Automated Crack Identification Using Deep Learning Based Image Processing

Link for the published journal article: https://doi.org/10.1016/j.cemconcomp.2020.103781 Investments in managing the ageing civil infrastructure assets worldwide have exceeded those for developing new public infrastructure. Yet, the condition assessment of the colossal backlog of ageing concrete structures is often based on subjective visual inspection. The results largely depend on the experience and judgment of the operator. Thus, there is a need for more robust inspection techniques using sound scientific principles. This research deploys deep learning techniques for automated image processing and recognition of cracks in ageing concrete structures. This approach allows to overcome operator subjectivity and can be executed using a robot and drone-based inspection, especially in hard to access areas of structures. Results indicate that using the Keras classifier combined with IPTs achieved an accuracy of 97.96% and 96.85% for training and testing data, respectively. The quantification error was minimal, namely 1.5%, 5%, and 2% for computing the length, width, and angle of orientation of the crack, respectively. The proposed approach is a promising method that could be integrated into various automated structure inspection platforms.