Fusing Multimodal CNN Features with Structurally-Masked Transformers for Hierarchical Sub-Pixel Tree Species Mapping

Authors: Mattia Ferrari, Lorenzo Bruzzone•Year: 2026

Journal: IEEE Transactions on Geoscience and Remote Sensing

Under Review for Pubblication

Sub-Pixel Forest Classification Fusing PRISMA Hyperspectral and DEM Data with a Hierarchy-Aware Multi-Branch CNN

Authors: Mattia Ferrari, Lorenzo Bruzzone•Year: 2026

Conference: IEEE International Geoscience and Remote Sensing Symposium (IGARSS 2026)

Upcoming presentation

Detection of Bark Beetle Attacks using Hyperspectral PRISMA Data and Few-Shot Learning

Authors: Mattia Ferrari, Giancarlo Papitto, Giorgio Deligios, Lorenzo Bruzzone•Year: 2025

Conference: IEEE International Geoscience and Remote Sensing Symposium (IGARSS 2025)

Novel approach using few-shot learning for detecting bark beetle attacks on hyperspectral data.

Hyperspectral data augmentation with transformer-based diffusion models

Authors: Mattia Ferrari, Lorenzo Bruzzone•Year: 2024

Conference: Artificial Intelligence and Image and Signal Processing for Remote Sensing XXX (SPIE)

Data augmentation techniques for hyperspectral imagery using transformer-based diffusion models.

Dimensionality Reduction Techniques for Hyperspectral Data

Role: Co-Supervisor•Year: 2025

This thesis evaluates different feature reduction techniques on hyperspectral satellite data to mitigate the curse of dimensionality and optimize deep learning models for forest classification.

Detection of Bark Beetle Attacks Using Hyperspectral Data and Self-Supervised Learning

Role: Co-Supervisor•Year: 2025

This thesis develops a self-supervised deep learning framework using Masked Autoencoders and hyperspectral satellite imagery to accurately detect and monitor bark beetle outbreaks in spruce forests.

Identification of Forest Changes Through the Analysis of Multispectral and Hyperspectral Images

Role: Co-Supervisor•Year: 2024

This thesis explores the use of multispectral and hyperspectral satellite remote sensing combined with machine learning to map forest damage and monitor recovery following the 2018 Storm Vaia in Northern Italy.

Data Augmentation Methods for the Classification of Remote Sensing Image Time Series with Recurrent Neural Networks

Role: Student (Author)•Year: 2022

This thesis evaluates various data augmentation techniques for remote sensing time series to mitigate labeled data scarcity and improve crop classification accuracy using Sentinel-2 imagery and LSTM neural networks.

Monitoring Agricultural Areas with Sentinel-2 Remote Sensing Image Time Series Using LSTM Recurrent Neural Networks

Role: Student (Author)•Year: 2019

This thesis presents an automated approach for agricultural land classification using Sentinel-2 multispectral imagery and LSTM neural networks trained on data generated from existing national thematic maps.