A lightweight integrated diagnostic approach for Macao Architectural Heritage using 3D laser scanning and NDT

Abstract

This study tackles the challenges of documenting architectural heritage in Macao’s dense urban settings. Traditional methods are often inefficient and invasive for the complex materials of actively used temples. We propose a lightweight method that integrates portable 3D laser scanning with non-destructive testing (NDT), using the Kuan Tai and Tin Hau Temple in Taipa as a case study. A millimetre-accuracy 3D point cloud was used as a unified spatial framework to correlate multi-source inspection data. This framework spatially correlates multi-source data on elemental composition, thermal anomalies, and surface hardness. Our approach achieves high-precision digital documentation. It reveals that the primary degradation mechanism is sulfate-induced corrosion, caused by incense pollutants under high humidity. The entire system weighs under 4 kg, making it portable and minimally invasive. It provides a scientific basis for the preventive conservation and digital management of heritage in similar complex environments.

Introduction

In the field of architectural heritage conservation, traditional recording and inspection methods have long relied on manual surveying and contact-based detection techniques, which present significant limitations in terms of efficiency and safety. Although manual measurement can capture basic geometric information of structures, the process is time-consuming and highly dependent on professional expertise, making it unsuitable for large-scale heritage surveys. Furthermore, the accuracy of conventional surveying methods is limited. They often fail to capture intricate details of complex architectural components, such as carvings and dougong brackets, resulting in incomplete documentation that can undermine subsequent conservation decisions¹.

Three-dimensional laser scanning technology has played a pivotal role in architectural heritage conservation in recent years, owing to its exceptional data acquisition accuracy and efficiency. This technique enables the non-contact, rapid capture of large-scale 3D point cloud data from building surfaces, producing real-scene 3D models with millimeter-level accuracy. These outputs provide a highly reliable geometric basis for heritage documentation, deformation monitoring, structural analysis, and restoration design^2,3. Its high resolution and precision are particularly suited for the digital reconstruction of complex structures, delicate finishes, and concealed building components, significantly enhancing the completeness and scientific rigor of architectural heritage documentation⁴. However, some high-precision 3D laser scanning systems remain bulky and heavy, posing operational challenges in spatially constrained and poorly accessible heritage environments⁵. In response, compact and lightweight portable 3D laser scanning systems have emerged. These devices maintain high accuracy while offering improved portability, broader environmental adaptability, and lower cost, effectively overcoming the limitations of traditional large-scale equipment in narrow and complex heritage spaces while providing flexible on-site support for integrated pathology detection^6,7,8. Nevertheless, a review of existing research reveals a persistent gap. Although the combination of portable scanning and non-destructive testing (NDT) technologies shows potential, most current applications still focus on the independent use of single techniques or the simple overlay and visualization of multi-source data. There is a marked lack of systematic frameworks to integrate geometric, chemical, and physical data for quantitatively revealing the intrinsic mechanisms of material degradation. Specifically, current methods often fail to adequately establish deep correlations between high-precision spatial models and material composition, internal defects, and mechanical properties. This limitation hinders the transition from merely “recording phenomena” to “analyzing degradation mechanisms”.

In the domain of pathology detection, traditional methods such as material sampling and percussion testing often require direct contact with the structure, posing risks of irreversible damage to fragile historical materials⁹. NDT has been increasingly adopted in cultural heritage conservation^10,11. Portable NDT equipment, including handheld X-ray fluorescence (XRF) analyzers, infrared thermography (IRT) cameras, and hardness testers, has demonstrated significant technical advantages in building heritage preservation¹². XRF enables in-situ elemental analysis, providing non-destructive and rapid identification of material composition, historical provenance, and degradation mechanisms, thereby offering reliable support for conservation decisions^13,14. IRT captures surface thermal radiation distribution to non-invasively detect internal defects, moisture penetration, and thermal anomalies, allowing precise localization of potential risks^15,16. Portable hardness testers, meanwhile, facilitate on-site quantitative assessment of material mechanical properties^17,18. These devices combine non-invasiveness with portability, greatly improving inspection efficiency and data accuracy while minimizing physical intervention—aligning closely with the principle of minimal intervention in heritage conservation¹⁹. With ongoing advancements in portable sensing technology, NDT plays an increasingly critical role in the digital documentation and preventive conservation of architectural heritage, contributing significantly to the sustainable preservation of cultural heritage sites^20,21. However, current research utilizing portable NDT often focuses on the independent application of single techniques or the simple parallel presentation of multi-source data. There is a lack of a systematic framework to deeply integrate multidimensional information, such as chemical composition and physical properties, in order to quantitatively reveal the intrinsic mechanisms linking material degradation to specific environmental factors. This shortcoming limits a profound understanding and predictive capability regarding heritage deterioration, especially for pathologies induced by complex interacting factors such as the coupling of pollution and humidity.

