Desheng Fortress, a typical example of ancient city wall building with loess rammed-earth fortifications, is strategically situated in Baoziwan Township, Xinrong District, Datong City, Shanxi Province. Built in 1549 CE (28^th year of the Jiajing reign during the Ming Dynasty), this fortress formed a critical node within the Ming Great Wall defensive network. Beyond its military significance, it served as a cultural crossroads fostering Mongol-Han ethnic integration along China's northern frontier. Due to centuries of weathering and environmental impacts such as wind erosion and sand abrasion, the Desheng Fortress has developed multiple structural deteriorations across its ancient walls, including partial collapses, cracks, surface spalling, and biological colonization. In accordance with core conservation principles of minimal intervention, reversibility, material compatibility, and distinguishable repairs, the local cultural relics protection bureau implemented pilot restorations in selected areas using two distinct methodologies: adobe brick masonry and traditional layer-by-layer ramming techniques^[1-3]. The quest for a non-destructive repair performance detection technology for Desheng Fortress presents both technically challenging and cultural significance. By leveraging advanced scientific methods and fostering interdisciplinary collaboration, it is possible to develop a solution that meets the needs of both preservationists and historians. Such a technology would not only protect the fortress but also serve as a model for the conservation of other historical sites worldwide.

Terahertz (THz) detection technology exhibits exceptional penetration capabilities when interacting with dielectric materials and non-polar substances, while ensuring no damage to the inspected objects. Its high resolution, strong penetration power, and excellent anti-interference properties make it highly suitable for diverse applications, including cultural heritage preservation^[4-7], security inspections^[8-11], composite material identification^[12-15], and medical diagnostics^[16-17]. These studies confirm that THz technology can simultaneously capture structural information and analyze material composition, earning significant recognition in the field of heritage conservation. Furthermore, compared to ultrasonic detection methods for cultural relics^[18-19], THz waves exhibit superior resolution (≤100 μm for THz waves versus ~1 mm for ultrasound). Additionally, unlike X-ray techniques,^[20] THz waves demonstrate enhanced penetrating capabilities in non-metallic materials. For example, the Krügener group in Germany employed THz technology to investigate internal cracks in a stone circular relief from the Lower Saxony State Museum in Hanover. By analyzing differences in THz time delays, they accurately measured hidden cracks ranging from 5 mm to 7 mm^[21] and effectively detected defects beneath 16th-century glazed ceramic layers^[22]. The Fukunaga research team utilized THz spectroscopy to mineral pigments identification, revealing distinct transmission peaks in the THz frequency band for most pigments (excluding carbon black and earth pigments). These spectral characteristics are determined not only by the presence of primary elements but also significantly influenced by molecular structures^[23]. In a complementary study, the Durand group utilized THz waves to analyze seals, successfully resolving internal structures and evaluating conservation states by localizing defects and cracks in depth^[24]. Additionally, the Huang et al. systematically identified 18 minerals with characteristic THz absorption peaks, including pyrophyllite (1.10 THz) and chamosite (1.15 THz). They also provided experimental evidence regarding the optical properties and pyrolysis mechanisms of silicate minerals within the THz frequency range^[25]. THz spectroscopy has been also used to assess disease-induced deterioration in stone cultural relics, providing critical insights into pathological degradation processes. Studies on lithological THz spectral features have revealed that different rock types (e.g., granite, sandstone, and limestone) exhibit distinct absorption coefficients and refractive indices in the THz band, enabling rapid classification and provenance tracing^[26-27]. This capability is especially significant for analyzing stone artifacts or architectural relics, as THz technology can non-invasively assess weathering patterns and structural integrity. Moreover, the penetration depth of THz wave's (up to several millimeters in non-polar materials) facilitates subsurface defect detection in materials such as terracotta or adobe walls, as validated through comparative studies on aged earthen walls^[28].

THz technology provides a transformative and innovative approach to cultural heritage analysis by integrating non-destructive valuation with molecular-level insights. As standardized spectral databases expand and instrumentation precision improves, THz technology is expected to become a fundamental pillar in interdisciplinary conservation science.

