Concrete & Pavement Crack Dataset

Published: 10 March 2023| Version 1 | DOI: 10.17632/429vzbgmbx.1
Oluwaseun Omoebamije


Title: Crack Detection in Concrete and Pavement using Convolutional Neural Networks Summary: This dataset contains 30,000 images of concrete and pavement surfaces, classified into two categories: crack and non-crack. The images were obtained from the Nigerian Army University Biu in Borno state, Nigeria, and collected by Omoebamije Oluwaseun, a civil engineering student, for his final year project. The images were collected using a DJI Mavic 2 Enterprise drone (for the high-ups) and a smartphone (for the ones beneath the average window height). The dataset was saved in RGB, JPEG format and downsized to 227 x 227 pixels. Content: The dataset consists of two folders: "positive" and "negative", containing images of cracked and non-cracked concrete surfaces, respectively. Each image in the dataset is in JPEG format, with a resolution of 227 x 227 pixels in RGB format. Usefulness: This dataset can be used for training and testing convolutional neural networks (CNNs) for crack detection in concrete. The dataset has been used by the author to achieve over 98% accuracy on his model, and it can be used for research purposes only. The author must be properly referenced if the dataset is used for any purpose. Details: Source: Nigerian Army University Biu, Borno state, Nigeria Collector: Omoebamije Oluwaseun Format: RGB, JPEG Resolution: 227 x 227 pixels Classes: crack, non-crack Total images: 30,000


Steps to reproduce

The collection process involved taking images of concrete and pavement surfaces in and around the university campus. Images were continuously captured using the 'timed-shots' feature of the DJI Mavic 2 Enterprise drone, and the Open Camera app's 'interval shots' feature on the Samsung phone. They were then later sorted one after the other and downsized on PowerToys. Additional information about the data can be found in the dataset description.


Machine Learning, Image Classification, Deep Learning