Information extraction from the GER 63-channel spectrometer data
暂无分享,去创建一个
The unprecedented data volume in the era of NASA's Mission to Planet Earth (MTPE) demands innovative information extraction methods and advanced processing techniques. The neural network techniques, which are intrinsic to distributed parallel processings and have shown promising results in analyzing remotely sensed data, could become the essential tools in the MTPE era. To evaluate the information content of data with higher dimension and the usefulness of neural networks in analyzing them, measurements from the GER 63-channel airborne imaging spectrometer data over Cuprite, Nevada, are used. The data are classified with 3-layer Perceptron of various architectures. It is shown that the neural network can achieve a level of performance similar to conventional methods, without the need for an explicit feature extraction step.
[1] Geoffrey E. Hinton,et al. Learning representations by back-propagating errors , 1986, Nature.
[2] Z. K. Liu,et al. Classification of remotely-sensed image data using artificial neural networks , 1991 .
[3] P. Swain,et al. Neural Network Approaches Versus Statistical Methods In Classification Of Multisource Remote Sensing Data , 1990 .
[4] J. Aczél. Measuring information beyond communication theory—Why some generalized information measures may be useful, others not , 1984 .