Mohammad Elahi
High-Resolution UAV-Based Vegetation classes Mapping Using Data-Centric Machine Learning Under Extreme Label Scarcity.
Rel. Francesca Matrone, Vittorio Fra, Alessandra Spadaro. Politecnico di Torino, Corso di laurea magistrale in Digital Skills For Sustainable Societal Transitions, 2026
|
Preview |
PDF (Tesi_di_laurea)
- Tesi
Licenza: Creative Commons Attribution Non-commercial No Derivatives. Download (5MB) | Preview |
Abstract
Abstract Accurate vegetation classification from very high-resolution UAV imagery remains a fundamental challenge in environmental monitoring and the management of aquatic ecosystems. Although multispectral UAV data provide detailed spectral information, factors such as spectral overlap among vegetation species, intra-class structural heterogeneity, and the severe scarcity of reference data particularly under pixel-based approaches can significantly limit classification reliability. These challenges are especially pronounced in complex alpine environments, where vegetation patterns are highly heterogeneous and dynamic. This study develops a pixel-based machine learning framework for vegetation classification using multispectral UAV imagery acquired over an alpine lake environment. The proposed workflow follows a data-centric approach, including preprocessing and vegetation masking, feature engineering based on spectral bands and vegetation indices, and supervised classification using a gradient boosting model.
To enhance the descriptive power of the input data, first-order statistical texture features derived from local neighborhoods are incorporated into the feature space, complementing the spectral representation with spatial information
Relatori
Anno Accademico
Tipo di pubblicazione
Numero di pagine
Corso di laurea
Classe di laurea
Aziende collaboratrici
URI
![]() |
Modifica (riservato agli operatori) |
