Academic Assignment Submission
Course: Foundations of Data Science Core (CS3005)
Student: Harsh Dayal ///////
Instructor: Dr. Prerna Jha
This project implements advanced machine learning and deep learning techniques for crop vs. non-crop land segmentation using satellite and UAV imagery. The work focuses on developing robust classification models that can distinguish between agricultural and non-agricultural land areas using multi-temporal satellite data and spectral analysis.
- Multi-temporal satellite image analysis for crop segmentation
- NDVI-based vegetation index classification
- Deep learning models for improved accuracy
- Comprehensive data preprocessing pipeline
- Performance evaluation with multiple metrics
This project employs a comprehensive approach to crop vs. non-crop land segmentation:
- Data Preprocessing: Multi-temporal satellite image preparation and normalization
- Feature Engineering: NDVI calculation and spectral band analysis
- Classical ML Approaches: Traditional machine learning methods for baseline comparison
- Deep Learning Models: Advanced neural networks for improved segmentation accuracy
- Evaluation: Comprehensive performance assessment using multiple metrics
The methodology is based on research by Rose M. Rustowicz and incorporates modern deep learning techniques for enhanced accuracy in agricultural land classification.
├── config/ # Configuration files
│ └── config.yaml # Main configuration parameters
├── Dataset/ # Raw dataset storage
│ └── download.txt # Dataset download instructions
├── DL model/ # Deep learning model implementations
│ └── Crop_classification_DL_model.ipynb
├── docs/ # Documentation and references
│ └── references.md # Academic references and citations
├── NDVI based/ # NDVI-based classification methods
│ └── NDVI_based.ipynb
├── notebooks/ # Jupyter notebooks for analysis
│ └── __init__.py
├── results/ # Output results and artifacts
│ ├── figures/ # Generated plots and visualizations
│ ├── metrics/ # Performance metrics and evaluations
│ └── models/ # Trained model artifacts
├── src/ # Source code modules
│ ├── __init__.py
│ ├── data_preprocessing.py # Data preprocessing utilities
│ ├── models.py # ML/DL model implementations
│ └── utils.py # Utility functions
├── data-preprocessing.ipynb # Main preprocessing notebook
├── cover1.png # Project overview image
├── test3.kml # Sample KML file for testing
└── README.md # This file
- Python 3.7+
- GDAL library for geospatial data processing
- CUDA-compatible GPU (recommended for deep learning models)
# Create conda environment
conda create --name crop_segmentation python=3.7
# Activate environment
conda activate crop_segmentation
# Install geospatial dependencies
conda install gdal rasterio
# Install core data science packages
conda install numpy pandas geopandas scikit-learn jupyterlab matplotlib seaborn
# Install additional packages
conda install xarray rasterstats tqdm pytest sqlalchemy scikit-image scipy
conda install pysal beautifulsoup4 boto3 cython statsmodels future graphviz
conda install pylint line_profiler nodejs sphinx
# For deep learning (optional - choose based on your system)
conda install pytorch torchvision torchaudio -c pytorch
# OR
pip install tensorflowThe project utilizes multi-temporal satellite imagery for crop segmentation:
- Satellite Data: 10 RapidEye satellite images from Planet.com
- Ground Truth: USDA Cropland Data Layer for pixel-level crop labels
- Format: Multi-spectral imagery with temporal sequences
- Coverage: Agricultural regions with diverse crop types
Dataset can be downloaded from the provided Google Drive link in Dataset/download.txt.
# Create and activate virtual environment
conda create --name crop_segmentation python=3.8
conda activate crop_segmentation
# Install requirements
pip install -r requirements.txt# Start Jupyter Lab
jupyter lab
# Execute notebooks in order:
# 1. data-preprocessing.ipynb
# 2. NDVI based/NDVI_based.ipynb
# 3. DL model/Crop_classification_DL_model.ipynb# Import preprocessing utilities
from src.data_preprocessing import preprocess_satellite_data
from src.models import CropSegmentationCNN, ClassicalMLModels, NDVIClassifier
from src.utils import load_config, calculate_comprehensive_metrics
# Load configuration
config = load_config('config/config.yaml')
# Process data and train models
X_processed, y_processed = preprocess_satellite_data('Dataset/raw/', config)📖 For comprehensive setup, troubleshooting, and execution instructions, see: Model Execution Guide
This detailed guide covers:
- Complete environment setup and installation
- Step-by-step model execution workflow
- Performance optimization techniques
- Troubleshooting common issues
- Expected results and benchmarks
- Multi-temporal image alignment and normalization
- Spectral band extraction and processing
- NDVI calculation for vegetation analysis
- Data augmentation for improved model robustness
- Classical ML: Random Forest, SVM, and ensemble methods
- Deep Learning: Convolutional Neural Networks (CNN) for image segmentation
- Feature Engineering: Spectral indices and temporal features
- Evaluation Metrics: Accuracy, Precision, Recall, F1-Score, IoU
- NDVI-Based Classification: Vegetation index thresholding
- Multi-Temporal Analysis: Time-series pattern recognition
- Deep CNN: Semantic segmentation for pixel-level classification
- Ensemble Methods: Combining multiple models for improved accuracy
The project achieves significant improvements over traditional mono-temporal approaches:
- Baseline Accuracy: ~75% (traditional spectral methods)
- NDVI-Enhanced: ~82% (with vegetation indices)
- Deep Learning: ~89% (CNN-based segmentation)
- Multi-Temporal: ~92% (temporal pattern analysis)
Detailed results and visualizations are available in the results/ directory.
This project demonstrates:
- Methodological Innovation: Integration of multi-temporal analysis with deep learning
- Technical Implementation: Robust preprocessing pipeline for satellite imagery
- Performance Analysis: Comprehensive evaluation of different approaches
- Practical Application: Real-world agricultural monitoring capabilities
- Model Execution Guide - Comprehensive setup and execution instructions
- Academic References - Complete academic references and related work
- requirements.txt - Complete list of Python dependencies
- Configuration Guide - Model and data processing parameters
- Integration of additional satellite data sources (Sentinel-2, Landsat)
- Real-time processing capabilities for operational deployment
- Extension to multi-class crop type classification
- Integration with IoT sensors for ground-truth validation
This project is submitted as part of academic coursework for CS30305 - Foundations of Data Science Core.
Note: This implementation focuses on demonstrating advanced data science techniques for agricultural applications, combining traditional machine learning with modern deep learning approaches for improved crop vs. non-crop land segmentation accuracy.