Passive infrared imaging technology operates by detecting the infrared radiation naturally emitted by objects, enabling the characterization of temperature and thermal distribution fields across target surfaces. This technology has demonstrated remarkable effectiveness in the nondestructive detection of diseases in cultural relics^[29-30]. By analyzing surface temperature field anomalies captured through thermal imaging, the technique allows for the identification of subsurface flaws without physical contact. Characterized by its noninvasive, intuitive, noncontact, and large-scale scanning capabilities, Passive infrared imaging stands as one of the most advanced and effective methods for thermal fault diagnosis and detection currently available^[31-33].

In this study, the conservation status of three critical sections of the Desheng Fortress city walls was systematically evaluated through the integrated application of Terahertz (THz) and infrared thermal imaging technologies. This non-destructive analytical approach provides scientific data references for heritage conservation management, including internal defect characterization, restoration interventions, structural reinforcement, and preventive maintenance strategies.

In the paper, we take Desheng Fortress (Ming Dynasty) as the research subject (Fig. 1, color online).

Three representative wall sections and detection areas of the Desheng Fortress city walls are illustrated in Fig. 2: Detection area 1 (Fig. 2(a), color online) : Adobe brick masonry repaired section, located at position F in Fig. 1. An adobe brick wall refers to a constructed type utilizing unfired earthen bricks. For surface restoration, adobe bricks were manufactured from locally sourced refined soil, and bonded using homogeneous refined soil mud. Detection area 2 (Fig. 2(b), color online) : Well-preserved eastern section situated between positions F and G on the eastern fortress face in Fig. 1. The east-facing orientation minimizes direct wind erosion, contributing to its well-preserved state. Detection area 3 (Fig. 2(c), color online): The traditional layer-by-layer ramming repaired section at position G in Fig. 1. Sequential compaction layers of lime-stabilized earth were applied over the original rammed earth stratum. The stabilization material comprised raw lime chunks and refined local soil, mixed in a 2∶8 weight ratio of raw lime to soil. Among them, the soil material must contain 15% clay particle sand and be sieved. After sieving, at least 90% of the quicklime should pass through a 40-mesh sieve (≤425 μm), and the loess material must be sieved to ensure particles ≤3 mm in diameter. Rammed earth materials must be compacted on the same day they are prepared. Prior to compaction, the rammed earth materials should undergo on-site wet mixing with water added at 16%−17% of the dry weight (i.e., 16%−17 parts water per 100 parts rammed earth material). Adjustments may be made based on ambient temperature and wind conditions. Manually mix the material in the field for 3−4 cycles to ensure uniform moisture distribution. The hydrated mixture must be used within one hour of preparation, and any remaining material cannot be reused. Continuous compaction of successive layers is strongly recommended. If completion within the same day is unachievable, the interval between layers must not exceed 12 hours. During the curing period of the rammed earth layer, compacted layers must be shaded to prevent rapid moisture evaporation. If necessary, apply light watering twice daily (morning and evening).

Terahertz time-domain spectroscopy (THz-TDS) is an advanced hybrid technique that integrates optical and electronic technologies, enabling the direct measurement of the time-dependent electric field waveform of THz pulses. Unlike conventional optical techniques, which only detect irradiance (the time-averaged squared electric field amplitude) without phase information, or electronic approaches limited to frequencies lower 100 GHz, THz-TDS uniquely combines ultrafast optical gating with antenna-coupled detection. This dual capability enables simultaneous measurement of both the amplitude and phase in THz electric fields, thereby providing direct access to the complex dielectric function (refractive index and absorption coefficient) of materials without requiring Kramers-Kronig relations^[34-35].

The THz-TDS system consists of four key components (Fig. 3(a) color online): (1) An ultrafast femtosecond laser: generates a pulse train (~100 fs duration) split into two beams. (2) A THz emitter: converts optical pulses into single-cycle THz transients (0.2−3 THz bandwidth). (3) A THz detector: measures the instantaneous THz electric field. (4) A Delay line: adjusts the optical path length of one beam to control relative timing between THz generation and optical pulses.

