🌲 VanRakshak YOLOv8n: Edge Aerial & Thermal Forest Drone Detector

VanRakshak YOLOv8n is a real-time object detection model fine-tuned for autonomous drone surveillance, anti-poaching patrol, wildfire detection, and wildlife conservation.

Part of the open-source VanRakshak Autonomous Drone Platform.

📊 Measured Benchmark Scores (Tesla T4 GPU)

Evaluated on 1,786 held-out validation aerial and thermal drone frames:

Metric	Score	Hardware / Note
mAP@50	80.76%	High-altitude bounding box detection quality
Precision (Box P)	83.76%	Ultra-low false alarm rate for autonomous ranger alerting
Recall (Box R)	72.45%	Robust target recall across canopy occlusion & night thermal
mAP@50-95	42.71%	Strict multi-IoU geometric precision
Inference Latency	2.7 ms	~190+ FPS throughput on GPU (5.2 ms end-to-end frame turnaround)
Model Size	3.0M params	6.0 MB PyTorch (`best.pt`) / 11.7 MB ONNX (`best.onnx`)

🏷️ Class List (6 Classes)

0: person        # Intruders, poachers, and hikers in RGB aerial & night thermal
1: vehicle       # Civilian cars, vans, and motorcycles on forest tracks
2: timber_truck  # Heavy timber transport and illegal logging flatbeds
3: fire          # Active wildfire flame fronts and spot fires
4: smoke         # Early wildfire smoke plumes and combustion clouds
5: elephant      # Wildlife (elephants and megafauna) for corridor monitoring

🚀 Quickstart & Inference

Python (Ultralytics)

from ultralytics import YOLO

# Load model directly
model = YOLO("best.pt")

# Run inference on drone aerial or thermal frame
results = model("drone_thermal_frame.jpg", conf=0.50)

# Visualize predictions
for r in results:
    r.show()

Edge ONNX Inference (Drone Companion Computer)

import onnxruntime as ort

session = ort.InferenceSession("best.onnx", providers=["CPUExecutionProvider"])
# Optimized for Raspberry Pi 5 and NVIDIA Jetson Orin Nano

🔗 Links & Resources

💻 GitHub Codebase: https://github.com/sanjeevafk/vanrakshak
📦 Training Dataset: sanjeevafk/vanrakshak-forest-aerial-thermal
📓 Colab Training Pipeline: vanrakshak_yolov8_training.ipynb

⚖️ License

Released under the Apache 2.0 License.

Downloads last month: 34