Concrete & Pavement Crack Dataset

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

标题：基于卷积神经网络的混凝土与路面裂缝检测
摘要：本数据集包含30000张混凝土及路面表面图像，分为裂缝（crack）与无裂缝（non-crack）两个类别。图像采集自尼日利亚博尔诺州的尼日利亚陆军大学比乌分校，由土木工程专业学生Omoebamije Oluwaseun为其毕业设计收集。图像采集设备包括DJI Mavic 2 Enterprise无人机（用于高空拍摄）及智能手机（用于普通窗口高度以下的拍摄）。数据集以RGB、JPEG格式存储，并统一调整至227×227像素分辨率。
内容：数据集包含两个文件夹："positive"与"negative"，分别存储带裂缝及无裂缝的混凝土表面图像。数据集中所有图像均为JPEG格式，采用RGB色彩空间，分辨率为227×227像素。
实用性：本数据集可用于训练和测试用于混凝土裂缝检测的卷积神经网络（Convolutional Neural Networks, CNNs）。作者基于本数据集搭建的模型准确率超过98%，本数据集仅可用于科研用途。若将本数据集用于任何用途，需正确标注原作者的引用信息。
详情：
来源：尼日利亚博尔诺州尼日利亚陆军大学比乌分校
采集者：Omoebamije Oluwaseun
格式：RGB、JPEG
分辨率：227×227像素
类别：裂缝（crack）、无裂缝（non-crack）
总图像数：30000张

搜集汇总

背景与挑战

背景概述

该数据集包含30,000张混凝土和路面表面图像，分为裂缝和非裂缝两类，图像来自尼日利亚，由土木工程学生收集，用于毕业项目，统一处理为227 x 227像素的RGB JPEG格式。数据集专为训练和测试卷积神经网络进行裂缝检测而设计，作者报告模型准确率超过98%，但仅限于研究用途，使用时需正确引用作者。