YOLO

YOLO (You Only Look Once) is a family of real-time object detection models that can identify and locate multiple objects in images and video at high speed. The current versions — YOLOv8, YOLOv9, and YOLOv11 — are developed and maintained by Ultralytics, a company that has become the de facto steward of the YOLO ecosystem. The models are free and open source under the AGPL-3.0 license (free for academic and personal use; commercial deployment requires a license).

What makes YOLO unique in aerospace is speed. Traditional object detection models process images in hundreds of milliseconds. YOLO processes a full frame in as little as 1–5 milliseconds on GPU hardware — fast enough for real-time video from a drone camera at 30+ frames per second. This speed-accuracy tradeoff is why YOLO dominates edge deployment: drones, inspection robots, airport surveillance systems, and satellite image processing pipelines.

Students can install YOLO with a single command (pip install ultralytics), train custom models on their own datasets in hours using free Google Colab GPUs, and deploy trained models to NVIDIA Jetson, Raspberry Pi, or any device that runs Python. The barrier to entry is the lowest of any production-grade computer vision system.

High School

Start with the Ultralytics quickstart: install the library, run a pre-trained model on images and video, and see bounding boxes appear around detected objects. No training required — just inference with existing models. Use Google Colab (free GPU) if your computer lacks a dedicated GPU.

• Install: pip install ultralytics
• Run pre-trained YOLOv8 on your webcam or sample images
• Follow the Ultralytics documentation tutorials — they are exceptionally well written
• Try detecting aircraft in Google Earth screenshots or airport webcam feeds

Undergraduate

Train custom YOLO models on aerospace-specific datasets. This is where real skill develops.

• Collect or download a labeled dataset (aircraft types, runway FOD, drone imagery)
• Use Roboflow (free tier) or CVAT (open source) to annotate your own images
• Train a YOLOv8 model on Google Colab or your university's GPU cluster
• Evaluate performance with mAP (mean Average Precision) metrics
• Deploy to an NVIDIA Jetson Nano or Raspberry Pi for edge inference
• Integrate with a drone (PX4 or ArduPilot) for real-time airborne detection

Advanced / Graduate

Push into research-grade applications: modify the YOLO architecture for specific aerospace tasks, implement tracking (not just detection), work with satellite imagery at scale, or combine YOLO with other models for multi-task systems.

• Fine-tune on overhead/satellite imagery datasets (DOTA, xView, FAIR1M)
• Implement multi-object tracking with ByteTrack or BoT-SORT
• Optimize models with TensorRT for maximum throughput on edge hardware
• Publish results — aerospace computer vision is an active research area at CVPR and AIAA SciTech