MilitaryConvoy-YOLO11L — Military Vehicle & Convoy Detection

Part of OmniSense / MediaSense / DeSense — Connectivia Labs' defense AI platform.

Performance

Metric	Value
mAP@0.5	50.0%
mAP@0.5:0.95	30.4%
Precision	57.0%
Recall	47.7%
Architecture	YOLO11L
Input size	640×640
Training date	2026-04-05

Classes (15 total)

Aerial/civilian (VisDrone): pedestrian · person · bicycle · car · van · truck · tricycle · awning-tricycle · bus · motor

Military (DeSense): military-vehicle · tank · apc · afv · artillery

Training

• Base: YOLO11L (COCO pretrained)
• Dataset: VisDrone2019-DET + MilitaryVehicleRecognition v7 (~9,300 images merged)
• Epochs: 75 | Batch: 32 | Imgsz: 640 | Optimizer: AdamW + cosine LR
• Augmentation: mosaic, mixup, copy-paste, flips, scale

Usage

from ultralytics import YOLO
model = YOLO('MuayThaiLegz/MilitaryConvoy-YOLO11L')
results = model('aerial_frame.jpg')

# Filter military classes only (indices 10-14)
mil = [b for b in results[0].boxes if int(b.cls) >= 10]
print(f'Military vehicles: {len(mil)}')

DeSense Pipeline

Camera/Satellite → MilitaryConvoy-YOLO11L → ByteTrack
→ DBSCAN convoy clustering → Stone Soup fusion → Geofence alert

OmniSense · Pillars: MediaSense · DeSense · Connectivia Labs 🔒