During operation, the laser beam is split into two paths: the first drives the THz emitter, generating a broadband electromagnetic pulse, while the second, delayed by a motorized mirror system, gates the detector in synchronization with the THz pulse arrival. By incrementally varying the delay time, the full THz waveform is reconstructed point by point through sampling measurement, a process enhanced by lock-in amplification to reduce noise (Fig. 3(b), color online). This sampling measurement reconstructs the complete time-domain electric field, E(t), which is subsequently Fourier-transformed to extract spectral amplitude and phase.

A photoconductive emitter (Auston switch) consists of a semiconductor substrate, such as high-resistivity GaAs, patterned with metallic electrodes that form an antenna gap (Fig. 3(b)). A DC bias (10 V) is applied across the gap. Upon illumination by an ultrashort laser pulse (<1 ps), photoexcited carriers accelerate under the bias field, generating a transient current. This time-varying current I(t), which is proportional to the derivative of the carrier velocity v(t), emits a broadband THz pulse (E_THz∝$\partial $I/$\partial $t).^[36]

In contrast, the photoconductive receiver mirrors the emitter’s structure but operates in reverse (Fig. 3(b)). A synchronized optical gate pulse illuminates the antenna gap of detector, temporarily increasing its conductivity. The incident THz field then drives the photoinduced carriers, generating a current proportional to the instantaneous THz electric field. For accurate field sampling, the semiconductor must have an ultrashort carrier lifetime (≤100 fs) to ensure the current response remains confined within the gating pulse duration. The measured current I(τ) is expressed as follows:

$ {I}(\tau ) \propto \int_{ - \infty }^{ + \infty } {\mu \cdot E_{{\text{THz}}} (t)} \cdot \phi (t - \tau ){\mathrm{d}}t \quad, $ (1)

where μ is the carrier mobility, Φ(t) represents the temporal gate profile, and τ is the delay between THz and optical pulses. With sub-picosecond carrier lifetimes, Φ(t) approximates a delta function, rendering I(τ)∝E_THz(τ).^[37]

In the THz transmission measurements system (Fig. 3(a)), the THz pulse propagates through the sample, undergoing absorption, Fresnel reflection, scattering, and phase retardation. The THz spectrum of the sample is then obtained. To ensure accurate measurements, the soil samples must have flat and parallel surfaces to minimize beam deviation caused by refractive index mismatches. Furthermore, the lateral dimensions of sample should exceed the cross-section of the THz beam (typically 1−3 mm in diameter for focused systems) to avoid edge diffraction or shadowing artifacts. Proper alignment of the sample at normal incidence to the THz propagation direction is crucial, as angular misalignment can induce lateral beam displacement via refraction, resulting in signal attenuation. Additionally, sample holders should include apertures with diameters greater than 5 mm to fully accommodate the THz beam spot, thereby preventing waveform distortions caused by partial beam blocking or aperture-induced diffraction. These precautions ensure that the measured signals accurately represent the intrinsic material properties rather than geometric artifacts (Fig. 3(c) color online).

Soil samples were collected from eight subzones across three test areas, with their spatial distribution illustrated in Fig. 2. The processing procedure involved homogenization of the regional samples using an agate mortar, followed by particle size fractionation through a 200-mesh sieve (75 μm aperture). For analytical preparation, 250 mg aliquots of the homogenized soil were precisely weighed and compressed into cylindrical test specimens (1.3 cm diameter) using a hydraulic tablet press under controlled pressure conditions (10−12 kPa). The samples were systematically labeled according to their geographical origin: No.1 (Detection area 1), No.2 (Detection area 2), and No.3 (Detection area 3).

The THz-TDS system shown in Fig. 3(a) demonstrates a peak signal-to-noise ratio (SNR) of ≥65 dB, spectral resolution of <3 GHz, and a THz spot size ranging from 1 to 3 mm. This system is driven by a femtosecond laser with a central wavelength of 780 nm, pulse width of 20 fs, a repetition rate of 80 MHz, and an output power of 1067 mW.^[38] In this study, the THz spectral scanning range was set to span 217−234 ps with incremental steps of 0.02 ps and an integration times of 0.5 s. Owing to its high SNR and broadband characteristics, the proposed methodology enabled highly sensitivity detection of soil sample responses. The acquired THz spectral data were analyzed using the physical model for THz optical parameters proposed by Dorney^[39] and Duvillaret^[40], thereby yielding the refractive index and absorption coefficient spectra of the soil samples.

