Metodi di tipo level set per la segmentazione di immagini di nevi

La tesi in oggetto propone un algoritmo che viene incontro alla necessità di segmentare in modo regolare immagini di nevi. Si è fatto uso di metodi level set region-based in una formulazione variazionale. Tale metodo ha permesso di ottenere una segmentazione precisa, adattabile a immagini di nevi con caratteristiche molto diverse ed in tempi computazionali molto bassi.

Studio e sviluppo di un metodo di detezione della camera atriale in ecografia intracardiaca

Il lavoro di tesi si è svolto in collaborazione con il laboratorio di elettrofisiologia, Unità Operativa di Cardiologia, Dipartimento Cardiovascolare, dell’ospedale “S. Maria delle Croci” di Ravenna, Azienda Unità Sanitaria Locale della Romagna, ed ha come obiettivo lo sviluppo di un metodo per l’individuazione dell’atrio sinistro in sequenze di immagini ecografiche intracardiache acquisite durante procedure di ablazione cardiaca transcatetere per il trattamento della fibrillazione atriale. La localizzazione della parete posteriore dell'atrio sinistro in immagini ecocardiografiche intracardiache risulta fondamentale qualora si voglia monitorare la posizione dell'esofago rispetto alla parete stessa per ridurre il rischio di formazione della fistola atrio esofagea. Le immagini derivanti da ecografia intracardiaca sono state acquisite durante la procedura di ablazione cardiaca ed esportate direttamente dall’ecografo in formato Audio Video Interleave (AVI). L’estrazione dei singoli frames è stata eseguita implementando un apposito programma in Matlab, ottenendo così il set di dati su cui implementare il metodo di individuazione della parete atriale. A causa dell’eccessivo rumore presente in alcuni set di dati all’interno della camera atriale, sono stati sviluppati due differenti metodi per il tracciamento automatico del contorno della parete dell’atrio sinistro. Il primo, utilizzato per le immagini più “pulite”, si basa sull’utilizzo del modello Chan-Vese, un metodo di segmentazione level-set region-based, mentre il secondo, efficace in presenza di rumore, sfrutta il metodo di clustering K-means. Entrambi i metodi prevedono l’individuazione automatica dell’atrio, senza che il clinico fornisca informazioni in merito alla posizione dello stesso, e l’utilizzo di operatori morfologici per l’eliminazione di regioni spurie. I risultati così ottenuti sono stati valutati qualitativamente, sovrapponendo il contorno individuato all'immagine ecografica e valutando la bontà del tracciamento. Inoltre per due set di dati, segmentati con i due diversi metodi, è stata eseguita una valutazione quantitativa confrontatoli con il risultato del tracciamento manuale eseguito dal clinico.

A framework for cyber-physical system simulation

In these last years, systems engineering has became one of the major research domains. The complexity of systems has increased constantly and nowadays Cyber-Physical Systems (CPS) are a category of particular interest: these, are systems composed by a cyber part (computer-based algorithms) that monitor and control some physical processes. Their development and simulation are both complex due to the importance of the interaction between the cyber and the physical entities: there are a lot of models written in different languages that need to exchange information among each other. Normally people use an orchestrator that takes care of the simulation of the models and the exchange of informations. This orchestrator is developed manually and this is a tedious and long work. Our proposition is to achieve to generate the orchestrator automatically through the use of Co-Modeling, i.e. by modeling the coordination. Before achieving this ultimate goal, it is important to understand the mechanisms and de facto standards that could be used in a co-modeling framework. So, I studied the use of a technology employed for co-simulation in the industry: FMI. In order to better understand the FMI standard, I realized an automatic export, in the FMI format, of the models realized in an existing software for discrete modeling: TimeSquare. I also developed a simple physical model in the existing open source openmodelica tool. Later, I started to understand how works an orchestrator, developing a simple one: this will be useful in future to generate an orchestrator automatically.

Methodologies for attitude determination from MEMS sensors in vehicles

The focus of the thesis is the application of different attitude’s determination algorithms on data evaluated with MEMS sensor using a board provided by University of Bologna. MEMS sensors are a very cheap options to obtain acceleration, and angular velocity. The use of magnetometers based on Hall effect can provide further data. The disadvantage is that they have a lot of noise and drift which can affects the results. The different algorithms that have been used are: pitch and roll from accelerometer, yaw from magnetometer, attitude from gyroscope, TRIAD, QUEST, Magdwick, Mahony, Extended Kalman filter, Kalman GPS aided INS. In this work the algorithms have been rewritten to fit perfectly with the data provided from the MEMS sensor. The data collected by the board are acceleration on the three axis, angular velocity on the three axis, magnetic fields on the three axis, and latitude, longitude, and altitude from the GPS. Several tests and comparisons have been carried out installing the electric board on different vehicles operating in the air and on ground. The conclusion that can be drawn from this study is that the Magdwich filter is the best trade-off between computational capabilities required and results obtained. If attitude angles are obtained from accelerometers, gyroscopes, and magnetometer, inconsistent data are obtained for cases where high vibrations levels are noticed. On the other hand, Kalman filter based algorithms requires a high computational burden. TRIAD and QUEST algorithms doesn’t perform as well as filters.

Geometric versus Model Predictive Control based guidance algorithms for fixed-wing UAVs in the presence of very strong wind fields.

The recent years have witnessed increased development of small, autonomous fixed-wing Unmanned Aerial Vehicles (UAVs). In order to unlock widespread applicability of these platforms, they need to be capable of operating under a variety of environmental conditions. Due to their small size, low weight, and low speeds, they require the capability of coping with wind speeds that are approaching or even faster than the nominal airspeed. In this thesis, a nonlinear-geometric guidance strategy is presented, addressing this problem. More broadly, a methodology is proposed for the high-level control of non-holonomic unicycle-like vehicles in the presence of strong flowfields (e.g. winds, underwater currents) which may outreach the maximum vehicle speed. The proposed strategy guarantees convergence to a safe and stable vehicle configuration with respect to the flowfield, while preserving some tracking performance with respect to the target path. As an alternative approach, an algorithm based on Model Predictive Control (MPC) is developed, and a comparison between advantages and disadvantages of both approaches is drawn. Evaluations in simulations and a challenging real-world flight experiment in very windy conditions confirm the feasibility of the proposed guidance approach.

A Distributed Support based on Kubeflow to Ease MLOps of Federated Learning Algorithms

The main objective of my thesis work is to exploit the Google native and open-source platform Kubeflow, specifically using Kubeflow pipelines, to execute a Federated Learning scalable ML process in a 5G-like and simplified test architecture hosting a Kubernetes cluster and apply the largely adopted FedAVG algorithm and FedProx its optimization empowered by the ML platform ‘s abilities to ease the development and production cycle of this specific FL process. FL algorithms are more are and more promising and adopted both in Cloud application development and 5G communication enhancement through data coming from the monitoring of the underlying telco infrastructure and execution of training and data aggregation at edge nodes to optimize the global model of the algorithm ( that could be used for example for resource provisioning to reach an agreed QoS for the underlying network slice) and after a study and a research over the available papers and scientific articles related to FL with the help of the CTTC that suggests me to study and use Kubeflow to bear the algorithm we found out that this approach for the whole FL cycle deployment was not documented and may be interesting to investigate more in depth. This study may lead to prove the efficiency of the Kubeflow platform itself for this need of development of new FL algorithms that will support new Applications and especially test the FedAVG algorithm performances in a simulated client to cloud communication using a MNIST dataset for FL as benchmark.