Currently, 3D laser scanning technology is increasingly being integrated with NDT methods such as IRT, XRF, and hardness testing, promoting a shift toward multi-source heterogeneous data fusion in architectural heritage conservation. Research has achieved significant progress through different technical integration pathways. For example, Giuffrida et al. conducted a comprehensive digital recording and diagnosis of the Church of Santa Maria delle Palate in Italy²². By fusing laser scanning, photogrammetry, IRT, and spectroscopy, they not only obtained an accurate digital replica of the building but also performed a non-invasive investigation of interior wall paintings. This work led to the creation of a “digital archive” platform integrating conservation, monitoring, and public dissemination. Similarly, Fang et al. performed a high-definition survey of the Beamless Hall at Linggu Temple in Nanjing by integrating 3D laser scanning, close-range photogrammetry, and IRT²³. They successfully mapped pathological information, such as moisture distribution, onto the 3D point cloud model using pseudo-color representation, achieving spatial visualization and efficient expression of wall deformation and dampness. Furthermore, Lerones et al. developed a method based on 3D laser scanning data itself—specifically, analyzing laser reflectivity—to automatically detect and quantify large-area moisture zones in buildings, providing a specialized technical solution for non-invasive humidity diagnosis²⁴. These advancements demonstrate that integrated methodologies can not only achieve high-precision reconstruction of architectural surface geometry but also support multi-dimensional, in-situ, and non-destructive diagnosis of material conditions and internal defects. Consequently, they help establish a multi-scale pathology identification system spanning from macroscopic morphology to microscopic material constituents. With continuous improvements in sensor portability and precision, along with ongoing optimization of data registration and fusion algorithms, multi-technology collaborative analysis based on 3D laser scanning and NDT has become a vital research direction in heritage diagnostics. It provides comprehensive scientific support for the preventive conservation and precise intervention of architectural heritage.

As unique embodiments of Sino-Western cultural convergence, Macao’s temples exhibit distinctive cross-cultural composite characteristics in their architectural form, materials, and craftsmanship. This presents both specific challenges and opportunities for the heritage pathology diagnosis central to this study. These structures not only preserve the timber-frame system and spatial layout of traditional Chinese temples but also incorporate materials and techniques such as Western lime mortar and Portuguese glazed tiles, resulting in representative hybrid construction practices²⁵. Cultural integration profoundly influences material combinations and aging mechanisms. Instances include the mixed masonry of Chinese gray bricks with Western mortar and the coexistence of Lingnan-style polychrome paintings with Portuguese tiles. These lead to more complex physical properties and degradation processes of the building materials. This complexity is particularly pronounced within the temple’s microenvironment, where high humidity coexists with persistent incense pollution, giving the deterioration mechanisms a distinct site-specificity^26,27,28. Concurrently, Macao’s location in a high-humidity maritime climate zone means perennial warm and wet conditions exacerbate salt weathering in masonry, wood rot, and surface erosion. Frequent typhoons pose an additional threat to heritage preservation, subjecting temple structures to continuous and compound environmental pressures²⁹. Within this context, traditional inspection methods often struggle to achieve rapid, accurate, and non-invasive diagnosis. In contrast, integrated digital technologies offer a novel pathway for systematically revealing the degradation mechanisms of materials under specific conditions of use and environment^30,31. As a digital diagnostic method oriented toward preventive conservation, the combination of high-accuracy 3D laser scanning with multi-source NDT enables the simultaneous, non-contact recording of a component’s geometric, chemical, and physical states. This supports the development of a degradation analysis model based on the coupled relationships between environment, material, and use. Owing to their cultural hybridity, material complexity, and environmental specificity, Macao’s temples thus serve as an exemplary research setting for validating and advancing lightweight, integrated diagnostic methodologies for architectural heritage.

The lightweight integrated methodology proposed in this study achieves high‑precision digital documentation and pathological diagnosis through a core mechanism of spatial alignment and information coupling of multi‑source heterogeneous data. The high‑accuracy 3D laser scanning point cloud first provides a millimeter‑level unified spatial reference. This ensures precise geometric correlation for all subsequent inspection data. Building upon this, the elemental composition data obtained from XRF, the internal defects and moisture distribution revealed by IRT, and the mechanical property data measured with a hardness tester are collectively mapped onto this 3D model. This integration establishes a diagnostic model with multi‑dimensional attribute associations, termed the ‘spatial‑composition‑physical‑mechanical’ model. By analyzing the statistical relationships among these attributes, such as the correlation between sulfur‑enriched areas and hardness reduction, the methodology transcends the surface‑level observations possible with any single technique. It uncovers the intrinsic chemo‑physical coupling mechanisms of material degradation, thereby achieving a cross‑scale diagnosis that connects macroscopic morphological anomalies to their microscopic causative factors.