$ n(\omega ) = 1 + \phi (\omega )\frac{{\mathrm{c}}}{{\omega d}} \quad, $ (2)

$ \alpha (\omega ) = \frac{2}{d} \cdot \ln \left(\frac{{4n(\omega )}}{{\rho (\omega ) \cdot {{(n(\omega ) + 1)}^2}}}\right) \quad, $ (3)

where n(ω) is the refractive index, α(ω) is the absorption coefficient, Φ(ω) is the phase difference between the sample and reference signals, c is the speed of light, ω is the angular frequency, d is the thickness of the samples, and ρ(ω) is the amplitude ratio of sample and reference signals, respectively.

The THz spectra of soil samples collected from three distinct test areas of the ancient city walls of Desheng Fortress are presented in Fig. 4 (color online). THz-TDS system is a phase-sensitive coherent detection technique capable of simultaneously acquiring both the amplitude and phase information of THz pulses. To establish a baseline, a reference (ref) signal was first recorded in the absence of any sample using transmission-mode THz-TDS. Subsequently, the time-domain spectra of the soil samples were measured under identical experimental conditions. Through this procedure, the THz time-domain spectra for the soil samples from all three test regions were obtained. The THz signal intensity corresponds directly to the amplitude of the THz electric field pulse. As shown in Fig. 4(a), the transmitted THz time-domain waveforms exhibit varying degrees of oscillation and attenuation compared to the reference waveform. These differences are attributed to the combined effects of THz wave reflection, dispersion, and absorption by the soil samples. Notably, the soil sample from the first test section (the adobe brick masonry repaired region) shows significant attenuation in terahertz time-domain signal. The main pulse peak amplitude decreases from the reference value of 198.96 μV to approximately 94.7 μV, along with an extended time delay. Compared to the reference signal’s main pulse peak position at 220.78 ps, the sample from this section shows a shifted peak position around 224.5 ± 0.02 ps. In contrast, the well-preserved second test section and the third test section (restored using the traditional layer-by-layer ramming technique) demonstrates closely matched THz time-domain waveforms. The measured THz main pulse peaks are approximately 109.7 μV and 117.3 μV, respectively, with peak positions at 223.8 ± 0.02 ps and 223.9 ± 0.02 ps. The relative time delays compared to the reference signal are 3.72 ps, 3.02 ps, and 3.12 ps for the three sections, respectively. These results indicates that compared to the adobe brick masonry restoration method, the data from the third section restored using the traditional layer-by-layer ramming technique closely align with those of the intact city wall in the second section. The material’s physical properties have essentially achieved the "restoration to original state" objective.

The complementary absorption coefficient data presented in Fig. 4(b) further elucidate the material properties. Specifically, the absorption coefficient quantifies a material’s capacity to absorb electromagnetic waves, which is influenced by parameters such as the refractive index, dielectric properties, and molecular structure. Notably, different materials, including metals, semiconductors, and insulators, exhibit distinct absorption characteristics within the THz frequency band. In paticular, its non-contact, high resolution and high sensitivity make it superior to the conventional detection methods, such as ultrasound, ground-penetrating radar and x-ray, in the field of cultural relics detection. Given the high sensitivity of THz waves, the THz absorption coefficient serves as a direct indicator of variations in the composition and structural properties of material used in the city walls. The first test area, restored using adobe brick masonry method, displays significantly higher absorption coefficients compared to other areas, suggesting substantial structural porosity from weathering effects. In contrast, the second and third test areas exhibit remarkable spectral congruence throughout the measured range. Notably, while all areas demonstrate spectral overlap below approximately 0.5 THz, the first test area manifests a pronounced absorption increase above 0.6 THz, whereas the second and third test areas maintain consistent absorption profiles up to 1.0 THz. Such data are generally low, indicating that the city walls have reached and slightly exceeded the original performance after practical repair. This empirical evidence strongly suggests that the traditional layer-by-layer ramming method demonstrates superior restoration efficacy on the city walls.

