Inspiration
Environmental changes such as deforestation, pollution, and glacier retreat are happening faster than ever, yet many go unnoticed until the damage becomes irreversible. We were inspired by the need for a tool that could provide early visibility into these transformations. EcoVision was conceived as an AI-powered solution to monitor and detect subtle environmental changes over time using satellite and drone imagery. Our goal is to bridge the gap between data and action — enabling governments, researchers, and organizations to respond before small shifts become global problems. If humans can’t be everywhere, let vision AI watch over our planet.
What it does
EcoVision is a visual difference engine that analyzes satellite and drone images taken over time to detect and classify environmental changes. It automatically compares images from different periods to identify patterns such as deforestation, water shrinkage, urban expansion, and glacier reduction. The system highlights areas of significant visual change and generates a detailed report with change type, confidence score, and estimated area affected. Users can view results through an interactive dashboard that displays heatmaps, side-by-side image comparisons, and time-series change analytics. In short, EcoVision transforms raw image data into actionable environmental insights — enabling timely decisions for conservation, urban planning, and disaster management.
How we built it
We designed EcoVision as a modular AI-powered visual difference engine to detect environmental changes using satellite and drone images.
- Data Collection *Sourced multi-temporal images from open datasets such as NASA Earth Observatory and Sentinel Hub.
- Preprocessing *Used OpenCV for image alignment, resizing, denoising, and color normalization to handle variations in lighting, angle, and scale.
- Feature Extraction *Extracted high-level spatial features using a pretrained ResNet model, converting images into comparable embeddings.
- Change Detection *Measured visual change using the Structural Similarity Index (SSIM): $$ \text{ChangeScore} = 1 - \text{SSIM}(I_1, I_2) $$ where ( I_1 ) and ( I_2 ) are images from different time points.
- Classification and Visualization *Classified detected change regions with a fine-tuned CNN into categories such as vegetation loss, urban growth, and water reduction. *Displayed results in an interactive dashboard with heatmaps and time-lapse comparisons.
- Reporting and Insights *Generated quantitative summaries, e.g.: $$ \Delta A = A_{t_2} - A_{t_1} $$ representing the area of change over time. *Provided actionable insights for researchers and policymakers.
Challenges we ran into
While building EcoVision, we faced several technical and practical challenges: Data Variability: *Satellite and drone images often differed in lighting, angle, resolution, and seasonal conditions, making alignment and comparison difficult. Dataset Limitations: *Collecting balanced labeled data for multiple environmental scenarios was challenging, especially for rare events like glacier retreat or sudden water shrinkage. Change Detection Accuracy: *Simple pixel differencing produced many false positives. We had to refine feature-based methods and integrate deep learning for semantic understanding. Explainability: *Ensuring that detected changes were interpretable and trustworthy required additional visualization layers and confidence scoring. Computational Constraints: *Processing high-resolution multi-temporal images efficiently while maintaining accuracy was computationally demanding. *Despite these challenges, we were able to develop a robust system capable of detecting meaningful environmental changes and producing actionable insights.
Accomplishments That We’re Proud Of
- Effective Change Detection
- Developed a visual difference engine that can reliably detect environmental changes such as deforestation, urban expansion, and water body reduction over time.
- Interactive Visualization
- Built a dashboard that clearly displays detected changes through heatmaps, side-by-side comparisons, and time-series visualizations, making the results easy to interpret.
- Quantitative Insights
- Successfully quantified changes using metrics such as: $$ \Delta A = A_{t_2} - A_{t_1} $$ providing actionable data for researchers and policymakers.
- Robust AI Integration
- Combined classical image processing techniques with deep learning (CNNs, feature embeddings) for higher accuracy and reduced false positives.
- Scalable Design
- Designed the system to handle multi-temporal images efficiently, enabling potential real-time monitoring and future expansion to additional environmental domains.
- Practical Impact Potential
- Created a tool that can support conservation efforts, urban planning, disaster management, and sustainability initiatives globally.
What We Learned
- Handling Real-World Data
- Working with multi-temporal satellite and drone images taught us how to manage inconsistencies in lighting, angle, resolution, and seasonal variations.
- Importance of Feature-Based Methods
- Deep learning-based feature extraction outperformed simple pixel differencing, highlighting the importance of semantic understanding in visual change detection.
- Balancing Accuracy and Efficiency
- We learned to optimize computational resources while maintaining detection accuracy for high-resolution images.
- Explainability Matters
- Designing clear visualizations and confidence scores was critical to ensure that results could be trusted and actionable by non-technical users.
- Interdisciplinary Thinking
- Combining environmental knowledge with AI techniques helped us translate technical outputs into meaningful insights for real-world decision-making.
What’s Next for Visual Difference Engine
- Real-Time Monitoring
- Integrate live satellite and drone feeds to provide continuous, up-to-date environmental change detection.
- Weather and Seasonal Normalization
- Incorporate models to normalize images for lighting, cloud cover, and seasonal variations to improve detection accuracy.
- Expanded Environmental Domains
- Extend the system to monitor additional areas such as wetlands, coral reefs, and desertification.
- Automated Reporting and Alerts
- Develop automated dashboards and alert systems for policymakers, researchers, and conservation organizations.
- Scalability and Cloud Deployment
- Deploy the system on cloud platforms to handle larger datasets and enable global monitoring.
- Enhanced Explainability
- Add features that explain why specific changes were flagged, helping users make informed decisions quickly.
Goal: Transform EcoVision from a prototype into a global environmental monitoring tool that provides actionable insights and supports sustainability initiatives.
Log in or sign up for Devpost to join the conversation.