Methods

Research object

The Temple of Kuan Tai and Tin Hau in Taipa is situated adjacent to Chok Ka Village in northern Taipa Island, Macao. It is located on the mountainside of the southern foothills of Kun Iam Tong Hill, facing the historic Cross Gate waterway. This geographical setting reflects the connection between maritime folk beliefs and port location (Fig. 1)³². The temple was founded in 1717, during the 56th year of the Kangxi reign in the Qing Dynasty. It integrates the worship of Guandi and Mazu^33,34 and has undergone multiple renovations throughout its history, demonstrating the sustained vitality of community faith.

Fig. 1: Location analysis.

a Map of Macao; b satellite imagery of the study site; c surrounding architectural and road context. This schematic was created by the authors for this study to illustrate the location and surroundings of the case study site. a A regional map of Macao showing the location of Taipa Island. b Satellite imagery where the Kuan Tai and Tin Hau Temple on Taipa Island is marked by a red rectangle. c Contextual view depicting the surrounding architectural fabric and road network of the temple.

The selection of this case study is based on the specific requirements that its architectural characteristics impose on the inspection methodology. The temple consists of two parallel two-hall structures with flush gable roofs (Fig. 2a). Its most notable structural feature is the use of traditional Lingnan brick-timber and rammed earth construction, with load-bearing walls made of rammed earth up to 70 cm thick (Fig. 2b). This is a unique surviving example among Macao’s temples. This massive, monolithic earthen structure is highly sensitive to invasive inspection methods. Any drilling or sampling could cause irreversible damage to its structural integrity and historical authenticity. This fundamentally precludes the feasibility of destructive testing and establishes the exclusive applicability of non-destructive testing (NDT) for this project.

Fig. 2: Site documentation.

a Aerial drone survey; b interior view of the connected front halls; c detail of gray plaster decoration on the external wall. This composite figure comprises three panels of photographs taken by the authors during the field survey. The left panel a provides an aerial overview of the temple complex captured by an unmanned aerial vehicle (Unmanned Aerial Vehicle, UAV), showing its complete layout and roof structures. The central panel b details the substantial rammed-earth wall inside the building, showcasing its material texture and construction technique. The right panel c presents a detailed close-up of the gray plaster decoration on an external wall, highlighting its craftsmanship and current surface condition.

Furthermore, as a continuously used living heritage site, the temple remains active for religious rituals. The long-term presence of incense smoke, combined with Macao’s humid maritime climate, creates a distinctive environment for material degradation. Various degrees of weathering, efflorescence, and local spalling are already visible on the building surfaces (Fig. 2c). This persistent environmental action and the apparent pathological signs indicate that the material degradation is likely not merely superficial. It may involve profound changes in internal chemical composition and mechanical properties. Therefore, visual documentation alone is insufficient for assessing its long-term preservation state. There is an urgent need for a detection method capable of simultaneously acquiring geometric form, chemical composition, and physical performance to clarify its pathological mechanisms. The structural particularity of this temple, coupled with its status as a living heritage site, together constitute an ideal and urgent case for applying the lightweight integrated methodology described in this paper. This approach, which fuses 3D laser scanning with multi-source NDT technologies, enables in-situ, cross-scale diagnosis.

Methodological framework

The research strategy encompasses three core phases (Fig. 3). The first phase is data acquisition. Terrestrial laser scanning (TLS) and an unmanned aerial vehicle (UAV) were employed to obtain high-precision point clouds and orthophotos of the building’s interior and exterior surfaces. This established a millimeter-accurate 3D digital base. Concurrently, XRF, IRT, and a hardness tester were used to perform elemental analysis, thermal anomaly detection, and surface hardness measurements on wall areas showing pathological symptoms identified during the laser scanning. This enabled the non-destructive characterization of material chemical composition and physical properties. The second phase involved point cloud data processing. This included the registration, filtering, and trimming of multi-station scan data to construct a clean, unified 3D model. This model also served as the spatial reference for the multi-source inspection data. The final phase was data analysis. Here, the chemical, thermal, and mechanical data were fused and spatially correlated with the geometric model. This served to evaluate modeling accuracy, analyze material degradation mechanisms, and provide a scientific basis for conservation interventions.

Fig. 3: Workflow of the integrated diagnostic methodology.

This flowchart was created by the authors to illustrate the three-phase framework of the proposed lightweight integrated approach. The workflow progresses from left to right, where rectangular boxes represent the core operational phases: data acquisition, point cloud data processing, and integrated data analysis. Arrows indicate the sequential flow and interdependence between phases. The data acquisition phase illustrates the concurrent use of terrestrial laser scanning, UAV photography, and portable Non-destructive testing (NDT) devices. The subsequent phases detail the key processing and analytical steps, culminating in the generation of high-precision digital documentation, the analysis of degradation causes and correlations, and the establishment of a referential framework for studies in comparable regional contexts.

