Design of a Wavelet-Based Hyperspectral Compression System

by
Daniel Griffie
Northeast Senior High School
in Pasadena, Maryland.

Our ability to collect information of the earth's surface is astounding. With today's remote sensing technology, it is possible for an unmanned aerial vehicle to fly over a site and take extremely precise information at hundreds of different spectral wavelengths. This data, termed hyperspectral imagery, may consist of hundreds of such spectral bands per image. This represents a tremendous research tool for scientists, but an equally tremendous burden on storage capacity. One hyperspectral image may exhaust anywhere from 50 to 200mb of storage space. My project is to design and implement a system to efficiently compress this data using a wavelet transform basis.

Certainly there is nothing new about wavelet image compression, but the goal here is to develop a system specifically for images taken by aerial remote sensing imagers. These images are meant for research purposes and therefore possess features that set them apart from aesthetic images. For sure, the reconstruction fidelity of hyperspectral is far more important because of the precision necessary for scientific data.

The design focuses around three components, a transform to represent the image as translations and dilations of wavelet functions, a quantizer to convert wavelet coefficients into small integers, and an entropy encoder to take advantage of redundancies and rewrite the information as its theoretic minimum. Each area has its own unique problems associated with multidimensional images.

Walt Houser CPCUG Coordinator.

CPCUG Science Fair Home Page