AEGIS-SEAS

SOVEREIGN C++20 INDIAN EEZ

Autonomous Satellite Radar Oil Slick Intelligence & Sovereign Polluter Attribution System

OPERATIONAL THEATER:

RADAR PLATFORM Sentinel-1 (C-Band, 10m)

AI ENGINE Real S1 Trained (5,969 Params)

PROCESSING LATENCY 1.10 s

THREAT STATUS ACTIVE SPILL CONFIRMED

🌐 TACTICAL RADAR SCENE & SITUATIONAL LAYERS Mumbai High Basin (71.35°E, 19.35°N)

LIVE STREAM Initializing AEGIS-SEAS Indian Ocean Maritime Surveillance Engine...

⚠️ DETECTED OIL SLICKS (REAL U-NET)

🚢 SOVEREIGN POLLUTER ATTRIBUTION

0 Suspects

🌊 REAL-TIME HYDRODYNAMIC DRIFT SIMULATOR

Wind Velocity (Fay 3% Windage): 6.8 m/s

Wind Direction: 240° (SW Monsoon)

Surface Current (Arabian Sea): 0.24 m/s (Konkan Drift)

Marangoni Capillary Damping: +8.9 dB to +9.6 dB

2D CA-CFAR Contact Sensitivity: P_fa = 10^-5 (Adaptive CUT)

Monte Carlo Dispersion Particles: 1,000 Particles / -4.0h Hindcast

TRAINING DATASET

Copernicus Sentinel-1 C-Band SAR

Authentic dual-pol (VV+VH) oil spill segmentation tiles (open CC BY 4.0 benchmark). 10m spatial resolution.

EMBEDDED ARCHITECTURE

Pure C++20 2D U-Net

5,969 parameters, 6 convolutional layers, bilinear upsampling, skip connections, SIMD OpenMP parallelized.

VALIDATION METRICS

Dice Score: 41.2% | mIoU: 29.4%

Trained with combined Focal/BCE + Dice loss with class-weight compensation for severe marine surface imbalance.

DEPLOYED WEIGHTS

data/unet_weights.bin

Verified binary weights loaded into C++ memory layout. Latency: ~1.1 seconds for 512x512 scene.

Epoch	Train Loss	Val Loss	Val Dice	Val mIoU
01	0.9477	0.8940	7.1%	3.7%
04	0.8957	0.8043	28.8%	18.3%
08	0.8229	0.7486	45.8%	33.1%
12	0.8089	0.7498	41.2%	29.4%