Evaluation of Damage Level for Ground Settlement Using the Convolutional Neural Network

In this study, a convolutional neural network (CNN)-based deep learning was applied to evaluate settlement of the ground. Firstly, the database of 1200 images was captured and labeled for three classes of damage levels. Seven CNN architectures were then selected for the transfer learning, in which the highest accuracy of approximately 96.11% for the testing set was observed from the DenseNet121 architecture. Herein, a comparison in terms of accuracy with various optimizers-algorithms for optimizing the loss function in machine learning-have been implemented in the DenseNet121 architecture. The goal of this study is to propose a better architecture with higher accuracy for practical applications in geotechnical engineering using the CNN technique. The results indicated that the DenseNet121 architecture using the Adam optimizer performed the most effectively with accuracies of 97.59%, 95.00%, and 96.11% on training, validation, and testing sets, respectively.