Deep Learning Based Surface Crack Detection in Battledore of Darbhanga Fort

Heritage structures across the globe are degrading because of their age. In India, the Archaeological Survey of India (ASI) recommends preserving heritage structures to safeguard their structural integrity. This study aims to implement You Only Look Once version 8 (YOLOv8), a deep learning (DL) architecture, as a tool to detect and localize surface damages in the form of cracks. As a case study, the Darbhanga Fort, made of brick masonry and located in Darbhanga, Bihar, has been considered. Because of the absence of periodic maintenance, huge vegetation, cracks, and spalling are present. The custom YOLOv8 model has been trained on the dataset containing 4056 cracked images. Performance parameters like mean average precision (mAP), precision, recall, F1-score, and confusion matrix are obtained to check the efficacy of the model. The YOLOv8 model attains a mAP_0.5 of 98.6%, a mAP_0.5:0.95 of 87.9%, a recall of 97.5%, and a precision of 96.6%. The obtained results show that the deployed model is capable of sensing and localizing cracks present in the structure. An open-source crack dataset of the present study along with its bounding box labels has been published on the Mendeley database.