AI Capstone Project: Concrete Crack Detection Using Deep Learning

This project is part of the AI Capstone Project with Deep Learning, which serves as the final project of the IBM AI Engineering Professional Certification through Coursera. In this project, I applied my knowledge in deep learning to solve a real-world challenge: detecting faults, cracks, or ruptures in concrete structures using image classification techniques.

Project Overview

The goal of this project is to build a deep learning model capable of detecting cracks in concrete structures from images. I utilized a ResNet18 architecture, modifying the final layer to classify between two categories: Positive for cracked and Negative for not cracked.

Key elements of the project include:

• Data Preprocessing: Images were preprocessed and normalized for input into the neural network.
• Model Architecture: A pre-trained ResNet18 model was used, with the final layer adapted to a (512, 2) output for binary classification.
• Training: The model was trained on labeled images of concrete structures, with the goal of maximizing accuracy and minimizing loss.
• Evaluation: The model's performance was evaluated using standard metrics such as accuracy and loss evolution, as well as misclassified samples.

Project Highlights

1. Model Architecture

The ResNet18 model was fine-tuned by replacing the last fully connected layer with a new layer (512, 2), allowing for binary classification between cracked and intact concrete structures.

Example output of the modified model:

2. Accuracy and Loss

The model achieved 97.03% accuracy during validation. Below are some visualizations of the training process:

• Accuracy and Loss Plot:

3. Misclassified Samples

Here are a few examples of misclassified samples, which provide insights into the challenges faced by the model when distinguishing between cracked and non-cracked images:

What I Learned

Through this project, I developed and tested a deep learning model that detects structural faults in concrete. Key learnings include:

• Building and fine-tuning a deep learning model using ResNet18.
• Executing the full pipeline of data preprocessing, model training, and evaluation.
• Improving model performance by experimenting with different techniques.
• Understanding the limitations of the model through error analysis.

Conclusion

This project demonstrates how deep learning can be applied to real-world problems, such as fault detection in critical infrastructure. The model shows promising results and could be further refined for practical deployment.