Abstract
Magnetic Resonance Imaging (MRI) exams suffer from undesirable structure replicating and overlapping effects on certain acquisition settings. These are called Spatial Aliasing Artefacts (SAA) and their presence interferes with the segmentation of other anatomical structures. This paper addresses the segmentation of the SAA in T1-weighted MRI image sets, in order to effectively remove their influence over the legitimately positioned body structures. The proposed method comprises an initial thresholding, employing the Triangle method, an aggregation of neighboring voxels through Region Growing. Further refinement of the objects contour is obtained with Convex Hull and a Minimum Path algorithm applied to two orthogonal planes (Sagittal and Axial). Some experiments concerning the extension of the pipeline used are reported and the results seem promising. The average contour distance concerning the Ground Truth (GT) rounds 2.5mm and area metrics point out average overlaps above 64% with the GT. Some issues concerning the fusion between the output from the two planes are to be perfected. Nevertheless, the results seems sufficient to neutralize the influence of SAA and expedite the downstream anatomical segmentation tasks.

Abstract
Localized plasmons in metallic nanostructures present strong analogies with Quantum Mechanical problems of particles trapped in potential wells. In this paper we take this analogy further using the Madelung Formalism of Quantum Mechanics to express the fluid equations describing the charge density of the conduction electrons and corresponding interaction with light in terms of an effective generalized Non-linear Schrodinger equations. Within this context, it is possible to develop the analogy of a plasmonic atom and molecule that exhibits Rabi oscillations, Stark effect, among other Quantum Mechanical effects.

Abstract

Abstract
Traditional coverage path planners create lawnmower-type paths in the operating area completely ignoring the uncertainty in the vehicle's position. However, in the presence of significant uncertainty in localization estimates, one can no longer guarantee that the vehicle will cover all the area according to plan. Aiming to bridge this gap, we present a coverage path planning technique for search operations which takes into account the vehicle's position and detection performance uncertainties and tries to minimize this uncertainty along the planned path. The objective is to plan paths, using a localization error model as input, to reduce as much uncertainty as possible and to minimize the extra path length (swath overlap) while satisfying mission feasibility constraints. We introduce an algorithm that calculates what will be the best moments for bringing the vehicle to surface to ensure a bounded position error. We also consider time and energy constraints that may influence the planned trajectory as path overlap is increased to account for uncertainty. Additionally we challenge the assumption frequently seen in coverage algorithms where two observations of the same target are considered independent. © 2017 IEEE.

Abstract
This study explored the viability of out-the-box, pre-trained ConvNet models as a tool to generate features for large-scale classification tasks. A juxtaposition with generative methods for vocabulary generation was drawn. Both methods were chosen in an attempt to integrate other datasets (transfer learning) and unlabelled data, respectively. Both methods were used together, studying the viability of a ConvNet model to estimate category labels of unlabelled images. All experiments pertaining to this study were carried out over a two-class set, later expanded into a 5-category dataset. The pre-trained models used were obtained from the Caffe Model Zoo. The study showed that the pre-trained model achieved best results for the binary dataset, with an accuracy of 0.945. However, for the 5-class dataset, generative vocabularies outperformed the ConvNet (0.91 vs. 0.861). Furthermore, when replacing labelled images with unlabelled ones during training, acceptable accuracy scores were obtained (as high as 0.903). Additionally, it was observed that linear kernels perform particularly well when utilized with generative models. This was especially relevant when compared to ConvNets, which require days of training even when utilizing multiple GPUs for computations.

Abstract