Ensemble-based approach using inception V2, VGG-16, and Xception convolutional neural networks for surface cracks detection

Abstract

: Manual road crack detection is time-consuming. However, deep learning-based solutions
are quick and accurate. Various deep learning-based convolutional neural networks (CNN) have been
recently proposed. This study implies a comprehensive assessment of the performance of inception
V2, VGG16, and Xception CNN utilizing the surface cracks dataset. The research approach comprises
four distinct steps. Training and validating these pre-trained models are necessary by immobilizing
certain foundational layers. The previously frozen layers are thawed during the second stage, and the
training and validation process is repeated. Subsequently, the performance of the model is evaluated.
To enhance the performance of the models in detecting surface cracks in dataset images, after
completion of the model training and validation process for both frozen and unfrozen layers, the
models are combined using the ensemble technique to increase the overall performance for surface
crack detection. The performance of the models, including inception V2, VGG16, Xception, and the
ensemble model, is evaluated using evaluation metrics including accuracy, precision, recall, and F1
score. The ensemble has the highest precision 99.97% and the highest recall 99.92%. along with the
highest accuracy 99.93% and F1 score 99.92%, compared to the other CNN models.