Real Time Building Crack Visual Measurement System using Metaheuristics with Deep Learning Model

Abstract

Cracks in concrete allow aggressive chemicals to enter the reinforcement and cause corrosion, affecting reinforced concrete longevity. Crack identification is crucial to damage assessment. Visual examination is the most common concrete infrastructure monitoring method. Inspectors visually estimate flaws using skill, engineering judgment, and experience. However, this process is subjective, time-consuming, and requires access to numerous challenging structures. One progress hinges on improving or combining conventional digital image processing methods. Deep learning (DL) methods like CNN can now overcome image processing's crack detection limitations. This study introduces the Real-Time Building Crack Visual Measurement System utilizing Metaheuristics with Deep Learning (RBCVMS-MDL) model. RBCVMS-MDL detects construction cracks using DL principles. Three main steps are involved in RBCVMS-MDL. First, ResNet is used to build feature vectors. Salp Swarm Algorithm (SSA) also tunes ResNet method hyperparameters Finally, Radial Basis Function (RBF) can detect and classify cracks. RBCVMS-MDL outperforms other methods in crack image dataset performance validation.