The THz refractive index spectra of soil wall samples across three test areas are presented in Fig. 4(c). THz waves may experience multiple refractions, reflections, and scattering among different components within soil wall samples, resulting in the refractive index of the soil being a complex and frequency-dependent parameter. In the THz band, the refractive index shows relatively stable characteristic, with its specific value influenced by factors such as soil composition, density, and particle size distribution. Significant discrepancies are observed between the first test area and the subsequent two regions, with the third area showing closer alignment to the second in terms of refractive index characteristics. Quantitative analysis at 0.56 THz (the spectral intensity maximum frequency) reveals average refractive indices of 2.224 for the adobe brick masonry repaired region, 2.098 for the well-preserved region, and 2.107 for the layer-by-layer ramming repaired region. These variations principally originate from distinct material compositions, as THz radiation exhibits exceptional sensitivity to molecular-level structural variations, enabling discrimination between different wall restoration techniques. Fig. 4(d) provides a clear and precise representation of the mean, standard deviation, and confidence intervals of the THz spectral parameters of the samples. This figure shows that the average refractive indices of the three areas are 2.224, 2.107, and 2.098, respectively. The confidence intervals remain within approximately 95% of the mean refractive index, and the standard deviations are indicated on the spectrum.

The infrared thermal imaging systems used in this study have the following technical specifications: infrared spectral range: 8−14 μm, temperature sensitivity: ≤0.025 °C, spatial resolution: ≤0.45 mrad, minimum image resolution of ≤58 μm, field of view (FoV): 23°×17°, operational temperature range: −40 °C−500 °C, and instrument accuracy: ±2 °C. Infrared thermal imaging systems operate by detecting the target's infrared radiation energy distribution through a synchronized detection mechanism. The integrated system, comprising an infrared detector and optical focusing components, first acquires spatial radiation patterns from the target surface. These radiation patterns are then precisely projected onto the photodetector array of the infrared sensor, where the incoming photons undergo photoelectric conversion to generate a digital thermal image. The resultant image provides a two-dimensional representation of the thermal gradient field across the target surface, with pseudo-color encoding visually distinguishing temperature variations. The color scale typically progresses from cool (blue) to hot (red) across the electromagnetic spectrum. Through advanced optoelectronic measurement techniques, the system establishes quantitative relationships between detected radiation parameters and surface temperature characteristics^[41-43]. Essentially, this technology performs dual physical conversions: first translating imperceptible mid-wave or long-wave infrared emissions into electronic signals, then algorithmically transforming these measurements into a color-coded visual display where chromatic variations directly correspond to thermal differentials across the imaged surface.

The visible light and infrared thermal images measured across three small zones in the first test area (Fig. 2(a)) of the Desheng Fortress city walls are presented in Fig. 5 (color online). The infrared thermal image (Fig. 5(b)) reveals negligible temperature variations along Line A (16.2 °C−16.9 °C), indicating uniform thermal distribution in the wall’s superficial layers and no structural anomalies. In contrast, Line B exhibits significant temperature fluctuations, with the highest recorded temperature reaching 17.5 °C. Analysis of the visible light image (Fig. 5(a)) suggests that surface cracks in the zone serve as seepage pathways, enabling localized heat absorption. These cracks likely result from mechanical damage caused by uneven stress distribution in the structure. Stress-related phenomena such as settlement, vibration, and tilting can induce cracking, faulting, and textural degradation while simultaneously promoting capillary water infiltration. Capillary water carries soluble salts (e.g., NaCl, Na₂SO₄) from soil into the wall matrix. Subsequent evaporation during temperature fluctuations leads to salt crystallization within wall fissures. The hydration expansion of these crystallized salts further exacerbates wall cracking. An uneven temperature distribution is observed in the region between Lines B and C, likely due to a local depression in the city walls. Comparative analysis of visible and infrared images (Fig. 5(c) and Fig. 5(d)) indicates that anomalies between Lines A and C primarily arise from material heterogeneity in the rammed-earth fill used during historical repairs. Major anomalies in Fig. 5(e) and Fig. 5(f) cluster near the midpoint of Line A (14.5 °C), where vegetation root systems have formed subsurface aqueducts. These conduits facilitate water accumulation in cracks (evidenced by localized cooling in thermal imagery) while creating microhabitats for microbial growth and reproduction. Post-mortem microbial degradation of plant and animal matter produces humic acids, which accelerate rock decomposition through chemical weathering. These findings indicate the need to address biological proliferation and associated physicochemical degradation mechanisms in conservation efforts to mitigate long-term structural risks.

