The intersection of Unmanned Aerial Vehicles (UAVs), commonly known as drones, with various human labor-intensive activities such as last-mile delivery, agriculture, and surveillance has emerged as a compelling frontier of research. This dissertation explores the transformative impact of drones in two key domains: last-mile delivery and agriculture. In the context of last-mile delivery, drones are rapidly reshaping logistics practices, offering opportunities to overcome geographical barriers, reduce delivery times, and minimize carbon footprints. The cost-effectiveness and adaptability of drones make them accessible to businesses and communities. However, this transition to drone-based delivery systems is not without its challenges. The first part of the dissertation focuses on last-mile delivery, exploring innovative solutions to optimize operational efficiency and enhance sustainability. Chapter 1 investigates the impact of wind dynamics on drone-based delivery systems defining a tunable model for representing drone energy consumption according to wind conditions. Then, an algorithm to plan minimum-energy trajectories, and one to analyze mission feasibility for drones in windy conditions are proposed. Chapter 2 explores the cooperation between a truck and a fleet of drones in the context of last-mile package delivery by introducing the Scheduling Conflictual Deliveries Problem (SCDP). The SCDP models each delivery as an interval, and each drone as channel of limited capacity, i.e., energy, and aims to maximize the collected reward associated to each delivery, i.e., the delivery priority. Finally, optimal and approximation algorithms are provided presented with a thorough performance evaluation on synthetic data. The second part of the dissertation delves into agriculture, specifically crop monitoring, using drones to collect data efficiently from specific locations. The possibility to visualize with constant update the crop health state helps the farmer in the decision process allowing to take actions and measurements promptly towards the root of the problem. By discretizing the observation position inside a crop, the monitoring task ca be seen as a Orienteering Problem (OP) where a reward function shapes the relevance of visiting certain positions, and a cost function represents the energy consumption due to drone operations. Under this framework Chapter 3 presents a solution for the data collection problem in a special orchard type, i.e., modern orchards protected by nets, providing a novel data structure to model it, and introducing the Single-drone Orienteering Aisle-graph Problem (SOAP), a specialized version of the OP. In SOAP, the objective is visiting the most relevant observation points in the orchard for collecting images without exceeding the battery capacity of the drone. For solving it are devised an optimal, two approximation algorithms and two fast heuristics. Chapter 4 follows on the same research line by addressing data collection from wireless sensor networks (WSN) strategically deployed within monitoring areas, introducing the Multiple-drone Data-collection Maximization Problem (MDMP). MDMP represents a further generalization of the OP, where in addition the original constraints is added a novel one, the storage capacity. Therefore, the drone during its data harvesting mission is subjected to the battery capacity, i.e., it has a maximum flight range, and a storage capacity, it can store only a limited amount of data during its mission. The problem is proved to be NP-hard, and for it are proposed an optimal ILP solution, an approximation, and two fast heuristics. The third part of the dissertation focuses on smart agriculture, particularly pest detetction in orchards. The invasive Halyomorpha Halys (in short HH), commonly known as Brown Marmorated Stink Bug, poses a significant threat to agriculture. The dissertation details the development of an Integrated Pest Monitoring (IPM) system, combining drones and machine learning algorithms for pest detection, with a specific emphasis on this invasive bug. Chapter 5 concentrates on improving pest detection using machine learning models trained on field-captured images. To the best of our knowledge, we explore for the first time the possibility to detect this invasive bug employing a computer vision algorithms trained on images taken inside orchard aisles. The chapter covers every phase of the training process, ending with wide performance analysis based on several factors, including variations in blurriness levels. Finally, we devise a cloud service architecture for extending the accessibility of the algorithms devised. Chapter 6 focuses again on the creation of a detection algorithms for scouting the HH, posing a special attention on the automatizion of the entire process, i.e., mission planning, data acquisition, and bug detection. Therefore, we first define a drone-based protocol which allows to collect images from the top of the orchard, at an altitude of 10 meters, pushing on strict camera settings, and photograph techniques. To the best of our knowledge, relying on this protocol we create the first datasets for scouting HH completely based on drone-captured images. Then, employing a tailored training algorithm based on image slicing and sub-portion attention we train several models exploiting the YOLOv5 algorithm. Finally, we analyze performance of the networks and models predictiveness using well-known and custom metrics. Throughout the dissertation, the fusion of technology and agriculture is highlighted, paving the way for a more sustainable and efficient future in farming. The research contributions discussed in each chapter have been published in various conferences and journals, showcasing the practical relevance and impact of the work. Overall, this dissertation addresses critical challenges and provides innovative solutions for the integration of drones into last-mile delivery, agriculture, and pest monitoring, contributing to the ongoing evolution of drone-based systems in these domains.

Optimizing Drone-Based Applications for Delivery and Smart Agriculture / Lorenzo Palazzetti. - (2024).

