Infrastructure Defect Detection using YOLOv8

This project implements an end-to-end Machine Learning Computer Vision pipeline to detect and classify structural defects in infrastructure (such as concrete cracks, spalling, and delamination) using deep learning.

📌 Project Overview

Monitoring infrastructure health is a highly manual and time-consuming process. This project leverages the YOLOv8 object detection model to automate the identification of structural anomalies. The model was trained on an academic computing cluster to rapidly process high-resolution imagery.

Key Features:

• Automated Data Pipeline: Securely downloads and extracts dataset from Roboflow via API.
• Programmatic Label Correction: Automatically maps and fixes corrupted/default YOLO class labels.
• High-Speed Inference: Achieves ~80ms inference times per image on standard CPU hardware (12+ FPS).

🛠️ Installation & Setup

1. Clone the repository:

   git clone [https://github.com/yourusername/infrastructure-defect-detection-yolo.git](https://github.com/yourusername/infrastructure-defect-detection-yolo.git)
   cd infrastructure-defect-detection-yolo
   
   

2. Set up the virtual environment:

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate
pip install -r requirements.txt

3. Download the Data: Create a .env file in the root directory and add your Roboflow API key: ROBOFLOW_API_KEY=your_key_here

Then run the download script:

python download_data.py

📊 Dataset & Discoveries

The model was trained on the "Infrastructure-Defects-Detection-4" dataset. During exploratory data analysis and validation, a severe class imbalance was identified.

The validation set was heavily skewed, containing exclusively spalling defects. Consequently, the model exhibits a bias toward predicting spalling in uncertain inference scenarios. This highlights the critical real-world MLOps necessity for strict data curation and balanced validation splits prior to production deployment.

🚀 Results

The trained custom weights (yolov8_infrastructure_defects.pt) are available in the assets/ folder.

Normalized Confusion Matrix:

Training Loss Results:

💻 Running Inference

You can test the trained model on new images using the provided weights:

from ultralytics import YOLO

# Load the custom trained model
model = YOLO('assets/yolov8_infrastructure_defects.pt')

# Run inference
results = model('path/to/test/image.jpg')
results[0].show()