Performance Analysis of InceptionV3, VGG16, and Resnet50 Models for Crevices Recognition on Surfaces

Cracks on surfaces can occur for a variety of reasons, including age, wear and tear, environmental factors, and structural issues. Cracks can occur on a variety of surfaces, including pavement, concrete, walls, and ceilings. Crevices majorly found on roads are always an indication of the quality of the road. Roads are major means of transportation. Any country’s development is indicated by the quality of roads laid. Also, the major problem faced by most of the countries is crevices on roads. These crevices/cracks are due to poor maintenance, usage of low-quality materials for construction, passage of heavy load vehicles for a long period of time. Improving the conditions of roads by following the standards is a solution but cannot be done overnight. Meanwhile we can reduce the effect of crevices on road accidents by identifying them. Different machine learning models can be used to construct an automated system which could detect the crack. The proposed work compares different machine learning models like VGG16, Inception, and Resnet50. From these models, we can identify the best suitable model for crack detection. Transfer learning is used to implement the Visual Geometry Group (VGG16), Inception, and Resnet50 models. These are pre-trained on ImageNet database. All the three models are compared to identify the accurate performing model for crack identification. It is identified that InceptionV3 outperformed the other two models.