Enhancing Concrete Damage Detection by Optimizing ConvolutionalNeural Networks for Efficient Environmental Monitoring

Environmental, earthquake, and unforeseen loads can all cause building damage. Advanced damagedetection technologies use cameras mounted throughout the structure to take images, which are thenprocessed by deep learning algorithms to determine damage types and locations. This information isforwarded to a centralized processing system for rapid repair. In the field of artificial intelligence, theresearch focuses on identifying an appropriate design for deep neural networks employed in damage statecategorization. This paper presents a novel approach to concrete crack detection utilizing a minimalisticConvolutional Neural Network (CNN) architecture. The CNN model is intentionally designed to be smallerin comparison with pretrained architectures, featuring a reduced parameter count and shallower layer depth.By leveraging this compact design, the paper aims to achieve a balance between model performance andcomputational efficiency. The decision to develop a custom CNN architecture is motivated by the need fora solution tailored specifically to the task of concrete crack detection, optimizing for effectiveness withinthis domain. Through rigorous evaluation and validation, the proposed CNN model demonstrates superiorperformance in concrete crack detection while maintaining computational feasibility. This researchcontributes to the advancement of infrastructure monitoring and maintenance practices by offering apractical and efficient solution for detecting structural vulnerabilities

Keywords:

Image classification , Convolutional Neural Networks (CNNs) , Concrete CrackDetection