The beach bedrock coast in northern island was eroded seriously especially during the period between 2010 and 2013. In addition, the shoal head shape in the western island changed a lot.
As hyperspectral remote sensing image have the features of high spectral resolution, high dimension, it brings a serious problem of choosing the appropriate band combination from numerous bands. To solve the problem，the present paper, basing on Optimum Index Factor, proposes a modified band selection algorithm. It aims to reduce the dimension through the improved block adaptive band selection. According to the strong correlation among the bands of hyperspectral remote sensing image，the paper introduces hyperspectral band selection method based on the correlation between bands. The modified algorithm can overcome some shortages of Optimum Index Factor, such as inefficient of selecting bands. However, it can also remain some characteristics from Optimum Index Factor. For instance, the selected band combination, through the modified algorithm, can retain rich amount of information and weak correlation between the selected bands. And the compute complication is decreased rapidly. Taking Hyperion hyperspectral remote sensing image from EarthObserving-1 as example, the present paper contrast the practical effect of the modified algorithm of sub regional selection band and Optimum Index Factor.