Aircraft Detector using YOLOv8

Overview

This project implements an Aircraft Detector using YOLOv8 to detect aircraft in images. The dataset is automatically split into train, validation, and test sets, and a YOLOv8 Nano model is trained on this data.

Project Structure

├── dataset.yaml  # YOLO dataset config
├── weights/
│   ├── best.pt  # Best trained model
│   ├── last.pt  # Last trained model
├── train_model_v2.py  # Training script
├── test_v2.py  # Inference script
├── README.md  # Project documentation

Installation

Ensure you have Python 3.8+ installed. Then, install dependencies:

pip install ultralytics numpy matplotlib opencv-python 

Dataset Preparation

The dataset is automatically split into train (80%), validation (10%), and test (10%) during execution. Dataset: https://www.kaggle.com/datasets/khlaifiabilel/military-aircraft-recognition-dataset/data

Training the Model

To train the YOLOv8 model, run:

python train_model_v2.py

This will:

• Load the YOLOv8-Nano model (yolov8n.pt)
• Train the model for 15 epochs (increase to 50 if your machine can handle)
• Save trained weights in the weights/ directory

Testing the Model

Run inference on test images:

python test_v2.py

This will:

• Detect objects in all test images
• Draw bounding boxes and count objects
• Display results using Matplotlib

Evaluation Metrics

The trained model is evaluated using:

• Precision
• Recall
• F1 Score

Results : F1 Curve : PRECISION Curve : PRECISION RECALL Curve : RECALL Curve :

Model Weights

The trained YOLOv8 model weights are saved in the weights/ directory:

• best.pt – Best performing model checkpoint
• last.pt – Model from the last training epoch

Author

Agneya Pathare – Robotics Engineer | AI & Computer Vision Developer