| dc.description.abstract | This report details development, verification and results for implementation of JPEG 2000 compression on a Zynq-7000~\cite{zynq7000} system-on-chip. Software/hardware co-development of the system is intended for on-board satellite processing of hyper-spectral image cubes and is focused on lossless compression.
The NTNU Small Satellite Lab is an initiative to bring activities related to space to the university, including the Hyperspectral Imager (HSI)-mission which purpose is to observe oceanographic phenomena by using a hyperspectral camera onboard a cube satellite. The prototype payload electronics include a high sensitivity, CMOS image sensor as well as a PCB-mounted system-on-module enabling onboard processing.
Limited data transfer windows and a limited bandwidth necessitate this onboard processing to ensure as much useful data as possible can be transferred to the base station. Digital image compression is thus a way of processing the raw image data into lesser storage sizes to allow more data to be transferred for reconstruction on ground. JPEG 2000 is the standard chosen for this project because of its superior compression ratios compared to JPEG, and its lossless and error-resilient properties.
Software/hardware co-design is employed by implementing processing software for the system-on-chip's CPU with hardware accelerators for intermediate processing steps. Development efforts are comprised of four main parts: a) interchangeable general-purpose hardware processing modules, b) a hardware implementation of a multidimensional discrete wavelet transform module, b) software entropy coding, d) a system-level integration of the processing system's and programmable logic's interaction and sharing of memory.
Benchmarking of compression performance has been performed on data sets of hyperspectral images to optimize compression ratios and hardware-specific parameters. The benchmarks also provide comparable performance metrics between the JPEG 2000 and the ESA CCSDS 123 standards, on image cubes in their entirety and dimensionally reduced through principal component analysis.
Discrete wavelet transforms are implemented in hardware through high-level synthesis and model verification, and ensures bit-correct conformity with the standard compared with reference software and MATLAB models. The module can perform uninterrupted processing at frequencies up to 312.2 MHz.
With the EBCOT-based, bit plane entropy coder done in software, compression of an entire 2024x1088 frequency band is performed within a worst of 6.5 seconds. | en |
