Drone, Helicopter, and Aircraft Detection using YOLOv11n 🚁

This project demonstrates an end-to-end pipeline for detecting drones, helicopters, and airplanes in both images and videos using the YOLOv11n object detection model. It features a clean web interface built with FastAPI and allows users to upload multiple images or a video file, view previews, and see real-time predictions.

---

🌐 Project Overview

This repository provides a complete solution to:

• Detect drones, helicopters, and aircraft in both images and videos
• Use a custom-trained YOLOv11n object detection model
• Preview uploaded media and visualize detections on the web
• Run a FastAPI backend with a modern HTML/CSS/JS frontend
• Upload multiple images (batch detection) or a single video for processing
• Organize and serve predictions without saving unnecessary files permanently

---

📁 Project Structure

drone-detection-app/
├── app/
│   ├── main.py              # FastAPI app with endpoints
│   ├── model.py             # YOLOv11n model loading and prediction logic
│   ├── utils.py             # Utility functions (e.g., video processing)
│   ├── uploads/             # Temporary storage for uploaded files
│   └── output/              # Stores processed video results
│
├── static/
│   ├── web.css              # CSS styling for the web interface
│   └── web.js               # JS logic for uploading, previews, detection
│
├── templates/
│   └── web.html             # Jinja2 HTML template for the main UI
│
├── data/                    # YOLO formatted dataset
│   ├── images/
│   │   ├── train/ test/ val/
│   ├── labels/
│   │   ├── train/ test/ val/
│   └── dataset.yaml
│
├── csv_to_yolo.py           # Converts annotation CSV to YOLO format
├── drone_detection.ipynb    # Model training notebook
├── test.jpg                 # Example image
├── yolov11n.pt              # Trained YOLOv11n model weights
└── runs_n/, runs/           # YOLO training outputs (optional)

---

⚙️ Installation

1. Clone the Repository

git clone https://github.com/AksharKher-30/AeroDetect.git
cd drone-detection-app

2. Install Dependencies

pip install -r requirements.txt

Make sure you have Python 3.8+ and pip installed on your system.

3. Start the Web Server

uvicorn app.main:app --reload

Visit: http://127.0.0.1:8000

---

📁 Data Preparation

• The dataset contains annotations in Pascal VOC (XML) format for drones, helicopters, and airplanes.
• To convert them into YOLO format, use the provided script:

  python csv_to_yolo.py

This script will:

• Parse all XML annotation files from the dataset
• Convert bounding box annotations to YOLO-compatible .txt format
• Automatically place the converted .txt files into the corresponding labels/ directoies.

Ensure your dataset is structured as follows before training:

data/
├── images/
│   ├── train/
│   ├── val/
│   └── test/
└── labels/
    ├── train/
    ├── val/
    └── test/

⚠️ Make sure the image and label filenames match exactly for YOLO training or you can change the path in the code if different names of files/folder.

The trained weights will be saved under:

Aero Detect/runs_n/drone_train_n/drone_detector_n/weights/best.pt
├── best.pt
└── last.pt

best.pt is used for inference.

🧪 How to Use

After starting the webpage:

🌄 Image Prediction

• Upload up to 8 images
• Preview appears instantly
• Click Detect to see prediction overlays

🎞️ Video Prediction

• Upload one video (MP4 format recommended)
• Preview shows the original video
• Click Detect to view the result in-place

📊 Performance

Validation results on the test set using YOLOv11n
Image size: 512, Confidence threshold: 0.35, IoU threshold: 0.5

Class	Images	Instances	Precision	Recall	mAP@0.5	mAP@0.5:0.95
All	603	497	0.920	0.928	0.936	0.592
Drone	220	224	0.887	0.879	0.874	0.447
Helicopter	140	140	0.985	0.964	0.981	0.676
Airplane	133	133	0.887	0.940	0.952	0.655

---

🚀 Working Demo

demo.mp4