MACHINE LEARNING - XXXVIII cycle - Machine Learning for Automated Recognition of Seabed Morphologies

Aug 4, 2026

Upload by Aurelia

Seabed environments contain diverse morphological features that document geological and geomorphological processes along continental margins. Their systematic recognition is essential for marine geohazard assessment and for designing offshore infrastructures, yet conventional bathymetric mapping depends largely on manual expert interpretation, making results slow, subjective, and hard to reproduce consistently over large areas. This thesis investigates machine learning methods to automatically recognize seabed morphological elements from high-resolution bathymetric derivatives. A U-Net-based convolutional approach is trained on MaGIC Project data, incorporating preprocessing, data balancing, and a bidirectional neighborhood metric tailored for elongated and discontinuous structures, improving segmentation quality while supporting reproducible geological interpretation.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages110

Reading Time4 h 37 min

AccessFree to download

Summary

Introduction

Motivations and Challenges

Research Questions

Structure of the Thesis

Publications

Context and Background

Background and Related Works

Seabed Data

Seabed Morphological Elements

Machine Learning Techniques in Marine Geology

Problem Statement and Experimental Framework

Problem Statement and Dataset

Methodology

Experiments and Results

Main multi-class model

Simplified model configurations

Discussion

Limitations

Conclusions and Open Research Questions

Conclusions

Open Research Directions

UNIVERSITÀ DEGLI STUDI DI TRIESTE XXXVIII CICLO DEL DOTTORATO DI RICERCA IN APPLIED DATA SCIENCE AND ARTIFICIAL INTELLIGENCE Istituto Nazionale di Oceanografia e di Geofisica Sperimentale (OGS) MACHINE LEARNING FOR AUTOMATED RECOGNITION OF SEABED MORPHOLOGIES Settore scientifico-disciplinare: INF/01 DOTTORANDO UMBERTO DI LAUDO COORDINATORE PROF. FRANCESCO PAULI SUPERVISORE DI TESI PROF. LUCA MANZONI SUPERVISORE DI TESI DOTT.SSA SILVIA CERAMICOLA ANNO ACCADEMICO 2024/2025 U!"\#$%&"’( )$*+" S’,)" )" T%"$&’$ Ph.D. in Applied Data Science & Arti!cial Intelligence XXXVIII cycle Machine Learning for Automated Recognition of Seabed Morphologies Candidate Umberto Di Laudo Supervisors Prof. Luca Manzoni Dr. Silvia Ceramicola Summary The seabed hosts a wide variety of morphological features that record the geological and geomorphological processes shaping continental margins. Their systematic recognition is fundamental for marine geohazard assessment and for the planning of o!shore infrastructures. However, conventional mapping relies heavily on manual interpretation of bathymetric data, which is time-consuming, subjective, and di"cult to reproduce consistently across large areas. This thesis explores the use of machine learning for the automated recognition of seabed morphological elements from high-resolution bathymetric derivatives. A convolutional architecture based on U-Net is developed and applied to adataset from the MaGIC Project, covering selected regions of the Italian continental margins. The proposed framework includes dedicated preprocessing, data balancing, and abidirectional neighborhood-based metric speci\#cally designed to evaluate elongated and discontinuous structures that are poorly captured by standard pixel-wise scores. Three model con\#gurations are tested, multi-class (reduced-class, and binary) to analyze how class granularity a!ects performance and interpretability. Overall, the proposed framework bridges quantitative image segmentation with geological interpretation, demonstrating that deep learning can e!ectively support and accelerate seabed mapping, while preserving consistency and reproducibility across extensive marine domains. Contents Summary i " Introduction " $ . $ Motivations and Challenges ......................... $ $ . % Research Questions .............................. % $ .& Structure of the Thesis ............................ & $ .’ Publications .................................. ’ I Context and Background \# $ Background and Related Works % % . $ Seabed Data .................................. ( % . % Seabed Morphological Elements ....................... ) % .& Machine Learning Techniques in Marine Geology ............. $* % .& . $ From statistical to machine learning approaches .......... $$ % .& . % Object-Based Image Analysis (OBIA) ................ $$ % .& .& Deep learning for sea+oor characterization ............ $% % .& .’ Towards the detection of linear features .............. $& % .’ Deep Learning Foundations for Image Analysis .............. $& % .’. $ Convolutional Neural Networks ................... $& % .’. % Image Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . $, % .’.& U-Net architecture . . . . . . . . . . . . . . . . . . . . . . . . . . %& II Problem Statement and Experimental Framework $% & Problem Statement and Dataset $’ & . $ Problem De\#nition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . %) & . % Dataset Description and Analysis . . . . . . . . . . . . . . . . . . . . . . &* ( Methodology &’ ’. $ U-net-based Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . &) ’. % Pipeline +owchart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ’$ ’. % . $ Input preprocessing ......................... ’$ ’. % . % Training ................................ ’& ’. % .& Post processing ............................ ’- ’

FAQs

Why is automated recognition of seabed morphologies important?

Systematic recognition supports marine geohazard assessment and offshore infrastructure planning. Manual mapping is slow, subjective, and difficult to reproduce consistently over large regions.

What modeling approach does the thesis propose?

The thesis develops a convolutional architecture based on U-Net for image segmentation of seabed morphological elements from high-resolution bathymetric derivatives.

How does the framework evaluate structures that are hard for standard metrics?

It includes preprocessing, data balancing, and a bidirectional neighborhood-based metric specifically designed to better capture elongated and discontinuous structures that are poorly handled by pixel-wise scores.

Aug 4, 2026

Upload by Aurelia