André Marçal

Abstract
<div>Unmanned Aerial Vehicles (UAVs) have emerged as suitable platforms for transporting and positioning communications nodes on demand, including Wi-Fi Access Points and cellular Base Stations. This paved the way for the deployment of flying networks capable of temporarily providing wireless connectivity and reinforcing coverage and capacity of existing networks. Several solutions have been proposed for the positioning of UAVs acting as Flying Access Points (FAPs). Yet, the positioning of Flying Communications Relays (FCRs) in charge of forwarding the traffic to/from the Internet has not received equal attention. In addition, state of the art works are focused on optimizing both the flying network performance and the energy-efficiency from the communications point of view, leaving aside a relevant component: the energy spent for the UAV propulsion.</div><div>We propose the Energy and Performance Aware relay Positioning (EPAP) algorithm. EPAP defines target performance-aware Signal-to-Noise Ratio (SNR) values for the wireless links established between the FCR UAV and the FAPs and, based on that, computes the trajectory to be completed by the FCR UAV so that the energy spent for the UAV propulsion is minimized. EPAP was evaluated in terms of both the flying network performance and the FCR UAV endurance, considering multiple networking scenarios. Simulation results show gains up to 25% in the FCR UAV endurance, while not compromising the Quality of Service offered by the flying network.</div>

Abstract
Machine learning, a subfield of artificial intelligence, has been widely used to automate tasks usually performed by humans. Some applications of these techniques are understanding network traffic behavior, predicting it, classifying it, fixing its faults, identifying malware applications, and preventing deliberate attacks. The goal of this work is to use machine learning algorithms to classify, in separate procedures, the errors of the network, their causes, and possible fixes. Our application case considers the WiBACK wireless system, from which we also obtained the data logs used to produce this paper. WiBACK is a collection of software and hardware with auto-configuration and self-management capabilities, designed to reduce CAPEX and OPEX costs. A principal components analysis is performed, followed by the application of decision trees, k nearest neighbors, and support vector machines. A comparison between the results obtained by the algorithms trained with the original data sets, balanced data sets, and the principal components data is performed. We achieve weighted F1-score between 0.93 and 0.99 with the balanced data, 0.88 and 0.96 with the original unbalanced data, and 0.81 and 0.89 with the Principal Components Analysis. © 2020, Springer Nature Switzerland AG.

Abstract
The extraction of accurate geometric measurements from images normally requires the use of metric cameras and stereoscopic observations. However, as good-quality digital cameras are widely available in mobile devices (smartphones, tablets), there is great interest to develop alternative approaches, suitable for these devices. This paper presents a methodology to compute the surface area and volume of a spheroid-shaped object, such as many types of fruit, from a single image acquired by a standard (non-metric) camera and a basic calibration target. An iterative process is used to establish a 3D spheroid out of the observed 2D ellipse, after which auxiliary images of height, resolution and surface area of each pixel on the 3D object surface are created. The method was tested with a data set of 2400 images, of 10 different objects, 2 calibration targets, 2 cameras and 2 mark types. The average relative errors (< d>) in establishing the 3D object semi-diameters were 0.863% and 0.791%. The semi-diameters are used to compute the object’s surface area (< d> = 1.557%) and volume (< d> = 2.365%). The estimation of the sub-region (mark) surface area over the modelled 3D object resulted in < d> = 2.985%, much lower that what is obtained ignoring the fact that the mark is not on the reference (calibration) plane (< d> = 50.7%), thus proving the effectiveness of the proposed iterative process to model the 3D object (spheroid). © 2020, Springer Nature Singapore Pte Ltd.

Abstract
An Automatic Calibration of Fertilizers (ACFert) system was developed, for use with centrifugal, pendulum or other types of broadcast spreaders which distribute dry granular agricultural materials on the top of the soil. The ACfert is based on image processing techniques and includes a specially designed mat, which should be placed in the ground for spreaders calibration. A set of images acquired outdoor by a standard device (simple camera) is used to extract information about the spreader distribution pattern. Each image is processed independently, providing as output two numerical values for each grid element present in the image - the number of fertilizers/seeds counted, and its numerical label. The performance of ACFert was evaluated for automatic granules detection using a set of manual counting measurements of nitrate fertilizer and wheat seeds. A total of 185 images acquired with two mobiles devices were used with a total of 498 quadrilateral elements observed and analysed. The overall mean absolute relative error between counting and computed by the ACFert system, were 0.75 +/- 0.75% for fertilizer and 2.12 +/- 1.68% for wheat. This near real-time calibration tool is a very low cost system that can be easily used on field, providing results to support accurate spreader calibration in near real time for different types of fertilizers or seeds.

Abstract
The purpose of this work is to address the imageCLEF 2019 coral challenge - to develop a system for the detection and identification of substrates in coral images. Initially a revision of the 13 classes was carried out by identifying a number of sub-classes for some substrates. Four features were considered - 3 related to greyscale intensity (1) and texture (2), and 1 related to the colour content. The Breiman's Random forest algorithm was used to classify the corals in one of 13 classes defined. A classification accuracy of about 49% was obtained.