Crack Recognition in Masonry Structures: CNN Models with Limited Data Sets

Abstract

This research focuses on the task of Masonry Wall Crack Identification using limited data, employing state-of-the-art Convolutional Neural Network (CNN) models. The models investigated include VGG16, MobileNetV2, Xception, and DenseNet121. The dataset, consisting of 946 masonry wall images containing cracks, is used to evaluate the effectiveness of each model in this specific domain. The training set comprises 642 images, the validation set consists of 90 images, and the test set includes 214 images. The models are pretrained on large-scale datasets to extract robust features and are then fine-tuned on the masonry wall crack dataset. Among the models, DenseNet121 stands out, achieving a commendable accuracy of 85.98% in accurately identifying masonry wall cracks. This result underscores the efficacy of DenseNet121 for the challenging task of crack identification in masonry structures using limited data. This study not only contributes to the field of structural health monitoring but also emphasizes the practicality of employing CNN models for real-world applications, particularly in the critical domain of masonry crack identification.