A deep transfer learning-based framework is proposed for automatic damage detection and classification in historical buildings and contributes to the development of data-driven decision-support tools for digital documentation and condition assessment in heritage conservation.
Abstract
Historical buildings are important cultural assets that reflect the identity of cities and preserve the collective memory of societies. However, these structures are increasingly exposed to environmental degradation and human-induced impacts, making their systematic documentation and condition assessment essential for effective conservation strategies. Recent advances in artificial intelligence have provided powerful tools for image-based analysis in the field of heritage preservation. In particular, transfer learning enables the adaptation of pre-trained deep learning models to domain-specific tasks with limited labeled data. In this study, a deep transfer learning-based framework is proposed for automatic damage detection and classification in historical buildings. A new near-balanced dataset of 20,000 images spanning six deterioration categories was developed and made publicly available. Ten convolutional neural network and transformer architectures pre-trained on ImageNet were systematically compared under a unified Bayesian optimization protocol. Experimental results on a held-out test set show that EfficientNetB3 achieves the highest classification accuracy (97.65%), while AlexNet obtains the lowest performance (83.89%); the validation set was used exclusively for hyperparameter tuning. The results demonstrate that transfer learning-based models can effectively identify visually observable deterioration patterns and provide reliable support for automated documentation processes. The proposed framework contributes to the development of data-driven decision-support tools for digital documentation and condition assessment in heritage conservation.
Architectural heritage connects us to the cultural achievements of past civilizations. Patan Durbar Square in Nepal is home to many such structures, yet identifying them remains a challenge for tourists. This paper presents an automated monument recognition system built on the backbone of convolutional neural networks (CNNs). A dataset of 1832 images of 9 important monuments from Patan was created, and build a detection system with MobileNetV2, a light-weight CNN, to detect monuments in Patan Durbar Square with a near-perfect F1 score of 98.94%. The approach utilizes transfer learning to adapt the model to local architectural styles. A detailed ablation study is performed to determine the optimal network design and augmentation strategies. Class-wise performance is further analyzed to verify robustness against visual occlusion and similarity. Finally, the model is deployed as a mobile application using Flutter and the FastAPI framework. This work demonstrates the viability of lightweight CNNs for real-time cultural heritage preservation.
Santosh Giri, Jebish Purbey, S. Adhikari et al.· IAES International Journal o...· 0 citations
The Indian heritage sites represent several centuries of history, architecture, and diversity of various cultures. Creative approaches are required to secure accessibility and awareness, and to maintain and promote these places. Traditional methods of heritage interpretation often fail to deliver a detailed, scalable, and automated experience because they rely on human guides, fixed displays, or limited mobile applications. This paper proposes a deep learning-based system that integrates a virtual tour system and heritage site classification. We predict the 55 Indian monuments using the datasets (Kaggle + GitHub) with transfer learning using EfficientNetB0 and ResNet50. To improve generalisation, a unified dataset of 1,091 images was processed and augmented. The trained models achieved good classification performance and high validation accuracy (EfficientNetB0: 99.39%; ResNet50: 99.08%). To enhance cultural interaction, we combine the classifier with a virtual tour module that captures the inside and outside views of the architectural site, along with textual descriptions in JSON files, with each landmark having its own. Users can upload an architectural image, label the monument, and automatically navigate its interior via a structured virtual tour using a web-based GUI (React frontend + FastAPI backend). The proposed system addresses gaps in existing systems by being automated, scalable, and accessible. Future developments for this system include multilingual support, AR/VR support, and implementation as an offline mobile/web application to increase digital heritage preservation.
Deepti Singh, Raghavender K.V, O.Obulesu3 et al.· DMPedia Lecture Notes in Com...· 0 citations
Transportation networks are critical for emergency response after earthquakes, but national-scale bridge and viaducts inventories often lack vulnerability-related attributes such as material and structural system. This paper presents an image-based approach, developed within the SAFENET project, to automatically classify bridges/viaducts according to a practical Material-Structure (MS) labeling scheme that reduces sparsity compared to finer taxonomies that also include construction period. Using a Portuguese bridge image dataset, we compare three visual model families for Material-Structure classification: a ResNet-50 convolutional baseline, a self-supervised vision transformer (DINOv2-Large), and a contrastive vision encoder (CLIP). Models are evaluated with a strict 5-by-5 Nested CrossValidation (NCV) protocol with bridge-level splits to prevent information leakage across train and test sets. Results show that DINOv2 achieves the best overall performance, with a mean accuracy of 0.903, a macro-F1 of 0.773, and a weighted-F1 of 0.897, outperforming ResNet-50 and CLIP especially on minority classes. These findings support the use of self-supervised vision transformers to enrich bridge inventories from imagery and to provide scalable inputs for regional seismic risk assessment.
