Characterization of Cracks in the Building Using Deep Learning Technique

Detecting and characterizing cracks in buildings using deep learning techniques is a crucial research area. Cracks in buildings could lead to catastrophic structural failures, which can be hazardous to human life and property. Deep learning techniques can aid in addressing this problem by enabling the detection and classification of different types of cracks with high accuracy. This study investigates the effect of image pre-processing on the performance of DL crack detection using a data set of 5000 images. The results showed that using a pretrained model with RGB weights does not affect the performance of the CNN model for detecting cracks in concrete structures. A CNN model is developed using the Keras Python package and pretrained VGG16. The original image data set was converted into five comparison sets using the SciKit Image Python package. The performance of the developed model in terms of accuracy and F1 values were all above 98%.