Search

Deep learning methods have traditionally been difficult to apply to compression of hyperspectral images onboard spacecrafts due to the large computational complexity needed to achieve adequate representational power, as well as the lack of suitable datasets for training and testing. In this article, we depart from the traditional autoencoder approach, and we design a predictive neural network, called line receptance weighted key value (LineRWKV), which works recursively line by line to limit memory consumption. In order to achieve that, we adopt a novel hybrid attentive-recursive operation that combines the representational advantages of Transformers with the linear complexity and recursive implementation of recurrent neural networks (RNNs). The compression algorithm performs the prediction of each pixel using LineRWKV, followed by entropy coding of the residual. Experiments on multiple datasets show that LineRWKV is highly memory-efficient, significantly outperforms state-of-the-art deep learning methods, and is the first deep learning approach to outperform CCSDS-123.0-B-2 at lossless and near-lossless compression. Promising throughput results are also evaluated on a 7-W embedded system.

Images have always been considered a reliable source of evidence in the past. Today, the wide availability of photo editing software urges the authors to investigate the origin and the integrity of a digital image before trusting it. Although several algorithms have been developed for image integrity verification, a comprehensive tool that allows the analyst to synergically exploit these algorithms, and to reach a final decision based on their output, is still lacking. In this work the authors propose an image forensic tool trying to fill this gap. The proposed tool exploits state of the art algorithms for splicing detection, with forgery localization capabilities, and make them available to the analyst through a graphical interface. In order to help the analyst in reaching a final assessment, a decision fusion engine is employed to intelligently merge the output of different algorithms, boosting detection performance. The tool has a modular architecture, that makes it easily scalable.

Two hundred twenty-seven patients (median age 5.4 months) in whom a Waterston anastomosis was done for initial palliation of tetralogy of Fallot between 1966 and 1979 were studied. Twelve patients died in the hospital (5.3%; 70% confidence limits, 3.8% to 7.3%). Young age, low weight, and poor clinical condition did not appear to be incremental risk factors, whereas a too large or a too small shunt was largely responsible for the hospital mortality and morbidity. Follow-up information was available in all the 215 patients discharged from the hospital. At the last follow-up visit, before any further surgical procedure, 74% of the patients were clinically in good condition. By actuarial methods, 97.7% of hospital survivors were alive and 95.8% were event-free at and beyond 3 years postoperatively. Eighty-six patients have been catheterized in preparation for secondary repair (mean interval between Waterston shunt and catheterization, 2.9 ± 1.38 years). One patient developed pulmonary vascular disease, four acquired pulmonary atresia, and 14 had a severe kinking of the right pulmonary artery at the site of the anastomosis.

Sixty-three corrective operations have been performed in patients with Tetralogy of Fallot and a previously constructed Waterston shunt. The patients were from 1.4 to 8 years of age (median 4.3 years), and the mean interval between the Waterston shunt and the repair was 39.3 ± 16.05 months. Three patients developed pulmonary atresia and six required a second anastomosis prior to intracardiac repair. Kinking and stenosis of the right pulmonary artery at the site of the Waterston anastomosis occurred in 12 (19 percent; 70 percent confidence limits 14 to 26 percent) patients who required patch enlargement of the right pulmonary artery at repair. Seven (11 percent; 70 percent confidence limits 7 to 17 percent) patients died in the hospital and three (5 percent; 70 percent confidence limits 2 to 11 percent) during the period of late follow-up. None of the early or late deaths was specifically related to the presence of the previously performed Waterston anastomosis.