Surface Crack Detection Using Data Mining and Feature Engineering Techniques

Surface cracks on any concrete structure can cause a lot of damage to its surroundings as well as to the people around it. These cracks need to be identified as soon as possible, especially on large and important structures like ancient structures thereby requiring a maintenance and quality inspection period. This requires a lot of money, resources, skilled labor, and time. Also, the detection of cracks is done manually which limits the places they can detect cracks and also makes it vulnerable to human errors. In a report published by the National Crime Records Bureau, they stated that more than thirteen lakh people had lost their lives due to the collapse of various structures. Therefore, to address these issues, this study offers a fully automated surface crack detection model using CNN with different variations and basic preprocessing with data augmentation to distinguish it into two classes, positive meaning crack exists and negative meaning crack doesn't exist. Six different CNN architectures were used in this paper. After comparing all the models, it was found that the second architecture gave the best performance amongst other four architectures.