An Integrated Method Using a Convolutional Autoencoder, Thresholding Techniques, and a Residual Network for Anomaly Detection on Heritage Roof Surfaces

The roofs of heritage buildings are subject to long-term degradation, resulting in poor heat insulation, heat regulation, and water leakage prevention. Researchers have predominantly employed feature-based traditional machine learning methods or individual deep learning techniques for the detection of natural deterioration and human-made damage on the surfaces of heritage building roofs for preservation. Despite their success, balancing accuracy, efficiency, timeliness, and cost remains a challenge, hindering practical application. The paper proposes an integrated method that employs a convolutional autoencoder, thresholding techniques, and a residual network to automatically detect anomalies on heritage roof surfaces. Firstly, unmanned aerial vehicles (UAVs) were employed to collect the image data of the heritage building roofs. Subsequently, an artificial intelligence (AI)-based system was developed to detect, extract, and classify anomalies on heritage roof surfaces by integrating a convolutional autoencoder, threshold techniques, and residual networks (ResNets). A heritage building project was selected as a case study. The experiments demonstrate that the proposed approach improved the detection accuracy and efficiency when compared with a single detection method. The proposed method addresses certain limitations of existing approaches, especially the reliance on extensive data labeling. It is anticipated that this approach will provide a basis for the formulation of repair schemes and timely maintenance for preventive conservation, enhancing the actual benefits of heritage building restoration.

Cette oeuvre a été publiée sous la license Creative Commons Attribution 4.0 (CC-BY 4.0). Il est autorisé de partager et adapter l'oeuvre tant que l'auteur est crédité et la license est indiquée (avec le lien ci-dessus). Vous devez aussi indiquer si des changements on été fait vis-à-vis de l'original.