SAR Image Intelligence for Defensive Situational Awareness
An end-to-end data and machine-learning project for turning Synthetic Aperture Radar imagery into structured, analyst-ready information.
GeoSpace Defence AI explores how SAR imagery can move through a complete analytical workflow — from raw files to validated data, exploratory analysis, machine-learning experiments, evaluation, and eventually an analyst-facing interface.
The goal is not to build another notebook that trains a classifier. The project is being developed as a small, reproducible data system where every stage can be inspected, questioned, and improved.
| Data Foundation Inspect, validate, resize and encode SAR imagery. |
Analysis Understand classes, distributions, image statistics and outliers. |
Machine Learning Build baselines, evaluate mistakes and compare experiments. |
Analyst Layer Turn predictions and metrics into usable visual reporting. |
Raw SAR Data
↓
Inspection & Validation
↓
Preprocessing → 128 × 128
↓
Processed Dataset (.npz)
↓
EDA + Train / Validation / Test
↓
Baseline Model
↓
Evaluation & Error Analysis
↓
Structured Results
↓
Analyst Interface
The current build uses the MSTAR-10-Classes dataset package. It contains separate train/test directories and 11 supplied class folders. Images are standardized to 128 × 128, labels are encoded, and reusable arrays are exported to mstar_processed_128.npz.
The processed artifact keeps X_train, y_train, X_test, and y_test separate so later experiments do not need to repeat the raw preprocessing stage.
| Complete | In Progress | Next |
|---|---|---|
| Dataset acquisition | Validation split | Baseline classifier |
| Structure inspection | Exploratory analysis | Model evaluation |
| Image preprocessing | Dataset statistics | Error analysis |
| Label encoding | Experiment storage | |
| Processed dataset export | Analyst dashboard |
Radar imagery is useful in conditions where optical imagery can be limited. That makes SAR an interesting foundation for infrastructure monitoring, disaster-response analysis, logistics resilience, terrain/object analysis, and defensive situational awareness.
The project is moving toward reproducible experiment tracking, queryable metadata, class-level evaluation, confusion analysis, misclassification inspection, and a compact interface for comparing results.
Data → Analysis → Intelligence → Analyst Support
This is an educational and defensive analytics project. It is intended for analysis, resilience research, monitoring, and decision support — not weapon targeting, strike planning, or autonomous engagement.
Built incrementally. Documented as it evolves.