Hyperspectral Change Detection and Change Captioning
About This Project
This research project focuses on developing advanced deep learning frameworks for **Hyperspectral Change Detection ** and Change Captioning using remote sensing imagery. The work aims to automatically identify significant temporal changes in hyperspectral satellite images and generate semantic textual descriptions of detected changes.
The project explores robust AI-driven techniques combining:
- Deep Learning
- Vision Transformers
- Mamba / State Space Models (SSM)
- Attention Mechanisms
- Multimodal Learning
- Remote Sensing Image Analysis
The primary objective is to improve accurate land-cover change analysis, disaster monitoring, environmental assessment, and intelligent geospatial understanding from hyperspectral imagery.
Research Focus
The project investigates:
- Pixel-level and semantic change detection
- Temporal feature extraction from hyperspectral data
- Lightweight and efficient Transformer architectures
- Mamba-based sequence modeling for long-range dependencies
- Vision-language models for automated change caption generation
- Robustness against illumination and seasonal variations
Methodology
The proposed pipeline includes:
- Hyperspectral image preprocessing and normalization
- Feature extraction using Transformer and Mamba architectures
- Spatial-spectral attention modeling
- Change map generation
- Semantic caption generation using multimodal learning frameworks
Applications
Potential applications include:
- Environmental monitoring
- Forest and vegetation analysis
- Urban expansion detection
- Agricultural monitoring
- Disaster assessment
- Defense and surveillance systems
Results
The project aims to achieve:
- Improved change detection accuracy
- Better semantic understanding of scene changes
- Lightweight and scalable architectures
- Enhanced interpretability in remote sensing AI systems
Future Scope
Future directions include:
- Foundation models for remote sensing
- Large-scale multimodal geospatial learning
- Real-time change captioning systems
- Edge-AI deployment for satellite analytics
- Self-supervised hyperspectral representation learning
