Intelligent Infrastructure Crack Detection Using MSEDBO-Optimized Deep Learning
Infrastructure surface crack detection is a vital task in structural health monitoring, directly influencing the safety, durability, and serviceability of civil engineering assets. Although deep learning methods have achieved notable success in automated crack detection, their performance is often constrained by inefficient hyperparameter tuning, susceptibility to local optima, and suboptimal feature extraction. This study addresses these limitations by proposing an intelligent optimization-driven crack detection framework. Limitations of Conventional Deep Learning-Based Crack Detection Traditional deep learning models rely heavily on manual or heuristic-based hyperparameter selection, which can lead to unstable training outcomes and reduced generalization performance. Moreover, commonly used optimization techniques may become trapped in local optima, resulting in inaccurate crack localization and increased false positive rates, particularly when dealing with complex backgroun...