Detecting Fissures in Concrete Structures using CNNs and Transfer Learning

Abstract

Due to the inefficiency of manual inspection techniques to detect faults in concrete structures, image processing techniques, and machine learning models can be used to detect the cracks in these concrete structures without manual supervision. In this research, the use of Unmanned Ariel Vehicles for the effective inspection of concrete structures is discussed. Various binary classification models are trained to detect the cracks in concrete structures using the concrete crack classification dataset which consists of images having cracks collected from METU campus buildings. The results obtained are compared with different algorithms and the best model is selected for the proposed problem. The accuracies of the trained classifier models are improved by retraining the models after applying data augmentation techniques on the dataset.