Automatic evaluation of cracks with semantic segmentation model with U-Net and an Efficient Net-b2

Velasquez R.A.; Vanessa Mejia Lara J.; Velasquez R.A. · Proceedings of the 2023 IEEE 30th International Conference on Electronics, Electrical Engineering and Computing, INTERCON 2023 (2023) · DOI: 10.1109/intercon59652.2023.10326087

Resumen

The appearance of cracks in underground infrastructures when presenting gives meaning of weakening, Peru being a seismic country it's crucial to have the capability to locate the cracks, because it could produce an irreparable problem. It proposes a semantic segmentation model with U-Net and an efficient Net-b2 process with visual evaluation to carry out a detection system without contact, more fast and provide greater safety to workers. The system is based on digital image processing because it is a modern technique that allows combining various techniques in order to identify a particular object, in addition to creating a database with the processed data for subsequent control or monitoring. Our findings are an accuracy of 98.3% between the images labeled as cracks and non-cracks, and it indicates 97.61% by using filtering process according to the morphological operations and surfaces, better results were obtained in extracting the edges of the cracks.

Instituciones: Universidad Nacional de San Agustín de Arequipa; Universidad Tecnológica del Perú