Ship As Wave Buoy: Data-Driven Sea State Estimation Based on Ship Motion Data/Судно как волновой буй: оценка состояния моря на основе данных о движении судна

Издание на английском языке
This book is devoted to modern methods for assessing sea state using vessel motion data using deep learning. It examines basic neural network architectures, models with attention mechanisms, multi-scale and prototype-based learning, approaches to working with imbalanced data, and methods for transferring knowledge between vessels. Special attention is paid to the robustness of models to noise, missing values, and disturbances, the use of graph neural networks, experimental validation of algorithms, and the potential for their use in real-time assessment of the maritime situation.

Content
1 Introduction
1.1. Background
1.2. State-of-the-Art
1.2.1. Traditional Methods
1.2.2. Model-Based Methods
1.2.3. Data-Driven Methods
1.3. Outline of This Book
References
Part I. Foundational Deep Architectures for Ship Motion Data Based SSE
2. Multi-domain Deep Representation Learning for SSE
2.1. Introduction
2.2. Methodology
2.2.1. Overview
2.2.2. Data Preprocessing
2.2.3. Sensitivity Analysis
2.2.4. Learning the Temporal Dependency
2.2.5. Learning the Local Features
2.2.6. Learning the Frequency Features
2.2.7. Feature Fusion
2.3. Experiments
2.3.1. Datasets
2.3.2. Comparison With SOTA Time Series Classification Methods
2.3.3. Baseline Comparison
2.3.4. Ablation Study
2.3.5. Sensitivity of Data Pre-processing
2.3.6. Real-Time Estimation
2.4. Conclusion
References
3. Dual-Attention CNN Architectures for Multi-parameter SSE
3.1. Introduction
3.2. Methodology
3.2.1. Definition of Multi-parameter Sea State
3.2.2. Overview
3.2.3. Data Processing
3.2.4. Convolutional Block and Dense Connection
3.2.5. Channel Attention Module
3.2.6. Feature Attention Module
3.3. Experiment
3.3.1. Dataset Description
3.3.2. Benchmark Comparison on Public Datasets
3.3.3. Data Analysis and Feature Selection
3.3.4. Baseline Comparison on SSE Dataset
3.3.5. Ablation Study
3.3.6. Comparison with Other Attention Mechanisms
3.3.7. Sensitivity Analysis
3.4. Conclusion
References
4. Parallel Multi-branch CNN Architectures for SSE
4.1. Introduction
4.2. Methodology
4.2.1. Overview
4.2.2. Parallel Convolution Module with Variable Kernels
4.2.3. Distance-Based Classification
4.2.4. Loss Function
4.2.5. Theoretical Analysis
4.3. Experiment
4.3.1. Datasets
4.3.2. Evaluation Metrics
4.3.3. Baseline Comparison on Public Datasets
4.3.4. Baseline Comparison on SSE Task
4.3.5. Ablation and Sensitivity Analysis
4.3.6. Visualization Analysis
4.3.7. Discussion
4.4. Conclusion
References
Part II. Imbalance-Aware Prototype Learning for Ship Motion Data Based SSE
5. Attention-Enhanced Cross-Scale Prototype Learning for Imbalanced SSE
5.1. Introduction
5.2. Methodology
5.2.1. Overview
5.2.2. Multi-scale Feature Learning
5.2.3. Cross-Scale Feature Learning
5.2.4. Relation to Other MSFL and CSFL Module
5.2.5. Prototype-Based Classifier
5.2.6. Loss Function
5.3. Experiments
5.3.1. Bseline Comparison on Public Datasets
5.3.2. Baseline Comparison on SSE Task
5.3.3. Comparison with Other Class Imbalance Learning Approaches
5.3.4. Ablation and Sensitivity Analysis
5.3.5. Discussion
5.4. Conclusion
References
6. Adaptive Selective Kernel Prototype Learning for Imbalanced SSE
6.1. Introduction
6.2. Methodology
6.2.1. Overview
6.2.2. Multiscale Learning
6.2.3. Prototype Classifier
6.2.4. Loss Function
6.3. Experiments
6.3.1. Baseline Comparison on Public Datasets
6.3.2. Baseline Comparison on SSE Task
6.3.3. Comparison with Other Imbalanced-focused Method
6.3.4. Robust to Noise and Missing Values
6.3.5. Ablation and Sensitivity Analysis
6.3.6. Discussion
6.4. Conclusion
References
7. Frequency-Guided Prototype Learning for Imbalanced SSE
7.1. Introduction
7.2. Methodology
7.2.1. Overview
7.2.2. Frequency-Based Global-Scope Adaptive Mixer
7.2.3. Prototype Classifier
7.2.4. Training Procedure
7.3. Experiment
7.3.1. Datasets
7.3.2. Experiment Setting
7.3.3. Baseline Comparison on Public Datasets
7.3.4. Baseline Comparison on SSE Task
7.3.5. Sensitivity Analysis
7.3.6. Ablation Analysis
7.3.7. Real-Time Prediction
7.4. Conclusion
References
Part III. Cross-Vessel Transfer Learning for Ship Motion Data Based SSE
8. Semi-supervised Transfer Learning for Cross-Vessel SSE
8.1. Introduction
8.2. Methodology
8.2.1. Problem Description
8.2.2. Framework
8.2.3. Data Alignment
8.2.4. Attention-Enabled Encoder
8.2.5. Adversarial Discriminator
8.2.6. Training Process
8.3. Experiment
8.3.1. Dataset
8.3.2. Comparison with Simple Transfer Learning Methods
8.3.3. Comparison with SOTA Deep Transfer Learning Methods
8.3.4. Comparison with Different Encoders
8.3.5. Comparison with Different Attention Mechanisms
8.3.6. Sensitivity Analysis
8.3.7. Discussion
8.4. Conclusion
References
9. Few-Shot Siamese Transfer Learning for Cross-Vessel SSE
9.1. Introduction
9.2. Methodology
9.2.1. Problem Setup and Model Structure
9.2.2. Data Pairing
9.2.3. Siamese Convolutional Neural Network
9.2.4. Loss Function
9.3. Experiment
9.3.1. Experimental Settings
9.3.2. Comparison with Models Learning from Scratch
9.3.3. Comparison with Direct Transfer and Fine-tune
9.3.4. Comparison with SOTA Transfer Learning Methods
9.3.5. Comparison with SOTA Attention Modules
9.3.6. Ablation Analysis
9.4. Discussion
9.5. Conclusion
References
Part IV. Robust and Structure-Aware Learning for Ship Motion Data Based SSE
10. Adversarially Robust Deep Models for SSE
10.1. Introduction
10.2. Methodology
10.2.1. Overview
10.2.2. SSE Module
10.2.3. Perturbation Examples Training Module
10.3. Experiments
10.3.1. Experiment Setup
10.3.2. Performance Evaluation
10.3.3. Model Analysis
10.3.4. Discussion
10.4. Conclusion
References
11. Structure-Aware Dynamic Graph Neural Networks for SSE
11.1. Introduction
11.2. Methodology
11.2.1. Data Preprocessing
11.2.2. Overall Architecture
11.2.3. Initial Feature Extraction Module
11.2.4. Dynamic Graph Structure Learning Module
11.2.5. Algorithm Description
11.3. Experiment
11.3.1. Baseline Comparison on SSE Task
11.3.2. Ablation Study
11.3.3. Visualization Analysis
11.4. Conclusion
References
12. Conclusion