Equipment specifications and technical parameters

To achieve high-precision digital documentation and pathological diagnosis of the architectural heritage at the Temple of Kuan Tai and Tin Hau in Macao, this study integrated a suite of portable equipment with a total weight of approximately 4 kg. This integrated system consists of the following core instruments (Table 1).

Table 1 Technical specifications of the portable devices used in this study

A 3D laser scanner, the Leica BLK360, was employed. This device exemplifies lightweight design, with a total weight of only 1.0 kg and compact dimensions, facilitating handheld use or placement within the complex interior spaces of historic buildings. It has an effective scanning range of up to 60 m and provides a point cloud accuracy of 6 mm at a 10 m distance, enabling the rapid acquisition of millimeter-accurate overall point cloud data.

A UAV, the DJI Mini SE, was used. This drone is renowned for its extreme lightness, with a take-off weight of less than 249 g, allowing it to be operated in parts of Macao without formal registration. Equipped with a 1/2.3-in. CMOS sensor capable of capturing 12-megapixel photographs, it assisted in obtaining texture information from building rooftops and contextual environmental data.

A handheld XRF analyzer, the Bruker Tracer 5i, was utilized. Weighing approximately 1.3 kg, it brings laboratory-grade elemental analysis to the field. The instrument can detect elements from magnesium to uranium with detection limits at the ppm level. It allows for the non-destructive and rapid analysis of pollutant element accumulation, such as sulfur, on wall surfaces, providing key chemical data for investigating material weathering mechanisms.

A portable hardness tester, the Equotip 550 Leeb (Type D), was applied. This is an integrated, portable device supporting multiple probes. With an impact energy of 11 N m, it performs on-site measurements of the Leeb hardness value. These values can be converted into other hardness standards via built-in algorithms, enabling the quantitative assessment of mechanical property degradation on material surfaces due to weathering.

A handheld IRT camera from the FLIR E85 series was used. Typically weighing around 700 g with a thermal sensitivity of 0.05 °C, it operates in the long-wave infrared spectrum. This tool can non-invasively detect temperature distribution differences on building wall surfaces, thereby accurately locating thermal anomalies caused by dampness or voids. It serves as a key instrument for identifying potential pathological areas.

Field data collection

First, the Leica BLK360 3D laser scanner was used to conduct a comprehensive scan of both the interior and exterior of the building. A total of 12 scanning stations were set up to acquire high-resolution 3D point clouds and texture information covering the entire structure. This established a unified geometric datum for the subsequent spatial integration of pathological data.

Building on this foundation, a FLIR handheld infrared thermography (IRT) camera was employed to perform a full screening of the wall surfaces. By identifying areas with abnormal temperature distributions, the survey preliminarily located potential hidden defects such as water seepage and voids. Notably, low-temperature zones indicated areas of moisture accumulation.

Subsequently, 40 representative measurement points were selected on the rammed earth walls for refined analysis. A Bruker handheld X-ray fluorescence (XRF) analyzer was used to conduct in-situ elemental composition analysis at these points, with a particular focus on quantifying the accumulation of pollutant elements such as sulfur (S). Concurrently, a Leeb hardness tester (Type D) was used at the same points to perform surface hardness measurements, enabling a quantitative assessment of the material’s mechanical properties. The XRF measurement time per point was approximately 30 seconds. For the hardness tests, three consecutive measurements were taken at each point and averaged to enhance data reliability (Fig. 4).

Fig. 4: On-site detection using lightweight equipment.

This figure presents photographs taken during the in-situ detection process at the temple, demonstrating the practical use of the lightweight equipment suite. The images show field personnel operating handheld devices, including an X-ray fluorescence (XRF) analyzer, infrared thermography (IRT) and a Leeb hardness tester (LHT), to collect pathological data on the rammed earth walls. The entire system, with a total weight of under 4 kg, exemplifies the high portability and minimal operational footprint of the proposed methodology, enabling efficient, non-destructive inspection within the constrained spaces of the active heritage site.

Model data processing

Point cloud data processing was completed using Leica Cyclone Register 360 software. First, the raw point cloud data collected from 12 scanning stations were imported. Preliminary alignment was performed using a target sphere-based registration method. A total of six target spheres were deployed as control points, with manual intervention applied to refine the matching and ensure point cloud stitching accuracy. Subsequently, fine registration was conducted using the iterative closest point (ICP) algorithm, ultimately generating 24 inter-station connections.

After registration, the merged point cloud underwent denoising. A statistical outlier removal algorithm was applied to eliminate floating points and noise. A distance filter was also used to remove redundant data points located far from the main structure. The final result was a high-accuracy 3D point cloud model with a unified coordinate system and clean data. This model complies with both the international ISO 31000 standard and the Chinese Technical Standard for Digitalization of Historic Buildings (JGJ/T 489-2021)^35,36.

