Session
Weekend Poster Session 1
Location
Utah State University, Logan, UT
Abstract
Many space-sensing platforms are capable of producing vast amounts of sensor data. While having additional data can further science goals, it also introduces the problems of storage and transfer. Space platforms have limited storage resources and transmission bandwidth. As such, data compression is a vital step to make the most efficient use of onboard resources. Sensors such as multispectral imagers can capture many times the amount of data of traditional visual sensors, posing difficulties for onboard storage and processing of the data. Spectral CNNJPEG is a state-of-the-art adaptive compression algorithm for multispectral and hyperspectral-imaging sensors. This algorithm has been previously demonstrated to provide excellent compression with high reconstruction quality. However, to determine the aptitude of this algorithm for space use, it must be tested under realistic use conditions. This research provides a performance assessment of the Spectral CNNJPEG compression system using the STP-H7-CASPR (Configurable Autonomous Sensor Processing Research) payload operating aboard the ISS. This algorithm leverages the onboard GPU computing environment as well as data captured by the Satlantis iSIM-90 multispectral imager aboard STP-H7-CASPR. The performance of the onboard compression system is evaluated in several categories, including total runtime, inference latency, compression power, and reconstruction quality. This research provides a flight demonstration of the Spectral CNNJPEG compression system. With Spectral CNNJPEG, we achieve compression ratios of nearly 20× while maintaining high structural similarity greater than 0.9. The compression ratio and quality can also be tuned depending on the desired level of compression. By employing Spectral CNNJPEG adaptive compression on CASPR, we can expand the boundaries of what is possible for onboard computing.
Evaluation of Spectral CNNJPEG Adaptive Compression on the STP-H7-CASPR Platform
Utah State University, Logan, UT
Many space-sensing platforms are capable of producing vast amounts of sensor data. While having additional data can further science goals, it also introduces the problems of storage and transfer. Space platforms have limited storage resources and transmission bandwidth. As such, data compression is a vital step to make the most efficient use of onboard resources. Sensors such as multispectral imagers can capture many times the amount of data of traditional visual sensors, posing difficulties for onboard storage and processing of the data. Spectral CNNJPEG is a state-of-the-art adaptive compression algorithm for multispectral and hyperspectral-imaging sensors. This algorithm has been previously demonstrated to provide excellent compression with high reconstruction quality. However, to determine the aptitude of this algorithm for space use, it must be tested under realistic use conditions. This research provides a performance assessment of the Spectral CNNJPEG compression system using the STP-H7-CASPR (Configurable Autonomous Sensor Processing Research) payload operating aboard the ISS. This algorithm leverages the onboard GPU computing environment as well as data captured by the Satlantis iSIM-90 multispectral imager aboard STP-H7-CASPR. The performance of the onboard compression system is evaluated in several categories, including total runtime, inference latency, compression power, and reconstruction quality. This research provides a flight demonstration of the Spectral CNNJPEG compression system. With Spectral CNNJPEG, we achieve compression ratios of nearly 20× while maintaining high structural similarity greater than 0.9. The compression ratio and quality can also be tuned depending on the desired level of compression. By employing Spectral CNNJPEG adaptive compression on CASPR, we can expand the boundaries of what is possible for onboard computing.
