In this study, the problem of computing a sparse representation of multi-dimensional visual data is considered. In general, such data e.g., hyperspectral images, color images or video data consists of signals that exhibit strong local dependencies. A new computationally efficient sparse coding optimization problem is derived by employing regularization terms that are adapted to the properties of the signals of interest. Exploiting the merits of the learnable regularization techniques, a neural network is employed to act as structure prior and reveal the underlying signal dependencies. To solve the optimization problem Deep unrolling and Deep equilibrium based algorithms are developed, forming highly interpretable and concise deep-learning-based architectures, that process the input dataset in a block-by-block fashion. Extensive simulation results, in the context of hyperspectral image denoising, are provided, which demonstrate that the proposed algorithms outperform significantly other sparse coding approaches and exhibit superior performance against recent state-of-the-art deep-learning-based denoising models. In a wider perspective, our work provides a unique bridge between a classic approach, that is the sparse representation theory, and modern representation tools that are based on deep learning modeling.
Satellite communication channels can be well described as non-linear functions with memory. The Volterra series representation provides accurate modeling of satellite channel dynamics, and thus, it constitutes a widely used approach to mathematically describe them. In this work, iterative correction of the non-linear distortion introduced by such channels is considered, by employing a soft interference canceller operating in a turbo equalization framework.
Localization of an isotropic acoustic source using energy measurements from distributed sensors is considered. While most acoustic source localization algorithms require that distance estimates between the sensors and the source of interest are available, we propose a linear least squares criterion that does not make such an assumption. The new criterion can yield the location of the source and its transmit power in closed form. A weighted least squares cost function is also considered, and distributed implementation of the proposed estimators is studied. Numerical results indicate significant performance improvement as compared to a linear least squares based approach that utilizes energy ratios, and comparable performance to other estimators of higher computational complexity.
This paper presents a computer-aided diagnosis scheme for the detection of prostate cancer. The pattern recognition scheme proposed, utilizes fused dynamic and morphological features extracted from magnetic resonance images (MRIs). The performance of the proposed scheme has been evaluated through extensive training and testing on several patient cases, where the staging of their condition has been previously evaluated by both ultrasoundguided biopsy and radiological assessment. The classification scheme is based on Probabilistic Neural Networks (PNNs), whose parameters are estimated using the Expectation-Maximization (EM) algorithm during a training phase. Fusion of the image characteristics is performed by properly aligning the respective T1-weighted dynamic and T2-weighted morphological images, allowing accurate feature selection from both images. The proposed classification scheme as well as the effect of fusion on the extracted features is tested, with respect to the correct classification rate (CCR) of each case.