Object Detection from Hs/ms and Multi-platform Remote- Sensing Imagery by the Integration of Biologically and Geometrically Inspired Approaches

This paper presents a system that integrates biologically and geometrically inspired approaches to detecting objects from hyperspectral and/or multispectral (HS/MS), multiscale, multiplatform imagery. First, dimensionality reduction methods are studied and used for hyperspectral dimensionality reduction. Then, a biologically inspired method, SLEGION (Spatial Locally Excitatory Globally Inhibitory Oscillator Network), is developed to perform object detection on the multispectral and dimension-reduced hyperspectral data, which provides rough object shapes. Thereafter, a geometrically inspired method, GAC (geometric active contour), is employed for refining object boundary detection on the high-resolution imagery based upon the initial object shapes provided by S-LEGION. A geospatial database is compiled and used for experimental analysis that includes data from a selected test site at Silver Lake in the Mojave Desert, California. Multispectral (Landsat TM 4-5) and hyperspectral (EO-1) satellite imagery, high-resolution satellite imagery (IKONOS), and descent images and ground stereo images are included in this database. This paper presents the first year results of a two-year research project.