The spatial coordinates of all pathological detection points were recorded based on the point cloud model to ensure spatial consistency of the multi-source data. Elemental content and hardness data were exported for statistical analysis using OriginPro 2022. Normality tests were first conducted on the SO₃ content and hardness values. Subsequently, regression analysis was performed using a power function model to determine the correlation strength (R²) and significance level (p). Furthermore, a Pearson correlation matrix was employed to analyze the relationships among multiple elements, including SO₃, CaO, TiO₂, and MgO, and the hardness values. This aimed to reveal the chemomechanical coupling mechanisms of material degradation. A significance threshold of p < 0.05 was applied to all statistical analyses.

Data matching

The registration and fusion of multi-source data aimed to integrate anomaly regions identified in IRT images and optical images, along with point-based XRF compositional data, into a unified 3D point cloud model. This enabled the visual localization and spatial analysis of pathologies. The specific procedure was as follows. First, suspected defect areas, such as dampness, efflorescence, and spalling, were manually interpreted and annotated on the 2D images. The surface locations of XRF measurement points were simultaneously recorded alongside close-up photographs. Then, by manually selecting clear and stable homologous feature points shared between the 2D images and the 3D point cloud—for example, window corners and mortar joints—a mapping relationship from 2D to 3D space was established based on the principle of the collinearity equation. Finally, this relationship was used to project the annotated abnormal areas and XRF points onto the 3D point cloud. They were represented as textured overlays and attribute-tagged points, respectively. This method does not require complex sensor calibration and relies primarily on manual interpretation and basic software functions. It provides an efficient and intuitive means for the spatial integration of multi-source heterogeneous data, establishing a solid foundation for subsequent comprehensive pathological analysis.

Results

Overall registration results

A comprehensive data acquisition of the Temple of Kuan Tai and Tin Hau in Taipa was carried out using a lightweight 3D laser scanning system. Through the deployment of 12 scanning stations and the successful registration of 21 station connections, a complete point cloud model of the architectural heritage was constructed (Fig. 5). The registration results indicate a global point cloud registration strength of 90%. This reflects a high confidence level in the software-aligned output and a stable, reliable point cloud network (Table 2). The mean overlap reached 41%, significantly exceeding the industry-recommended threshold of 30%. This confirms that the station planning strategy effectively captured detailed regions of the complex structure. These areas include zones prone to data loss, such as under-eaves spaces and gray plaster decorations, ensuring comprehensive coverage of the architectural heritage.

Fig. 5: Software-processed point cloud registration.

This figure is a screenshot from the point cloud processing software, Leica Cyclone Register 360 (used under institutional license), visualizing the registration results. In this view, blue circles indicate the positions of individual scanning stations, and the turquoise dashed lines represent the registration connections established between them. The aligned point cloud data from these stations form a complete 3D model of the temple. The visualization uses color gradients or intensity to represent depth or reflectance, allowing for the assessment of registration completeness and geometric accuracy across the entire structure.

Table 2 Summary of overall point cloud registration quality

Accuracy analysis

The overall registration achieved a root mean square error (RMSE) of 9.2 mm and a mean absolute error (MAE) of 8 mm. The slightly higher value of RMSE relative to MAE is consistent with the mathematical expectation of squared-error averaging, indicating the presence of a limited number of moderately larger errors while still reflecting a highly concentrated error distribution with no significant statistical outliers. This result confirms that the registration quality exhibits good uniformity and stability overall. An RMSE of 9.2 mm fully meets the requirements for architectural heritage documentation, enabling accurate capture of geometric features of macro-elements such as beams, columns, and roofs, while also providing a reliable geometric basis for subsequent identification and quantification of surface pathologies.

Detailed analysis of connection quality

To further evaluate the distribution characteristics and local stability of the registration quality, a detailed analysis was performed on all 21 station connections (Table 3). The results indicate that the majority of the connections exhibited excellent registration accuracy. Among these, Connection 1 (between Setup8 and Setup9) performed particularly well, with an overlap of 95% and a mean absolute error (MAE) of only 3 mm, forming the core support for the high-precision foundation of the overall point cloud. Furthermore, most connections—such as 2, 4, 11, and 12—achieved MAE values better than the global average (<8 mm), while maintaining overlap rates above 30%, reflecting stable and reliable registration quality across most regions of the station network.

Table 3 Summary of point cloud registration quality by connection

It is particularly noteworthy that the maximum registration error across all 21 connections was controlled within 15 mm (Fig. 6). This metric further confirms the overall reliability of the point cloud alignment. As a key parameter for evaluating local registration accuracy, the strict control of the maximum error indicates that even in the worst-case regions, spatial deviations remain within the millimeter range, fully meeting the geometric accuracy requirements for detailed modeling of historic structures.

Fig. 6: Maximum error registration quality matrix.

