Retrieval of atmospheric compensation for hyperspectral imagery using data mining techniques

• Online Access: http://hdl.handle.net/2047/D20317901

One of the biggest interferences of hyperspectral remote sensing is the atmosphere, which can degrade the multi- and hyperspectral imageries. The quick atmospheric correction (QUAC) utilizes an in-scene approach, and is significantly faster than physics-based methods, but more approximate. In this research we attempt to use data mining techniques to retrieval the atmospheric corrections. The key QUAC assumption is that the average of diverse endmember reflectance spectra, excluding highly structured materials (e.g., vegetation, shallow water, mud), is always the same. The biggest contribution of this research is to derive comparable results without this key assumption. To this end, we assembled a training set and a validation set of hyperspectral cubes from AVIRIS sensor website. We built an endmember reflectance library retrieved using QUAC and the training cubes. For a testing cube, the gain featured in QUAC is derived using only the fact that the gain is independent of in-scene space coordinates. It is found that the gain derived is comparable with that from QUAC and yields similar reflectance. We also tested an unsupervised machine learning algorithm to compensate the atmospheric effects directly, featuring clustering and classification algorithms.