Detecting Cracks in Concrete Surfaces using Convolutional Neural Networks and Resnet 50

The maintenance and lifespan of structures made of concrete depend on the early detection and timely treatment of cracks, which can compromise the building's strength and need costly repairs. The study presents a novel approach for crack identification on surfaces of concrete using deep learning algorithms. The suggested approach makes use of convolutional neural networks (CNNs) to computationally identify and classify cracks in high-quality photographs of concrete flooring. The process begins with compiling a sizable dataset of distinct surface photos of concrete with varying crack dimensions, trends, and ambient conditions. It is easier to do supervised learning when a collection of data is thoroughly annotated. A preliminary processing workflow is built up to standardize input data and strengthen the appearance of images in order to achieve the highest potential accuracy of the model. A pretrained CNN model structure called Resnet 50 is developed, modified, and utilizes transfer learning to leverage the information from already trained algorithms on substantial image collections. With robust generalization to unseen data and outstanding accuracy in crack identification, the trained model demonstrates superior abilities in this area. The model's usefulness is increased by capturing pictures of concrete substrates in a variety of lighting environments and perspectives. The recommended method shows resilience to disturbances, variations in illumination, along with other external variables, indicating its applicability in practical scenarios. When comparing the effectiveness of the developed model with classic methods for image analysis, deep learning operates better in crack searches. The results indicate that not exclusively is the approach suggested powerful, but it also holds the assurance of significantly reducing the duration and assets required for personal inspection and planned preventive maintenance. All things considered, this work presents a unique method for surveillance of infrastructure through presenting an adaptable and robust deep-learning system for concrete's outer crack identification. The amalgamation of Zorin OS 16.3, x86-64 architecture, NVIDIA GPU support, and Linux kernel 5.15.0-83-generic enhances the computational capacity needed to enable efficient development and operation of a deep-learning approach for detecting concrete crack recognition. The study outlines how the advocated deep learning approach could be applied in practical conditions.