Geology & Environmental Science Colloquium: Dr. Joe Ortiz | Kent State University

Applications of multi- and hyperspectral image decomposition in visible near infrared remote sensing Presented by Joseph D. Ortiz | Professor, Kent State University Department of Earth Sciences Talk Abstract: The mixed pixel problem – the averaging of sensor information over the sensor’s ground sampling distance and instantaneous field of view – provides a fundamental challenge for optical remote sensing which limits the quality of classification results. A traditional approach to dealing with this problem has been to reduce the pixel size, which has been beneficial given the coarse resolution of early sensors and the large inherent spatial scales of terrestrial scenes, which are dominated by macroscopic targets, such as crops and tree canopy. This approach has been less helpful in aquatic systems, where the remote sensing targets (algae and suspended sediment) are microscopic in size. Over the past few decades, optical sensors have made great technological leaps forward resulting in smaller pixel size and rapidly increasing band count. Those technological changes have also resulted in rapid increases in the size of remote data sets. The transition from low band count multispectral to high band count hyperspectral sensors however carried with it the introduction of significant quantities of redundant information. Multicollinearity – the intercorrelation of visible and near infrared bands – is one of the primary sources of uncertainty in optical remote sensing. The focus of my research is to explore how numerical methods can be used to address the mixed pixel problem and multicollinearity in VNIR remote sensing. The result of that work has been the development of varimax rotated principal component analysis (VPCA) in conjunction with visible derivative spectroscopy to address the mixed-pixel and multicollinearity problems. An added benefit to VPCA is that it also serves as a data-adaptive filtering method, reducing redundant data volume to a smaller number of i