Comparative Analysis of Deep Learning Algorithms Based on Surface Image Classification

A surface crack or road crack is the most common problem in civil infrastructures. Accidents may occur due to this serious problem. Manually detecting surface cracks is a lengthy and difficult process since the inspector must concentrate on particular areas and it is subjective. The proposed work aims to develop an automated system for analyzing images of surface or road specimens. The system is also able to classify cracks and provide a rationale based on the results. A proposed multilayered convolutional neural network (CNN) and ResNet152 are applied and trained on a large number of images. The total number of images in the dataset is 40,000. The dataset is divided into a training set (75% of the total image) and a test set (25% of the total image) to evaluate the models. We have applied data augmentation on the training set to increase the number of images. The proposed surface classification model achieved the accuracy of 99.84% and 99.80% by applying the proposed sequential CNN model and ResNet152 model respectively. This proper classification of surface or road cracks will reduce accidents on roads because they will be diagnosed and examined automatically. The proposed system will also take less time compared to manual observation.