Optimizing Drone-Based Applications for Delivery and Smart Agriculture

Lorenzo Palazzetti
2024

Abstract

The intersection of Unmanned Aerial Vehicles (UAVs), commonly known as drones, with various human labor-intensive activities such as last-mile delivery, agriculture, and surveillance has emerged as a compelling frontier of research. This dissertation explores the transformative impact of drones in two key domains: last-mile delivery and agriculture. In the context of last-mile delivery, drones are rapidly reshaping logistics practices, offering opportunities to overcome geographical barriers, reduce delivery times, and minimize carbon footprints. The cost-effectiveness and adaptability of drones make them accessible to businesses and communities. However, this transition to drone-based delivery systems is not without its challenges. The first part of the dissertation focuses on last-mile delivery, exploring innovative solutions to optimize operational efficiency and enhance sustainability. Chapter 1 investigates the impact of wind dynamics on drone-based delivery systems defining a tunable model for representing drone energy consumption according to wind conditions. Then, an algorithm to plan minimum-energy trajectories, and one to analyze mission feasibility for drones in windy conditions are proposed. Chapter 2 explores the cooperation between a truck and a fleet of drones in the context of last-mile package delivery by introducing the Scheduling Conflictual Deliveries Problem (SCDP). The SCDP models each delivery as an interval, and each drone as channel of limited capacity, i.e., energy, and aims to maximize the collected reward associated to each delivery, i.e., the delivery priority. Finally, optimal and approximation algorithms are provided presented with a thorough performance evaluation on synthetic data. The second part of the dissertation delves into agriculture, specifically crop monitoring, using drones to collect data efficiently from specific locations. The possibility to visualize with constant update the crop health state helps the farmer in the decision process allowing to take actions and measurements promptly towards the root of the problem. By discretizing the observation position inside a crop, the monitoring task ca be seen as a Orienteering Problem (OP) where a reward function shapes the relevance of visiting certain positions, and a cost function represents the energy consumption due to drone operations. Under this framework Chapter 3 presents a solution for the data collection problem in a special orchard type, i.e., modern orchards protected by nets, providing a novel data structure to model it, and introducing the Single-drone Orienteering Aisle-graph Problem (SOAP), a specialized version of the OP. In SOAP, the objective is visiting the most relevant observation points in the orchard for collecting images without exceeding the battery capacity of the drone. For solving it are devised an optimal, two approximation algorithms and two fast heuristics. Chapter 4 follows on the same research line by addressing data collection from wireless sensor networks (WSN) strategically deployed within monitoring areas, introducing the Multiple-drone Data-collection Maximization Problem (MDMP). MDMP represents a further generalization of the OP, where in addition the original constraints is added a novel one, the storage capacity. Therefore, the drone during its data harvesting mission is subjected to the battery capacity, i.e., it has a maximum flight range, and a storage capacity, it can store only a limited amount of data during its mission. The problem is proved to be NP-hard, and for it are proposed an optimal ILP solution, an approximation, and two fast heuristics. The third part of the dissertation focuses on smart agriculture, particularly pest detetction in orchards. The invasive Halyomorpha Halys (in short HH), commonly known as Brown Marmorated Stink Bug, poses a significant threat to agriculture. The dissertation details the development of an Integrated Pest Monitoring (IPM) system, combining drones and machine learning algorithms for pest detection, with a specific emphasis on this invasive bug. Chapter 5 concentrates on improving pest detection using machine learning models trained on field-captured images. To the best of our knowledge, we explore for the first time the possibility to detect this invasive bug employing a computer vision algorithms trained on images taken inside orchard aisles. The chapter covers every phase of the training process, ending with wide performance analysis based on several factors, including variations in blurriness levels. Finally, we devise a cloud service architecture for extending the accessibility of the algorithms devised. Chapter 6 focuses again on the creation of a detection algorithms for scouting the HH, posing a special attention on the automatizion of the entire process, i.e., mission planning, data acquisition, and bug detection. Therefore, we first define a drone-based protocol which allows to collect images from the top of the orchard, at an altitude of 10 meters, pushing on strict camera settings, and photograph techniques. To the best of our knowledge, relying on this protocol we create the first datasets for scouting HH completely based on drone-captured images. Then, employing a tailored training algorithm based on image slicing and sub-portion attention we train several models exploiting the YOLOv5 algorithm. Finally, we analyze performance of the networks and models predictiveness using well-known and custom metrics. Throughout the dissertation, the fusion of technology and agriculture is highlighted, paving the way for a more sustainable and efficient future in farming. The research contributions discussed in each chapter have been published in various conferences and journals, showcasing the practical relevance and impact of the work. Overall, this dissertation addresses critical challenges and provides innovative solutions for the integration of drones into last-mile delivery, agriculture, and pest monitoring, contributing to the ongoing evolution of drone-based systems in these domains.
2024
Maria Cristina Pinotti
ITALIA
Goal 9: Industry, Innovation, and Infrastructure
Goal 11: Sustainable cities and communities
Goal 12: Responsible consumption and production
Lorenzo Palazzetti
File in questo prodotto:
File Dimensione Formato  
Phd_Thesis_signed.pdf

accesso aperto

Tipologia: Tesi di dottorato
Licenza: Open Access
Dimensione 12.3 MB
Formato Adobe PDF
12.3 MB Adobe PDF

I documenti in FLORE sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificatore per citare o creare un link a questa risorsa: https://hdl.handle.net/2158/1356051
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact