ROI-Based On-Board Compression for Hyperspectral Remote Sensing Images on GPU

Abstract

In recent years, hyperspectral sensors for Earth remote sensing have become very popular. Such systems are able to provide the user with images having both spectral and spatial information. The current hyperspectral spaceborne sensors are able to capture large areas with increased spatial and spectral resolution. For this reason, the volume of acquired data needs to be reduced on board in order to avoid a low orbital duty cycle due to limited storage space. Recently, literature has focused the attention on efficient ways for on-board data compression. This topic is a challenging task due to the difficult environment (outer space) and due to the limited time, power and computing resources. Often, the hardware properties of Graphic Processing Units (GPU) have been adopted to reduce the processing time using parallel computing. The current work proposes a framework for on-board operation on a GPU, using NVIDIA’s CUDA (Compute Unified Device Architecture) architecture. The algorithm aims at performing on-board compression using the target’s related strategy. In detail, the main operations are: the automatic recognition of land cover types or detection of events in near real time in regions of interest (this is a user related choice) with an unsupervised classifier; the compression of specific regions with space-variant different bit rates including Principal Component Analysis (PCA), wavelet and arithmetic coding; and data volume management to the Ground Station. Experiments are provided using a real dataset taken from an AVIRIS (Airborne Visible/Infrared Imaging Spectrometer) airborne sensor in a harbor area.

A Systematic Review of Hardware-Accelerated Compression of Remotely Sensed Hyperspectral Images

“…This is a drawback since the degradation of the image quality is not caused by an external factor but by the model itself [ 62 ]. Also, the Mean Square Error (MSE) and Root MSE (RMSE) metrics are employed in [ 46 , 49 , 51 , 55 , 63 , 64 , 65 , 66 ] while the Normalized MSE (NMSE) is used in [ 56 , 67 , 68 , 69 ]. The use of normalization facilitates comparison between different datasets.…”

Comprehensive review of hyperspectral image compression algorithms

“…AGU coder is applied at the last level of compression. Giordano and Guccione 16 proposed a combination of clustering and transformation for compression of HSI implemented on graphical processing unit (GPU). It is a region of interest (ROI)-based compression method that clusters the input image into five application-specific classes with an assumption that reflectance value of pixels is preloaded into memory.…”

“…Some acclaimed machine learning algorithms applicable in the technique are SVM, 13 artificial neural network (ANN), 83 backpropagation network, 84 CNN, 31 independent component analysis (ICA)/PCA, 85 and clustering algorithms. 16 Technique.…”

“…Reconstruction is obtained by LZMA decoder followed by residual learning to train the CNN at decoder side, which reverses downsampling to regain the original image. 3D-DCT-SVM, 13 PCA-DCT, 17 folded PCA, 18 WPCA, 19 and Giordano and Guccione 16 discussed in transformation method is an example of use of machine learning in compression. Similarly, VQPCA 57 from VQ method and CSDL_JP2 67 from compressive sensing can also be categorized under learning-based compression.…”

A Runtime-Scalable and Hardware-Accelerated Approach to On-Board Linear Unmixing of Hyperspectral Images

“…Hence, no experimental data or performance metrics under space constraints can be found in those works. However, it is also possible to find solutions throughout the literature using GPUs or FPGAs showing potential real-time results [8,9,[29][30][31].…”