Ica Aided Linear Spectral Mixture Analysis of Agricultural Remote Sensing Images

In this paper, we propose a new method of estimating pure spectra and the mixture ratio by applying the Independent Component Analysis (ICA) to the agricultural remote sensing images for recognizing fine structure vegetation change on farmland, where the covering plant is unknown. This technique enables to separate the change of vegetation into qualitative one due to ecological characteristics such as the chlorophyll quantity, and the quantitative coverage one. In the area of remote sensing several attempts using the ICA have been reported. These methods have defined the spectral reflectance pattern in the wavelength domain, as the independent component (IC), in order to extract pure spectra or only spectral features for the classification. In these cases, it is necessary to provide sufficient spectral bands to ensure the independence of each IC, as, for example, with hyperspectral images. In our technique, we define the periodical spatial distribution of crops along the farmland position as the IC, so that pure spectra of crops are estimated as the mixture ratio of the IC, the coverage, unlike the conventional ones. To the simulated mixed spectra, we demonstrated that this technique is useful even when the mixed spectra include vegetation covering fluctuation, additive noise such as thermal noise from the sensor and atmospheric noise, in the real data are involved. In addition, by interpreting the coverage as the IC, it is possible to reduce the number of spectral bands. This means that our method can be applied not only to hyperspectral images but also multispectral images.