A Biologically and Geometrically Inspired Approach to Target Extraction from Multiple-source Remote-sensing Imagery

This paper presents the research results on the integration approach using a biologically inspired algorithm (LEGION) and a geometrically-inspired method (GAC) for target extraction from multiple-source remote-sensing imageries, specifically EO-1 Hyperion hyperspectral (30-meter resolution), and IKONOS multispectral (4-meter resolution) images. An automatic road-extraction algorithm based on LEGION (Locally Excitatory Globally Inhibitory Oscillator Networks, a neurocomputational framework for image segmentation) was developed to extract main roads from EO-1 Hyperion imagery. A region-based geometric/geodesic active contour (GAC) which adopts Euclidean distance as the basic energy metric was used to perform target extraction from both the EO-1 Hyperion and IKONOS multispectral images. The candidate targets including roads detected on the EO-1 imagery were projected onto the IKONOS imagery and used as prior knowledge for target extraction. Experimental results show that this approach reduced the computational complexity on the IKONOS imagery. Also, the use of the LEGIONbased road-extraction algorithm increased the probability that major roads would be distinguished from other objects that are made of similar materials.