Drone Detection & Distance Estimation (YOLOv8)

Multi-class drone detection model trained on quadcopter and military fixed-wing drones.

Model Details

Architecture: YOLOv8n (nano)
Classes: 2
- quadcopter (class 0) — Multirotor drones (DJI Mavic, Mini, etc.)
- fixed-wing (class 1) — Military fixed-wing drones (Shahed-131/136, Lancet, Orlan-10)
Input Size: 640x640
Training: 30 epochs on RTX 3090
Framework: Ultralytics

Performance

Metric	Value
mAP50	91.3%
mAP50-95	62.1%
Precision	94.6%
Recall	85.9%

Training Data

Quadcopters: 9,000 images from Seraphim dataset
Fixed-wing: 9,000 images from Kaggle military drone dataset
Split: 80% train, 15% val, 5% test

Distance Estimation

Model integrates with monocular distance estimation using known object sizes:

Quadcopter: 0.35m wingspan → accurate distance for small drones
Fixed-wing: 2.5m wingspan → accurate distance for Shahed-type drones

Formula: distance = (real_wingspan * focal_length) / bbox_width_px

Usage

from ultralytics import YOLO

# Load model
model = YOLO('TomSmail/drone-yolo-v1')

# Run inference
results = model.predict('image.jpg', conf=0.3)

# Get detections
for result in results:
    for box in result.boxes:
        class_name = result.names[int(box.cls[0])]
        confidence = float(box.conf[0])
        x1, y1, x2, y2 = box.xyxy[0].tolist()
        print(f"{class_name}: {confidence:.2%}")

Integration with Distance Estimation

from drone_cv import Detector

detector = Detector(config_path="configs/")
result = detector.predict(frame)

for detection in result.detections:
    print(f"{detection.class_name}: {detection.distance_m:.1f}m")

Limitations

Fixed-wing class does not distinguish between specific military drone types
Distance estimation assumes perpendicular viewing angle
Accuracy degrades with occlusion, extreme angles, or poor lighting
Trained on synthetic + real data mix; performance on real-world military drones not validated

Citation

@misc{drone-yolo-v1,
  author = {FRAID Labs},
  title = {Drone Detection & Distance Estimation Model},
  year = {2026},
  publisher = {HuggingFace},
  url = {https://huggingface.co/TomSmail/drone-yolo-v1}
}

License

CC-BY-4.0

Training datasets used:

Seraphim: CC-BY-4.0
Kaggle military drone: CC0-1.0

Downloads last month: 101

TomSmail
/

drone-yolo-v1