Hyperspectral datasets are massive. This size makes them difficult to acquire, store, transmit, analyze and use. Hyperspectral imagers (HSI) are also costly and complex, taking significant time or portions of a focal plane to acquire each spectral band. In addition, orbital HSIs are susceptible to radiation damage and inaccessible for repairs, making any damage long-lasting.
In this effort, we will be using advanced algorithms for compressive sensing using inpainting and machine learning to develop a compressive sensing enhanced HSI system. Our Phase I results have shown that we can achieve a 1% subsampling rate with maintaining high quality imagery with minimal loss. This new system allows for over an order of magnitude improvement in the frame acquisition speed of existing detectors at a high compression ratio. The approach will automatically compress all images obtained by the level of sub-sampling used to form the image, reducing transmission and storage costs significantly, even without using additional lossless data compression before transmission.
In order to demonstrate the ability of these methods to provide these benefits, Sivananthan Laboratories will apply them by fabricating a new CS-enhanced HSI system. This leverages the wide range of IR detectors that the company, and its sister companies EPIR and Episensors, have pioneered over the last 20 years. We propose to develop new hyperspectral sensors that will intentionally subsampled data that will be recovered using computational imaging with learned representations. By multiplexing the input signals, and randomly selection spectral bands using an acousto -optical tunable filter we can leverage subsampling directly into higher signal-to-noise levels, and better resolution.
This technology can be applied to both existing hyperspectral imaging systems and those being developed. The approach can be applied to both division of time and division of space systems, maintaining resolution while reducing capture time and noise. Additional applications are available in robust data compression for storage and transmission over long distances. This approach also provides fault tolerance, system recovery and image enhancement for fielded imagers with lost pixels.
Commercialization will come in two forms: CS enhancement of existing technologies and through selling complete imaging systems as built in functionality. Hyperspectral imaging systems are used in many applications including chemical and biological detection, manufacturing, environmental surveys of CO2, pollution, and leakages.