Tesi etd-11262025-182033 |
Link copiato negli appunti
Tipo di tesi
Tesi di laurea magistrale
Autore
PANDOCCHI, GAIA ANNA ARIELE
URN
etd-11262025-182033
Titolo
Seasonal distribution of Caretta caretta in the Gulf of Manfredonia through aerial survey and machine learning for object detection
Dipartimento
BIOLOGIA
Corso di studi
CONSERVAZIONE ED EVOLUZIONE
Relatori
relatore Prof. Casale, Paolo
relatore Agabiti, Chiara
relatore Agabiti, Chiara
Parole chiave
- Caretta caretta
- deep learning
- distribuzione stagionale
- seasonal distribution
- UAV
- YOLOv8
Data inizio appello
15/12/2025
Consultabilità
Non consultabile
Data di rilascio
15/12/2065
Riassunto
Il Golfo di Manfredonia, nel Sud Adriatico, è un’importante area di foraggiamento per la tartaruga caretta (Caretta caretta). È una regione fortemente antropizzata che presenta tra i più alti livelli di bycatch nel Mediterraneo. Poiché i dati disponibili sulla densità delle tartarughe a mare sono limitati (i dati di abbondanza sono prevalentemente rappresentati da numero di nidi o numero di femmine sulle spiagge riproduttive), il monitoraggio ecologico su scala fine è utile per lo sviluppo di misure di conservazione adeguate a ridurre l’impatto della pesca. A questo proposito, il primo obiettivo di questa tesi era stimare la densità superficiale delle tartarughe, invernale ed estiva, utilizzando Unmanned Aerial Vehicles (UAV), e compararne i valori. Sono stati monitorati tre siti lungo la costa del Golfo di Manfredonia nell’inverno 2023 e nell’estate 2024, volando simultaneamente su due transetti speculari per aumentare il numero di voli e l’area campionata al giorno. Dato il tempo necessario per revisionare manualmente le registrazioni dei voli, il secondo obiettivo della tesi era sviluppare un modello di machine learning per object detection in grado di rilevare autonomamente le tartarughe. Da precedenti campionamenti aerei condotti con droni nell’Arcipelago delle Isole Pelagie sono stati estratti i frame contenenti tartarughe, revisionati da un operatore, e utilizzati per allenare un modello YOLOv8 della Ultralytics. La sua performance di rilevazione è stata confrontata con l’analisi manuale dei video e le densità stagionali finali sono state calcolate integrando i dati manuali e del modello.
The Gulf of Manfredonia, in the South Adriatic, is a key foraging area for the loggerhead turtle (Caretta caretta). This region is highly anthropized and experiences some of the highest bycatch rates in the Mediterranean. Since the data available on turtle density at sea are limited (abundance data are represented mainly by the number of nests or the number of females on reproductive shores), fine-scale ecological monitoring is useful for the development of adequate conservation measures for reducing the impact of fishing. In this concern, the first objective of this thesis was to estimate the turtle surface density during winter and summer using Unmanned Aerial Vehicles (UAVs) and compare the values to each other. Three sites along the coast of the Gulf of Manfredonia were monitored during winter 2023 and summer 2024, flying simultaneously on two specular transects to increase both the number of flights and the area surveyed per day. Given the time-consuming nature of manually reviewing the recordings of flights, the second goal of this thesis was to develop a machine learning model for object detection able to autonomously detect turtles. A YOLOv8 model by Ultralytics was trained on previous aerial surveys conducted with drones in the Pelagian Island Archipelago and used to extract only frames containing turtles, further human reviewed. Its detection performance was compared to manual video analysis and the final seasonal density estimates were derived by integrating manual and automated data.
The Gulf of Manfredonia, in the South Adriatic, is a key foraging area for the loggerhead turtle (Caretta caretta). This region is highly anthropized and experiences some of the highest bycatch rates in the Mediterranean. Since the data available on turtle density at sea are limited (abundance data are represented mainly by the number of nests or the number of females on reproductive shores), fine-scale ecological monitoring is useful for the development of adequate conservation measures for reducing the impact of fishing. In this concern, the first objective of this thesis was to estimate the turtle surface density during winter and summer using Unmanned Aerial Vehicles (UAVs) and compare the values to each other. Three sites along the coast of the Gulf of Manfredonia were monitored during winter 2023 and summer 2024, flying simultaneously on two specular transects to increase both the number of flights and the area surveyed per day. Given the time-consuming nature of manually reviewing the recordings of flights, the second goal of this thesis was to develop a machine learning model for object detection able to autonomously detect turtles. A YOLOv8 model by Ultralytics was trained on previous aerial surveys conducted with drones in the Pelagian Island Archipelago and used to extract only frames containing turtles, further human reviewed. Its detection performance was compared to manual video analysis and the final seasonal density estimates were derived by integrating manual and automated data.
File
| Nome file | Dimensione |
|---|---|
La tesi non è consultabile. |
|