Hyperspectral-Deep-Learning

🔍 Hyperspectral Image Classification and Segmentation

A comprehensive implementation of deep learning models for hyperspectral image classification and segmentation using PyTorch. The project implements various architectures and techniques in a single streamlined pipeline.

📚 Table of Contents

1. 🔄 Data Acquisition and Preprocessing

   • 📥 Downloading Datasets
   • 🔃 Data Format Conversion and Organization
   • ✂️ Dataset Splitting (80/10/10)
   • 📊 Band Quality Assessment
   • 🧹 Noise Removal

2. 📈 Data Exploration and Visualization

   • 📉 Spectral Signatures Visualization
   • 🔗 Band Correlation Analysis
   • 📊 Class Distribution Analysis
   • 🎲 3D Datacube Visualizations

3. 🎯 Dimensionality Reduction and Feature Selection

   • 🧮 PCA, ICA, LDA Analysis
   • 🎯 Band Selection Methods
   • ✨ Reduced Representations Evaluation

4. 🔄 Data Augmentation Strategies

   • 🌊 Spectral Augmentation
   • 🌍 Spatial Augmentation
   • 🔄 Combined Spectral-Spatial Augmentation

5. 🏗️ Model Architecture Design

   • 📋 Classification Models
     • 🔥 ResNet18 with Spectral Attention
     • 🤖 Vision Transformer (ViT)
     • 🧊 3D CNN
     • 🔄 Hybrid CNN-Transformer
   • 🎨 Segmentation Models
     • 🌈 U-Net with Spectral Attention
     • 🎲 3D U-Net
     • 🎯 FCN with Spectral Attention

6. ⚙️ Training Pipeline Implementation

   • 📉 Loss Functions
     • 💫 Cross-entropy with Class Weights
     • 🎯 Dice Loss for Segmentation
   • 🔧 Optimization
     • ⚡ Adam Optimizer
     • 📈 Learning Rate Scheduling
   • 🔒 Regularization
     • 🎭 Spectral Dropout
     • 🏋️ L1/L2 Regularization

7. 📊 Model Training and Monitoring

   • 🎯 Model Selection
   • 🏃‍♂️ Training and Validation
   • 💾 Checkpoint Management
   • 🛡️ Overfitting Handling

8. 📊 Results and Visualization

   • 📈 Performance Charts
   • 🎨 Color-graded and Normalized Visualizations

✨ Features

📚 Supported Datasets

• 🌾 Indian Pines
• 🏛️ Pavia University
• 🌱 Salinas Scene

🛠️ Core Functionality

• 🔄 Automatic dataset downloading and processing
• 📊 Band quality assessment and noise removal
• 🔄 Spectral and spatial data augmentation
• 📉 Dimensionality reduction techniques
• ⚡ Advanced training features with regularization
• 📊 Comprehensive visualization and evaluation tools

📋 Requirements

torch>=1.8.0
numpy>=1.19.2
pandas>=1.2.0
scipy>=1.6.0
scikit-learn>=0.24.0
matplotlib>=3.3.4
seaborn>=0.11.1
tqdm>=4.59.0
plotly>=4.14.0

🚀 Usage

1. Install the required dependencies:

pip install torch numpy pandas scipy scikit-learn matplotlib seaborn tqdm plotly

2. Run the main training pipeline:

# Select dataset and models
SELECTED_DATASET = ['indian_pines']  # Options: 'indian_pines', 'pavia_university', 'salinas'
models = ['fcn']  # Options: 'resnet', 'vit', '3dcnn', 'hybrid', 'unet', 'fcn'
# Training will automatically:
# - Download and process the dataset
# - Train the selected models
# - Generate visualizations and metrics

🔑 Key Components

📊 Data Processing

# Load and preprocess dataset
loader = HyperspectralDataLoader(dataset_name)
data, ground_truth = loader.load_dataset()
# Analyze band quality
analyzer = BandQualityAnalyzer(data, dataset_name)
noisy_bands = analyzer.identify_noisy_bands()

🏃‍♂️ Model Training

# Create model
model = create_model(dataset_name)  # For classification
# or
model = create_fcn_model(dataset_name)  # For segmentation
# Train
trainer = Trainer(config, task_config)
trainer.train()

📈 Visualization

# Plot results
plot_confusion_matrix(dataset_name, classification_models, device)
plot_segmentation_maps(dataset_name, segmentation_models, device)
plot_training_curves(metrics, dataset_name, model_type)

📊 Results

The code includes comprehensive evaluation tools that generate:

• 📈 Classification accuracy metrics
• 🎯 Segmentation IoU/Dice scores
• 📊 Confusion matrices
• 📉 ROC curves
• 🔍 Error analysis
• 📊 Band quality visualization
• 📈 Training progress curves