Text this: Interpretability Analysis of Convolutional Neural Networks for Crack Detection