A Novel Machine Learning System for Autonomous Classification of High-Resolution

NASA has a critical need to automate the identification and classification of features in Earth Science imagery, particularly new high-resolution imagery. In addition, commercial applications of Earth Science data as well as the viability of the remote sensing industry depend on the development of new tools for accelerated classification of remotely sensed images. Previous research by others to automate analysis of remote sensing data through development of feature extraction software has met with limited success. The proposed research involves a new automated feature extraction approach that uses innovative machine-learning algorithms and techniques, including neural network ensembles and foveal vision. Visual Learning Systems (VLS) proposes to demonstrate the technical feasibility of this approach for rapid extraction of features from a wide variety of high-resolution digital images. This Phase I research is expected to result in technology advances that will successfully demonstrate the feasibility of VLS's innovative machine-learning system for automated feature classification from diverse high-resolution multispectral, hyperspectral, and radar image data sets from leading satellite and airborne sensors. These results will establish a solid foundation for further research during Phase II that will lead to a fully operational automated feature classification system ready for subsequent commercial development.