Surface crack detection using Hierarchal Convolutional Neural Network

Cracks on surface walls may imply that a building possesses problems with its structural integrity. Evaluating these types of defects needs to be accurate to determine the condition of the building. Currently, the evaluation of surface cracks is conducted through visual inspection, resulting in occasions of subjective judgements being made on the classification and severity of the surface crack which poses danger for customers and the environment as it not being analysed objectively. Previous researchers have applied numerous classification methods, but they always stop their research at just being able to classify cracks which would not be fully useful for professionals such as surveyors. We propose building a hybrid web application that can classify the condition of a surface from images using a trained Hierarchal-Convolutional Neural Network (H-CNN) which can also decipher if the image that is being looked is a surface or not. For continuous improvement of the H-CNN’s accuracy, the application will have a feedback mechanism for users to send an email query on incorrectly classified images which will be used to retrain the H-CNN.