IMPROVED ROAD NETWORK EXTRACTION BY SUPER-RESOLUTION MAPPING OF HYPERION IMAGE DATA

摘要

This paper is concerned with the development of super resolution mapping approaches to extract the road network from a moderate resolution hyperspectral (Hyperion) image data. Super-resolution mapping is a technique which allows classification and mapping from coarse resolution images at the sub-pixel scale. In this work, the potential of Linear Spectral Unmixing and Support Vector Machine as inputs for Hopfield Neural Network based super resolution mapping. Further, integration of SVM as a soft-classified input to the SRM proved to be better than linear spectral unmixing (LSU) as soft classification input. The probability image obtained from support vector machine is used as an input for super resolution mapping designed on Hopfield Neural Network (HNN). For the evaluation of the technique, road segments are mapped using the Hyperion image (30m resolution and 242 bands) of Pondicherry city, south India. Producer's accuracy of the road class is 97.26% and user's accuracy is 91.03%. SRM with SVM as an input has achieved completeness of 0.594 and correctness of 0.863. The results proved that SVM as a soft classified input to the HNN based SRM is efficient and it results in overall accuracy of 92.97% for road network extraction from 30 m resolution image.