Fig. 6 (color online) presents visible light and infrared thermal images of two minimally weathered, well preserved eastern wall segments in the second test area of Desheng Fortress (Fig. 2(b)). The thermal anomalies above Line A and below Line C in Fig. 6(b) are mainly caused by the detachment of the weathered layer on the city wall surface (Fig. 6(a)), with temperature differences exceeding 17 °C. Meanwhile, the thermal anomalies above Line B and below Line C in Fig. 6(d) are caused by the celadon mildew microbial biofilms (biological erosion) in the corresponding region (Fig. 6(e)). Photosynthesis activity from plants such as algae, lichens, and mosses induces rapid temperature increases, leading to localized high temperatures. Additionally, the derivation of fungi also forms a complicated biological community on the city wall surface, which accelerates the biophysical weathering of cultural relics. Residual capillary water traces further exacerbate these degradation processes.

Fig. 7 (color online) presents visible light and infrared thermal images of three small zones in the third test area of Desheng Fortress (Fig. 2(c)). The infrared thermography reveals relatively uniform thermal profiles across these regions (ΔT≤0.56 °C), indicating structural continuity. Visual inspection confirms successful material compatibility between the restored and original wall sections in terms of color, texture, and bonding integrity, achieving the principle of "authenticity in preservation." Nevertheless, capillary moisture infiltration persists 1−3 m above ground level in Zones 2 and 3 (Fig. 7(d) and Fig. 7(f)), with incipient cracking observed in Zone 2.

In summary, while both the adobe brick laying technique (Detection area 1) and the traditional layer-by-layer ramming method (Detection area 3) enhanced structural stability, the layer-by-layer ramming method demonstrates superior conservation outcomes. This approach achieves three critical advantages:

(1) Material compatibility: Seamless integration with the original wall fabric through matched physicochemical properties (Fig. 4);

(2) Degradation resistance: Effective suppression of biodeterioration (e.g., microbial colonization) and capillary water infiltration by minimizing hygrothermal gradients (Fig. 6 and Fig. 7);

(3) Aesthetic fidelity: Preservation of historic visual continuity through texture and color alignment (Fig. 6 and Fig. 7).

The traditional layer-by-layer technique thus represents a sustainable solution that balances structural reinforcement with heritage authenticity, offering a replicable model for earthen architectural conservation.

This study evaluated three sections of Desheng Fortress walls (well-preserved, adobe brick masonry restoration and layer-by-layer ramming repaired) using integrated THz spectroscopy and infrared thermal imaging. The results reveal that adobe brick masonry, despite providing structural reinforcement and surface weathering resistance, introduced material incompatibilities—particularly in thermal properties—leading to structural fissures, moisture migration, and biological colonization, which accelerate weathering. Conversely, traditional layer-by-layer ramming technique achieves superior conservation through material compatibility, structural density, and resistance to biological and capillary degradation. The dual-method approach proves effective: infrared thermal imaging quantifies surface conditions (e.g., moisture, thermal variation), while THz spectroscopy enhances analysis by penetrating materials to detect internal molecular states and spectral signatures, enabling precise component identification. The synergy of these techniques offers non-destructive, multi-layered evaluation, with THz technology providing unique advantages for depth-resolved diagnostics. In summary, this work presents a novel framework for evaluating heritage restoration performance, offering a nondestructive, scientifically validated tool for cultural heritage conservation.