Learning from weakly representative data and applications in spectral image analysis

摘要: Spectral imaging has been extensively applied in many fields, including agriculture, environmental monitoring, biomedical diagnostics, etc. Thanks to the advances sensor technology, spectral systems nowadays provide finer and resolution needed characterize properties of materials. The high resolution, however, raises an issue as difference information between two adjacent wavelength bands is typically very small. As a result, much data scene seems be redundant. However, critical embedded that often can used identify This thesis aims at facilitating analysis by making use pattern recognition techniques, on one hand, improve visualization, other directly solve classification problems.