High-Level Project Summary
Our solution is aimed at providing real-time masking of hyperspectral images in the spectral domain. This application was chosen as it is low-power method for rapidly reducing the size of hyperspectral images on orbit, allowing roughly two orders of magnitude reduction in the data file size, depending on application and sensor. We achieved this by reimagining a hyperspectral imager as a reprogrammable multispectral imager, outputting a smaller number of bands that can be calibrated to serve the application in mind rather then generating excess data. The simplicity of the algorithm is a feature as it can be run in real-time on a SpaceEdge on-board computer.
Link to Final Project
Link to Project "Demo"
Detailed Project Description
Hyperband used NASA hyperspectral data downloaded from Earthdata. This data was converted from images in the x and y domains into images in the x and wavelength domains. This more accurately portrays the images as they would be collected by a pushbroom sensor. This was done in Matlab as the interface makes it easier to visualize the data each step
Once the data was properly formatted it was transferred to USB to be accessed on Raspberry Pi to more accurately represent the hardware available on a SpaceEdge OBC. The algorithm to reduce the data was written in python.
Processing time was key, the idea is to quickly and efficiently handle large hyperspectral data sets. It was discovered that the USB was just as fast as the SD card the Raspbian OS was running from. Once the algorithm was ready it was run over the entire dataset and timed.
Space Agency Data
This project used Earth-Observing 1 data from the Hyperion imaging spectrometer. While any data could have been used, the data captured on 1/1/2016 over California (EO1H0410362016001110K7) was of particular interest because the satellite unintentionally captured a methane leak in progress that was at the time undetected. Using a spectral satellites to monitor for these leaks inspired me to work with hyperspectral earth observation years ago and remains an unused resource due to the cost of hyperspectral satellite imaging.
Hackathon Journey
This experience helped me connect a lot of disconnected, underdeveloped skills I have gathered over the years into one project. Converting the data into spatial-spectral slices and increasing the algorithm's speed became the major time sinks in this process as with most projects making it work was simple, making it work well was a headache. I'd like to thank Spiral Blue and Taofiq Huq for pushing me to do this project, it was fun to make something again.
References
Hyperion data, including reference RGB images where accessed through Earthdata: https://search.earthdata.nasa.gov/search
Tags
#earthobservation #hyperspectral #spiralblue #edgecomputing