This matrix diagram visualizes the maximum registration error between each pair of scanning stations. Rows and columns correspond to individual station numbers. Each cell’s color indicates the connection quality based on the maximum error: green represents a high-quality connection with a maximum error below 0.015 m; yellow indicates an intermediate connection with a maximum error below 0.015 m; and red signifies a connection where the maximum error exceeds 0.020 m. The predominance of green cells across the matrix demonstrates a high-strength registration network with robust inter-station connections, which ensures the overall geometric reliability of the point cloud model.

Analysis of local precision also revealed several noteworthy characteristics in individual connections. Connection 16 (between Setup1 and Setup2) exhibited a MAE of 14 mm, significantly higher than the global average. This is likely attributable to complex occlusion or longer scanning distances along this path. Although the error remains within an acceptable threshold and did not substantially impact the overall model after global optimization, this observation indicates potential for optimization in data acquisition for local regions. Furthermore, connections with low overlap rates (<30%), such as 14, 15, 21, and 23, maintained reasonable error control but were structurally less robust. In future similar projects, increasing station density or optimizing scanning angles is advised to enhance data redundancy and registration reliability.

The point cloud registration results of this study are highly consistent with the conclusions reported by Sha Liu et al. from their work on the heritage building of Shanxi Yuci Dacheng Hall^37,38. Key metrics, including a mean overlap of 41% and a mean absolute error of 9.2 mm, align with best-practice standards for digital acquisition in architectural heritage. Although minor local variations are present, the overall registration network demonstrates high consistency and stability. All accuracy metrics comply with the Grade 2 precision requirements specified in the Standard for Measurement of Ancient Buildings (CH/T 6005-2018) and the Technical Standard for Digitalization of Historic Buildings (JGJ/T 489-2021). These results further validate the reliability and standardization of the lightweight scanning methodology when applied in complex heritage environments.

Point cloud semantic segmentation

On this basis, the building point cloud was further subdivided into three main categories according to material and functional attributes: rammed earth walls, roofs, and decorative moldings. This classification system was designed specifically for the characteristics of Macao’s architectural heritage and effectively distinguishes building materials with different weathering mechanisms and pathology types. The process took into full account the geometric complexity of various components. For instance, high-precision identification and extraction were achieved for intricate features such as roof tiles and ridge decorations (Fig. 7).

Fig. 7: Workflow of point cloud semantic segmentation.

This workflow diagram, created by the authors, outlines the key stages in semantically segmenting the heritage point cloud. Rectangular boxes represent either the input/output data (the raw registered point cloud) or the core processing actions (classification, extraction). Arrows between the boxes explicitly indicate the sequential direction and logical progression of the processing steps. The workflow typically begins with the complete 3D model, proceeds through iterative classification and refinement based on geometric and material features, and results in the extraction of distinct, labeled components such as rammed earth walls, roofs, and decorative elements for targeted analysis.

Based on the semantic segmentation results, this study further established a correspondence between different architectural components and their characteristic pathological features. This associative framework is particularly important for understanding and diagnosing temple heritage in Macao with its Sino-Western composite construction characteristics. As the primary load-bearing and enclosure system, the deterioration of rammed earth walls is primarily associated with moisture absorption, salt crystallization, and the resulting surface powdering and friability. The pathology of roof tiles mainly manifests as glaze detachment, tile misalignment, or loss, which is closely related to long-term exposure to weathering, thermal stress, and biological colonization. The delicate decorative moldings are more susceptible to pollution deposition, pigment fading, and micro-crack development. These degradation processes are directly influenced by their material composition, surface orientation, and immediate microenvironment. The clear correspondence between components and pathologies establishes a clear analytical framework for subsequent quantitative pathology assessment and targeted conservation intervention.

The semantic segmentation of the point cloud provides critical data support for in-depth research on architectural heritage. It enables targeted pathology analysis and evaluation of different material components, such as assessing surface weathering on rammed earth walls or detecting misalignment and loss of roof tiles, thereby offering a quantitative basis for component-level preventive conservation. Moreover, by isolating non-critical elements, semantic segmentation simplifies the model and highlights the core features of the structure, significantly improving the clarity and professionalism of 3D models in visualization and interactive analysis. In summary, this approach enhances both the organization and utility of data and provides direct technical support for evaluating core heritage value and informing conservation decisions.

Semantic segmentation of the point cloud in this study aligns with the classification approach used by Wang et al.³⁹ for pathological information in the James Gymnasium, achieving fine-grained identification and categorization of architectural components in both cases. Furthermore, drawing on the ontological model proposed by Messaoudi et al.⁴⁰, semantic information was linked to the 3D geometric model, enabling visual management and cross-analysis of multidimensional pathology data.

