Cracked concrete surface image classification on low-dimensional image using artificial intelligence algorithms

Abstract

The project aims to create a Convolutional neural network (CNN) to detect and classify building cracks. Cracks are a key factor in determining how well-built a concrete structure is since they affect its sturdiness, utility, and safety. Due to its superior image processing capabilities, CNN is rapidly gaining traction as a credible option to replace manual crack detection. Cracks on the concrete surface are one of the earliest signs of structural damage, which is important for maintenance and can cause significant environmental harm. The first step in a manual examination is to sketch the crack and note the conditions. The manual approach is dependent on the specialist’s expertise and experience, resulting in a lack of impartiality in quantitative analysis. As an alternative, automated image-based crack detection is suggested where a variety of detection methods are available, such as k-nearest neighbors (KNN), support vector machines (SVM), decision trees (DT), artificial neural networks (ANN), and convolutional neural networks (CNN). These techniques will be used in this project. Positive crack and negative crack are two classes that make up the dataset that will be used with the mentioned strategies, and there are 20,000 photos per class. The images are resized into five different sizes (50×50, 35×35, 25×25, 10×10, and 5×5), and then the results are analyzed based on the performance of the techniques used in the project. It is concluded that the performance with low-resolution images is at par with that of high-resolution images. In addition, for the 50×50 sample image, the accuracy score of the classifiers (KNN, SVM, DT, ANN, and CNN) was (89, 98, 97, 94, and 99) % respectively, while for the 5×5 sample image, the value of the accuracy was (91, 90, 89, 92 and 95) % respectively.