Enviar Registro por Correo electrónico: Deployable Deep Learning Models for Crack Detection: Efficiency, Interpretability, and Severity Estimation