Pathological findings on walls

Infrared thermography revealed significant surface temperature differences on the walls, with localized areas exhibiting distinct low-temperature anomalies (Fig. 8) compared to surrounding normal zones, indicating higher internal moisture content and identifying these as potential moisture accumulation areas. Along with field-observed pathologies such as cracking and spalling, these hygrothermally anomalous regions were selected as key areas for subsequent XRF elemental analysis and hardness testing. This approach aims to elucidate the intrinsic mechanisms of material degradation through correlative analysis of compositional and mechanical properties.

Fig. 8: Infrared thermography and corresponding field observations.

This composite figure presents a direct comparison between infrared diagnostic data and visual evidence from on-site investigation. The left panel shows the infrared thermography (IRT) image captured on-site, where color variations represent differences in surface temperature. Cooler areas (blue in common palettes) may indicate subsurface moisture or voids. The right panel displays the corresponding on-site investigation photograph of the same wall section, documenting visible surface conditions such as cracks or discoloration. Placing these images side-by-side demonstrates the integrated use of thermal imaging with direct observation to detect, correlate, and interpret pathological features on the heritage building.

In-situ measurements using XRF and a hardness tester revealed a significant correlation between the chemical composition and mechanical properties of the material (Fig. 9). Analysis demonstrated a strong negative correlation between Leeb hardness (HL) and SO₃ content (r = −0.53), along with a relatively strong positive correlation with CaO (r = 0.47). These correlations suggest that sulfur-bearing substances derived from incense combustion accumulate in the wall matrix and, combined with the loss of key cementitious components such as lime, contribute to the degradation of mechanical performance.

Fig. 9: Correlation analysis of elemental composition and hardness.

This correlation matrix visualizes the statistical relationships between the content of key chemical elements (e.g., SO₃ and CaO) and the Leeb hardness (HL) values measured on-site. Each cell in the matrix displays a colored circle at the intersection of a row (element) and a column (element or hardness). The color of the circle encodes the direction of the Pearson correlation coefficient: red indicates a positive correlation, while blue indicates a negative correlation. The size of the circle is proportional to the absolute value of the correlation coefficient, meaning a larger circle represents a stronger relationship. This visualization allows for the rapid identification of significant chemomechanical couplings, such as the strong negative correlation between sulfur content and surface hardness.

To quantify the relationship between material performance and pollutant accumulation, a regression model was established between SO₃ content and Leeb hardness (HL). The results indicate a strong negative correlation between the two variables, with a power-function regression yielding a goodness-of-fit R² of 0.76 (p < 0.001), demonstrating high statistical significance (Fig. 10). This outcome clearly shows that increased SO₃ content in the wall material significantly accounts for the reduction in hardness, statistically confirming that sulfation is the core mechanism responsible for the degradation of mechanical properties.

Fig. 10: Relationship between SO₃ content and material hardness.

This scatter plot illustrates the quantitative relationship between sulfur trioxide (SO₃) content and Leeb hardness (HL) measured on the rammed earth walls. Each data point represents a paired measurement from a specific location. The blue curve represents the fitted trend, revealing a strong negative correlation between the two variables. The goodness of this fit is indicated by a coefficient of determination (R²) of 0.76, statistically confirming that increasing SO₃ accumulation is a key factor in the mechanical degradation of the wall material under the studied environmental conditions.

As the primary load-bearing and enclosure system in architectural heritage, walls play a critical role in maintaining both the structural safety and historical authenticity of the building. In recent years, numerous scholars have focused on this key element, employing advanced technologies for pathology identification: Shuai Yang et al. utilized YOLOv8 to systematically identify holes, stains, discoloration, and cracks on traditional dwelling walls in Zhejiang⁴¹, while Xinrui Luo’s team achieved precise localization of weathering areas on brick walls of historical structures in Wuhan using YOLOv10⁴². Building on these efforts, this study further integrates lightweight 3D scanning and non-destructive testing technologies to conduct high-precision documentation and pathological mechanism analysis of temple walls in Macao. It was found that sulfation-induced corrosion triggered by incense-derived pollutants in high-humidity environments is the main degradation mechanism. This work extends previous research both in methodological innovation and mechanistic depth.

Quantitative pathology mapping based on semantic segmentation

To achieve precise quantification of the distribution and severity of material deterioration, this study performed a spatial correlation analysis. The analysis focused on an interior rammed-earth wall severely affected by long-term incense smoke. It was based on a high-precision point cloud model obtained via 3D laser scanning. Data on sulfur (S) content from XRF and surface mechanical properties from a hardness tester were integrated and mapped onto this unified spatial framework.As shown in Fig. 11, this spatial analysis is presented in three panels. Panel (Fig. 11a) displays the measurement points on the wall, which were located using the 3D point cloud model. Based on these points, an interpolated hardness distribution map was generated and is shown in panel (Fig. 11b). Similarly, panel (Fig. 11c) presents the interpolated SO₃ distribution map.This result provides direct spatial evidence for the degradation mechanism uncovered through the fusion of 3D geometric and material attribute data. The mechanism is that a high-humidity environment drives the migration of incense-derived pollutants (sulfur). This leads to sulfate crystallization, which subsequently causes the deterioration of the mechanical properties of the rammed earth material.

