Compressive sensing techniques can greatly reduce the complexity of imaging instruments used in space applications, by reducing sampling requirements and power consumptions without sacrificing image resolution. In this paper, we present the compressive imaging technique developed within the EU project "SURPRISE". A suitable CS model adapted to the optical layout of the instrument enables image reconstruction at a resolution higher than the sensing elements. Both classical CS reconstruction techniques and deep learning-based methods are investigated, showing that the latter achieves a better quality of the reconstructed images for different compression ratios. Interestingly, deep learning-based techniques also reduce the complexity of CS reconstruction, since most of the computations are performed in the training phase that can be executed offline.
COMPRESSIVE IMAGING AND DEEP LEARNING BASED IMAGE RECONSTRUCTION METHODS IN THE "SURPRISE" EU PROJECT
Donatella Guzzi;Cinzia Lastri;Vanni Nardino;Lorenzo Palombi;Valentina Raimondi;
2021
Abstract
Compressive sensing techniques can greatly reduce the complexity of imaging instruments used in space applications, by reducing sampling requirements and power consumptions without sacrificing image resolution. In this paper, we present the compressive imaging technique developed within the EU project "SURPRISE". A suitable CS model adapted to the optical layout of the instrument enables image reconstruction at a resolution higher than the sensing elements. Both classical CS reconstruction techniques and deep learning-based methods are investigated, showing that the latter achieves a better quality of the reconstructed images for different compression ratios. Interestingly, deep learning-based techniques also reduce the complexity of CS reconstruction, since most of the computations are performed in the training phase that can be executed offline.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.