Building Surface Crack Detections Using Deep Convolutional Neural Network (DCNN) Architectures

This paper examines the most common structural defect in concrete is surface cracking. Building inspections are carried out to assess the stiffness and tensile strength of a building. Crack detection is a crucial step in the inspection process since it helps locate cracks and assess the building’s condition. With the use of TensorFlow, several deep learning models, including VGG19, VGG16, and MobileNetV2, have been improved to recognize surface cracks. The files contain 40,000 photos of various concrete surfaces, both with and without cracks, each with a size of 227 by 227 pixels and an RGB color channel. One of the most cutting-edge vision model architectures, VGG16 is a Convolution Neural Network (CNN) with an accuracy of 99.62%. Dense Convolutional Network (DenseNet) is a deep network architecture used in deep learning (DL). 99.51% test accuracy can be attained by dividing the weights of the features collected from deeper layers among several inputs present in the same dense block and transition layers. The VGG19 architecture and VGG16, which have been tested with an accuracy of 99.62%, share a lot of similarities. MobilenetV2 has a 99.81% accuracy rate.