Spectrogram Dataset 🔉

Images of the feature extracted samples of the Custom UAV dataset can be found at UAV Classification Dataset

Papers 📜

1️⃣ 4,500 Seconds [Accepted, Preprint]: Arxiv 2505.23782
    ↪️4,500 Seconds Oral Presenation: YouTube Link
2️⃣ 15,500 Seconds [Under Review, Preprint]: Arxiv 2506.11049
3️⃣ The Unbearable Weight: TBD

Training Logs 🪵

Weights & Biases Training Logs

UAV Classification 🛩️

Code repository for UAV (Unmanned Aerial Vehicle) classification using deep learning techniques. The project is containerized using Docker and supports experiment tracking with Weights & Biases.

NOTE 📎

The Datasets used in this project are not included in the repository due to their visibility -> We have decided not to open-source the datasets.

If you would like to use the codebase, please use this example directory to store your datasets. and update the config.yaml file to point to your datasets.

Prerequisites 🔮

• Docker
• Python 3.8+
• CUDA-compatible GPU (recommended)

Setup 🏗️

1. Clone the repository:

git clone https://github.com/yourusername/UAV_Classification_repo.git
cd UAV_Classification_repo

2. (Optional) Copy the example environment file and configure your variables:

cp .env.example .env

3. Build and using Docker:

docker compose build app

Environment Variables 📨

Create a .env file in the root directory with the following variables (see .env.example):

Note: see setup section for more details

• WANDB_API_KEY: Your Weights & Biases API key
• BOT_TOKEN: Telegram bot token for notifications (optional)
• CHAT_ID: Telegram chat ID for notifications (optional)

Usage 🐳

1. Configure your experiment in src/config.yaml & orchestrate.yaml

2. Run training:

docker compose run app

License ⚖️

This project is licensed under the MIT License - see the LICENSE file for details.