Tomás Oliveira, Rui S. Moreira, Feliz Gouveia et al.· International Conference on...· 0 citations
While neural networks represent a promising approach for evaluating sensor data to assess damage presence, location and severity, large amounts of data are required for training. However, the generation of experimental data is both labor-intensive and costly. Transfer learning is a well-established method in deep learning that enables the reuse of knowledge from pre-trained models to improve performance and reduce data requirements across various domains such as computer vision and natural language processing. Its application within the field of strain-based structural health monitoring (SHM) has received little attention from currently published literature.
This work develops a resource-efficient transfer learning approach for strain-based SHM. While the Convolutional Neural Network (CNN) model learns relationships between strain distribution and crack geometry in an aluminum beam from Finite Element (FE) data, fine-tuning adapts it to experimental conditions, accounting for factors such as measurement noise, increasing overall accuracy and robustness.
An encoder-decoder CNN (UNet) is initially trained with synthetic data from FE simulations. The model is then fine-tuned based on a smaller experimental Digital Image Correlation (DIC) dataset obtained from an aluminum beam subjected to a four-point bending fatigue test. For this purpose, the encoder part of the CNN is frozen, while parameters of the final layers of the decoder are updated. The approach is validated with respect to its accuracy, robustness and applicability for SHM systems.
The presented approach significantly reduces the experimental data requirements while improving damage detection performance for an aluminum beam under four-point bending. This study demonstrates that transfer learning enables efficient adaptation to real-world conditions, offering a cost-effective and scalable solution for data-driven SHM.
Finja Rentzsch Holm, T. Schalm, Jorge Luis Jiménez Aparicio et al.· e-Journal of Nondestructive...· 0 citations
Accurate classification of the seabed defects is crucial in safeguarding the safety, reliability, and maintenance of underwater infrastructure, including pipelines, cables, and offshore platforms. Classification of underwater images remains a challenging task due to issues such as low visibility, scattering, color distortion, and multi-scale variations in underwater environments. To address these challenges, this study proposes IEResViT, a novel fusion-based hybrid architecture that integrates a Depthwise Inception Convolutional Neural Network (CNN) with a modified Vision Transformer (ViT) framework. The proposed model replaces the traditional multilayer perceptron (MLP) block in the transformer encoder with a ResNet-based residual connection (ResMLP) to enhance hierarchical feature extraction and reduce computational complexity. The architecture processes images through parallel CNN and transformer branches to capture both local and global feature representations, followed by feature fusion and classification. The proposed model was evaluated on the Aquatic Defect dataset, AQUA20 dataset and Marine_PULSE dataset, achieving high classification performance with an accuracy of 98.88% 98.65% and 95.65% respectively, outperforming several state-of-the-art deep learning models while using fewer parameters and reduced computational cost. The results demonstrate that the IEResViT model provides an efficient and lightweight solution for reliable underwater defect image classification.
Underwater infrastructure, such as pipelines, communication cables, offshore wind farms, and marine habitats, is a crucial interface on the seabed. Natural processes and human activities may cause seabed defects, e.g., cracks and erosion scars, or sediment liquefaction areas or uncovered buried utilities, over time. Early and precise diagnosis of these defects is critical in preventive maintenance, environmental safety, and navigation. Traditional seabed relies on manual interpretation, and its time consuming and labour-intensive. Recent advancements in machine learning help in automated detection. The graphical abstract presents a deep learning model to classify the aquatic defect images. This article presents the IEResViT model, which uses two parallel branches: a depthwise inception CNN branch that extracts local spatial features, and a ResMLP ViT branch that captures global dependencies through multi-head self-attention mechanisms. The extracted local and global features are then fused together, enabling enhanced representation of complex underwater defect image characteristics. The fused features are subsequently passed to fully connected dense layers, followed by a softmax classifier to predict the defect category. The Aquatic defect image dataset, AQUA20 dataset, and Marine_PULSE dataset are used to evaluate the performance of the model, and achieved an accuracy rate of 98.88%, 98.65%, and 95.65% respectively.
Sk Mahmudul Hassan, Kumar Sekhar Roy, Mehbub Alam et al.· Earth Systems and Environmen...· 0 citations
Satellite image categorization is a crucial component of applications such as land-use analysis, agricultural monitoring, environmental evaluation, and disaster response. This article presents a deep learning strategy utilizing Convolutional Neural Networks (CNNs) for the efficient categorization of satellite data. The suggested system incorporates effective data augmentation methods, contemporary CNN architectures, and refined training methodologies to improve accuracy and generalization capabilities. An experimental assessment on a benchmark dataset of satellite images demonstrates that our model exhibits robust performance, with a validation accuracy of 98.93% and elevated precision, recall, and F1 scores. The created model exhibits significant improvements in accuracy and computing efficiency relative to traditional classification methods. The results indicate that this methodology is exceptionally well-suited for satellite image classification and can be further enhanced through techniques such as transfer learning, hybrid deep learning architectures, and multimodal feature integration to address more complex real-world scenarios.
Potturi Chandrasekhar Raju, V. Burra· ITEGAM- Journal of Engineeri...· 0 citations