About Capstone

Project Overview

The project is focused on building a classifier to detect cracks in images of concrete, which is a crucial task for structural health monitoring and inspection. The goal is to efficiently build a classifier using a pre-trained model that can distinguish between images of cracked concrete (positive class) and images of concrete with no cracks (negative class).

Problem Formulation

  • Positive class: Images of cracked concrete
  • Negative class: Images of concrete with no cracks

Project Objective

The objective of the project is to develop a classifier that can accurately detect cracks in images of concrete, which can be used for structural health monitoring and inspection.

Relevance

Crack detection is a critical task in various industries, including construction, civil engineering, and infrastructure management. Accurate detection of cracks can help prevent catastrophic failures, reduce maintenance costs, and ensure public safety.

Machine Learning Approach

The project will employ a deep learning approach using pre-trained models to build a classifier. This approach is suitable for image classification tasks, and pre-trained models can be fine-tuned to achieve high accuracy on specific datasets.

Modules

The project is divided into 4 modules:

Module 1: Loading Data

  • Introduces the problem and dataset to be used for building an image classifier
  • Covers loading, manipulating, analyzing, and visualizing the image dataset

Module 2: Data Processing

Focuses on processing image data to prepare it for building a classifier using pre-trained models

Module 3: Building Classifiers

  • Covers building a linear image classifier using PyTorch
  • Covers building an image classifier using the ResNet50 pre-trained model in Keras

Module 4: Advanced Classifiers and Evaluation

  • In PyTorch, builds an image classifier using the ResNet18 pre-trained model
  • In Keras, builds an image classifier using the VGG16 pre-trained model and compares its performance with the ResNet50 model built in Module 3
  • Completes a peer review assignment