🛰️ Sat-Scan Terminal v3.0

Geospatial Intelligence Platform | Temporal Earth Observation & Change Analysis

---

---

🚀 Live Demo

https://satilite-image-detection.streamlit.app/

---

📌 Overview

Sat-Scan Terminal is a geospatial intelligence platform designed to detect, analyze and classify temporal surface changes using Sentinel-2 imagery.

Originally developed for Smart India Hackathon (SIH), Sat-Scan evolved into a complete engineering rebuild emphasizing:

• Temporal Synchronization
• Robust Feature Comparison
• Noise Resistant Classification
• Interpretable Spatial Analysis
• Large Scale Geospatial Processing

---

Status

Metric	Value
Current State	Functional MVP
Deployment	Streamlit Application Released
Validation	Real-world Satellite Evaluation
Version	v3.0
Status	Active

---

🌟 Core Innovation

Sat-Scan derives understanding from four complementary layers.

Layer	Objective
Temporal Alignment	Synchronize T1 and T2 observations
Feature Similarity	Siamese embeddings instead of pixel differencing
Probabilistic Interpretation	Dynamic World priors suppress environmental noise
Spatial Reasoning	Contour extraction enables interpretable outputs

Together these components create a robust pipeline capable of operating under real-world geospatial conditions.

---

📜 Engineering Journey

Initial prototypes suffered from:

• Coordinate desynchronization
• Seasonal false positives
• Region instability
• Poor spatial interpretation

Rather than applying incremental fixes, the architecture was redesigned from first principles.

Sat-Scan v3.0 represents:

• Root-cause debugging
• Architectural redesign
• Robustness over convenience
• Interpretability over complexity

---

🛰️ Sat-Scan Pipeline

Stage	Purpose
Sentinel-2 Acquisition	Retrieve temporal imagery
AOI Selection	Define analysis region
Temporal Alignment	Synchronize observations
Feature Extraction	Siamese ResNet embeddings
Distance Analysis	Compute change similarity
Clustering	Identify candidate regions
Contour Extraction	Generate interpretable boundaries
Dynamic World Classification	Region labeling
Probabilistic Filtering	Noise suppression
Visualization	Interactive DualMap rendering
Reporting	Change statistics

---

🛠️ Technology Stack

Domain	Stack
Deep Learning	PyTorch • Siamese Networks • ResNet-18
Remote Sensing	Sentinel-2 • Google Earth Engine
Image Processing	Scikit-Image
Geospatial Processing	Rasterio • Affine Transform
Visualization	Streamlit • Folium • DualMap
Classification	Dynamic World Probability Bands
Spatial Analysis	Thresholding • Contours • Clustering

---

📡 Classification Logic

IF Cluster_Area > Threshold:

    IF DynamicWorld_Mode == Built:
        LABEL = "HUMAN-MADE"

    ELSE:
        LABEL = "NATURAL"

---

📊 System Outputs

✔ Contour-based Region Detection

✔ Reduced Seasonal False Positives

✔ Human-interpretable Classification

✔ Temporal Synchronization

✔ Surface Change Statistics

✔ Interactive Visualization

---

📈 Improvements Over Initial Version

Component	Initial Prototype	Sat-Scan v3.0
Alignment	Drift Issues	Fully Synchronized
Noise Handling	High False Positives	Majority Rule Filtering
Detection	Bounding Boxes	Contours
Stability	Experimental	Consistent
Visualization	Static	Interactive
Classification	Pixel-based	Probabilistic

---

🚀 Local Setup

git clone https://github.com/Ganateju/Satilite-Image-Detection.git

cd Satilite-Image-Detection

pip install -r requirements.txt

Create:

.env

GEE_PROJECT_ID=your-google-project-id

Run:

streamlit run frontend/app.py

---

Engineering Domains

• Geospatial Intelligence
• Remote Sensing
• Computer Vision
• Deep Learning
• Spatial Analytics
• Earth Observation
• Data Visualization
• Systems Engineering

---

🧩 Developer Note

Failure is acceptable.

Not understanding failure is not.

Sat-Scan was rebuilt through architectural analysis and systematic debugging rather than incremental patching.

---

Built by Ganateju

"Perception under uncertainty."