Deep learning based harmful algal blooms modeling in inland water using multimodal monitoring

목차 (Table of Contents)

• Chapter 1 Introduction
• 1.1 Backgrounds and Motivations
• 1.2 Research Objectives
• 1.3 In publications
• 1.4 Literature Reviews
• 1.4.1 Harmful Algal Bloom Monitoring in Inland Water
• 1.4.2 Limitations of Algal Bloom Models
• 1.4.3 Data-driven approaches and deep learning to estimate algal blooms
• 1.4.4 Multimodal Learning for water quality management
• 1.4.5 Explainable AI for interpretable HAB predictions
• Chapter 2 Deep Learning for Super-Resolution Monitoring of Inland Harmful Algal Blooms
• 2.1 Introduction
• 2.2 Materials and Methods
• 2.2.1 Study area
• 2.2.2 Research overview
• 2.2.3 Data acquisition
• 2.2.4 Airborne and satellite image preprocessing
• 2.2.5 Super-resolution of satellite imagery using deep learning models
• 2.2.5.1 Super-resolution CNN (SRCNN)
• 2.2.5.2 Super-resolution generative adversarial network (SRGAN)
• 2.2.6 Generative of super-resolution map of Chlorophyll-a concentration
• 2.2.7 Performance evaluation
• 2.3 Results and Discussion
• 2.3.1 Spatial variability of water reflectance in the processed satellite and airborne data
• 2.3.2 Super-resolution of deep learning models
• 2.3.2.1 Trading and validation of SRCNN
• 2.3.2.2 Training and validation of SRGAN
• 2.3.2.3 Performance comparison of SRCNN with SRGAN
• 2.3.3 Fine resolution map of Chlorophyll-a distribution from SRCNN and SRGAN
• 2.3.4 Super-resolution using deep learning for water remote sensing
• 2.4 Conclusion
• 2.5 Supplementary information for Chapter 2
• Chapter 3 Phytoplankton Abundance Estimation Using Probabilistic Machine Learning and Hyperspectral Imaging
• 3.1 Introduction
• 3.2 Materials and Methods
• 3.2.1 Study area description
• 3.2.2 Data acquisition
• 3.2.2.1 Airborne remote sensing and image processing
• 3.2.2.2 In-situ monitoring and phytoplankton density analysis
• 3.2.3 Deep learning approach
• 3.2.3.1 Bayesian neural network
• 3.2.3.2 Natural gradient boosting
• 3.2.3.3 Hyperparameter optimization
• 3.2.4 Model performance evaluation
• 3.3 Results and Discussion
• 3.3.1 Temporal variation of algal blooms
• 3.3.2 Simulation results for the deep learning approach
• 3.3.2.1 Optimization of probabilistic deep learning algorithms
• 3.3.2.2 Training and testing of the Bayesian neural network
• 3.3.2.3 Training and testing of the NGBOOST algorithm
• 3.3.3 Spatial distribution map of algal blooms
• 3.3.4 Implications of uncertainty estimation for algal blooms
• 3.4 Conclusion
• Chapter 4 Multi-modal learning for algae classification using image and particle modalities
• 4.1 Introduction
• 4.2 Materials and Methods
• 4.2.1 Research overview
• 4.2.2 Modalities description
• 4.2.2.1 Data acquisition
• 4.2.2.2 Image and particle properties data
• 4.2.2.3 Modality processing for feature fusion
• 4.2.3 Model development and evolution
• 4.2.3.1 Multi-modal algorithm for algae identification
• 4.2.3.2 Model evaluation
• 4.2.4 Interpretation of deep learning model
• 4.2.4.1 Shapley additive explanation
• 4.2.4.2 Gradient-weighted class activation map
• 4.3 Results and Discussion
• 4.3.1 Algae phylum morphology
• 4.3.2 Algae phylum identification
• 4.3.2.1 Algae identification performance from modalities
• 4.3.2.2 Performance analysis of algae identification: Multi-modal, image, and particle-based identification
• 4.3.3 Interpretation of multi-modal algorithm for algae phylum identification
• 4.3.3.1 Ablation study from the particle properties
• 4.3.3.2 Contribution of model input features
• 4.4 Conclusion 123
• 4.5 Supplementary information for Chapter 4
• Chapter 5 High-frequency Cyanobacteria Bloom Monitoring and Multimodal Deep Learning and Tower-based Hyperspectral System
• 5.1 Introduction
• 5.2 Materials and Methods
• 5.2.1 Study area description
• 5.2.2 Data acquisition
• 5.2.2.1 Tower-based hyperspectral monitoring
• 5.2.2.2 In-situ monitoring
• 5.2.3 Deep learning approach
• 5.2.3.1 Multi-modal algorithm
• 5.2.4 Post-hoc process
• 5.3 Results and Discussions
• 5.3.1 High-temporal variation of tower-based hyperspectral system
• 5.3.2 Simulation results for the deep learning approach
• 5.3.3 The implication of the multi-modal algorithm
• 5.3.3.1 Comparative analysis of data modalities
• 5.3.3.2 Contribution of the input features for multimodal algorithm
• 5.4 Conclusion
• 5.5 Supplementary information for Chapter 5
• Chapter 6 Conclusions
• 6.1 Conclusion
• 6.2 Future plan
• Reference