Fig. 11: Spatial coincidence analysis.

a Measurement points on wall derived from 3D laser scanning. b Hardness distribution. c Generated SO₃ distribution. This three-panel figure visualizes the spatial correlation between material composition and mechanical property through integrated data mapping. Panel a shows the schematic of the study area on an interior rammed earth wall, with the spatial framework and the 40 specific measurement points (red dots) derived from the registered 3D laser scanning point cloud. Panels b and c present continuous distribution maps generated by interpolating the discrete point data. Panel b visualizes the spatial variation of Leeb hardness (HL), while panel c shows the distribution of sulfur trioxide (SO₃) content. In both maps (b, c), the same 40 measurement points are marked as red dots for spatial reference. The color gradient in the maps (from blue to yellow) represents the value scale of the respective parameter: blue tones indicate lower values (softer areas or lower SO₃ concentration), and yellow tones indicate higher values (harder areas or higher SO₃ concentration).

Discussion

This study confirms the significant advantages of an integrated non‑destructive testing methodology based on portable devices for heritage conservation. Through the synchronized, collaborative operation of lightweight 3D laser scanners, handheld XRF analyzers, and Leeb hardness testers, the workflow achieved the in‑situ, simultaneous acquisition of multidimensional parameters. These parameters include geometry, chemical composition, and surface mechanical properties in complex environments. This highly integrated workflow overcomes the inefficiency inherent in traditional multi‑stage, multi‑process inspection approaches. It requires no sampling throughout the entire procedure, thereby maximally preserving the integrity and authenticity of the heritage structure. The method not only reduces the cost and time barriers associated with conventional surveying and laboratory analysis but, more importantly, enables the correlation and cross‑verification of spatial position, elemental distribution, moisture anomalies, and hardness data within a unified framework. Consequently, this study constructs a complete chain of evidence from phenomenological observation to mechanistic inference. It provides a solid data foundation for scientifically deciphering material degradation mechanisms triggered by the coupling of specific environmental conditions and human activities, such as incense burning. This highlights the core value of the integrated approach in enabling a transition for heritage pathology from “qualitative description” to “quantitative mechanistic analysis”.

This study integrated 3D point clouds, thermal imaging, and optical data to support multi-dimensional pathology analysis of the Temple of Kuan Tai and Tin Hau in Taipa. However, challenges remain in achieving pixel-level automatic co-registration of multi-source data due to differences in sensor imaging principles and resolutions. In practice, this study relied on high-accuracy point clouds as a unified spatial reference and manually mapped pathology areas from 2D images to the 3D model via feature points, enabling reliable spatial localization of defects. Future work should focus on developing more advanced registration algorithms to improve the automation and accuracy of multi-source data integration (Fig. 12).

Fig. 12: Visualization of diagnostic results using digital measurement techniques.

This schematic illustrates the integrated methodology for multi-dimensional pathology analysis of the Temple of Kuan Tai and Tin Hau. It shows the fusion of 3D point clouds, infrared thermography, and optical data. The core workflow involves using the high-accuracy point cloud as a unified spatial reference. Pathological areas from 2D images are mapped onto the 3D model via manual feature-point matching, enabling reliable spatial localization despite challenges in fully automated pixel-level co-registration due to sensor differences. The figure highlights the potential for future algorithm development to enhance fusion automation and precision.

This integrated method provides a rapidly deployable, lightweight diagnostic solution for buildings such as the Temple of Kuan Tai and Tin Hau in Macao. Its advantages are particularly prominent in living heritage scenarios characterized by high humidity, high temperatures, and spatial constraints, which are common for similar sites. First, in response to the physical limitations that the narrow spaces and intricate decorations of temples impose on large equipment, this system, with a total weight of only about 4 kg, offers exceptional on-site adaptability and operational flexibility. More importantly, this method directly addresses the typical sulfate-induced corrosion pathology triggered by the combined action of persistent incense burning and Macao’s humid maritime environment. Through the synchronized and rapid application of XRF, IRT, and hardness testing, it can provide in-situ, non-destructive confirmation of the humidity-driven accumulation of sulfur pollutants and their deteriorating impact on material mechanical properties. This clearly reveals the underlying environment–pollutant coupling degradation mechanism. Furthermore, as a completely non-invasive inspection workflow, it minimizes disruption to the religious activities of this living heritage site. Therefore, the method not only effectively tackles the specific conservation challenges in this case but also offers a replicable and generalizable integrated technical scheme. This scheme enables rapid diagnosis and mechanistic analysis for similar living heritage located in hot, humid, and crowded environments.