From Sanxingdui to Shanxi Ancient Tombs: Innovative Applications of Hyperspectral Imaging in Cultural Relics Archaeology

三星堆遺址的考古發(fā)現(xiàn)持續(xù)為我們揭示古蜀文明的奧秘，而支撐這些發(fā)現(xiàn)背后的，是一項(xiàng)正在革新考古學(xué)研究方法的技術(shù)——高光譜成像。在最近的三星堆考古工作中，這項(xiàng)技術(shù)已成功應(yīng)用于青銅器等珍貴文物的無(wú)損檢測(cè)，幫助研究人員發(fā)現(xiàn)了許多被時(shí)間掩埋的歷史細(xì)節(jié)。

高光譜成像技術(shù)通過(guò)采集數(shù)百個(gè)連續(xù)波段的光譜信息，實(shí)現(xiàn)文物"圖譜合一"的精準(zhǔn)檢測(cè)，既能記錄空間形態(tài)，又能分析物質(zhì)組成，為考古研究提供多維數(shù)據(jù)支撐。本文以兩個(gè)案例為切入點(diǎn)，闡述高光譜成像技術(shù)在文物考古領(lǐng)域的創(chuàng)新應(yīng)用。

The archaeological discoveries at the Sanxingdui site continue to unveil the mysteries of the ancient Shu civilization, supported by a technology revolutionizing archaeological research methods—hyperspectral imaging. In recent Sanxingdui excavations, this technology has been successfully applied to non-destructive testing of precious artifacts such as bronze ware and ivory, helping researchers uncover historical details long buried by time.

Hyperspectral imaging captures spectral information across hundreds of continuous bands, enabling precise "image-spectrum integration" detection of cultural relics. It records spatial morphology while analyzing material composition, providing multidimensional data support for archaeological research. This article explores two case studies to illustrate the innovative applications of hyperspectral imaging in cultural relics archaeology.

三星堆文物，圖片來(lái)源網(wǎng)絡(luò) / Sanxingdui artifacts, image source: internet

「三星堆：探尋古蜀文明的奧秘」

「Sanxingdui: Unraveling the Mysteries of the Ancient Shu Civilization」

三星堆遺址作為20世紀(jì)最偉大的考古發(fā)現(xiàn)之一，其出土的青銅器等文物具有高歷史價(jià)值。然而，傳統(tǒng)分析方法如X射線熒光分析、電子探針等存在明顯局限：要么具有接觸破壞性，要么只能反映局部特征。中科院西安光機(jī)所聯(lián)合西北工業(yè)大學(xué)的研究團(tuán)隊(duì)將高光譜成像技術(shù)應(yīng)用于這一領(lǐng)域，實(shí)現(xiàn)了文物保護(hù)與研究的雙重突破。

As one of the greatest archaeological discoveries of the 20th century, the Sanxingdui site’s bronze ware, ivory, and other artifacts hold immense historical value. However, traditional analytical methods like X-ray fluorescence and electron microprobe have significant limitations—either being contact-destructive or only reflecting localized features. A research team from the Xi’an Institute of Optics and Precision Mechanics (Chinese Academy of Sciences) and Northwestern Polytechnical University applied hyperspectral imaging in this field, achieving breakthroughs in both cultural relic preservation and research.

遺址文物數(shù)據(jù) / Archaeological site artifact data

該團(tuán)隊(duì)構(gòu)建了一套完整的高光譜文物分析技術(shù)體系，在多個(gè)關(guān)鍵環(huán)節(jié)實(shí)現(xiàn)創(chuàng)新突破。

在數(shù)據(jù)采集與處理環(huán)節(jié)，開發(fā)了基于相對(duì)坐標(biāo)的區(qū)域拼接算法，通過(guò)精確計(jì)算重疊區(qū)域面積和多尺度特征匹配，大幅提升了多軌圖像拼接效率。該算法采用Dense SIFT算子進(jìn)行快速特征提取，再通過(guò)RANSAC算法消除誤匹配點(diǎn)，實(shí)現(xiàn)無(wú)縫拼接。

在特征提取方面，創(chuàng)新性地將空間注意力機(jī)制引入HRNet網(wǎng)絡(luò)架構(gòu)，通過(guò)特征圖分解和ResNet結(jié)構(gòu)聚合，顯著提升了高分辨率圖像的空間定位精度。同時(shí)構(gòu)建了金字塔3D殘差網(wǎng)絡(luò)，采用漸進(jìn)式特征維度增長(zhǎng)策略，實(shí)現(xiàn)了對(duì)812個(gè)波段光譜特征的深度解析。

該技術(shù)在實(shí)際應(yīng)用中展現(xiàn)出優(yōu)秀的性能，已成功處理821幀不同分辨率的高光譜圖像。通過(guò)對(duì)比實(shí)驗(yàn)，最終采用視覺(jué)映射模型的數(shù)據(jù)-決策級(jí)融合方案，使青銅器等目標(biāo)的平均識(shí)別準(zhǔn)確率達(dá)到0.84，ROC曲線分析顯示AUC值較傳統(tǒng)方法提升15%以上。

The team developed a comprehensive hyperspectral analysis system for cultural relics, making innovative advances in multiple key areas.

In data acquisition and processing, they created a region-stitching algorithm based on relative coordinates, significantly improving multi-track image stitching efficiency through precise overlap area calculation and multi-scale feature matching. The algorithm employs the Dense SIFT operator for rapid feature extraction, followed by the RANSAC algorithm to eliminate mismatched points, achieving seamless stitching.

For feature extraction, they innovatively integrated a spatial attention mechanism into the HRNet architecture, enhancing spatial localization accuracy in high-resolution images through feature map decomposition and ResNet-based aggregation. They also constructed a pyramid 3D residual network, adopting a progressive feature dimension growth strategy to enable in-depth analysis of spectral features across 812 bands.

This technology demonstrated exceptional performance in practical applications, successfully processing 821 frames of hyperspectral images at varying resolutions. Comparative experiments led to the adoption of a data-decision-level fusion scheme based on a visual mapping model, achieving an average recognition accuracy of 0.84 for targets like ivory and bronze ware. ROC curve analysis showed a 15%+ improvement in AUC values compared to traditional methods.

算法流程圖 / Algorithm Flowchart

「山西古墓：解讀地下的歷史篇章」

「Shanxi Ancient Tombs: Deciphering Buried Historical Narratives」

山西南部某古墓底部存在大量黑色殘留物，這些疑似古代文字或圖案的痕跡因年代久遠(yuǎn)和環(huán)境侵蝕，肉眼已難以辨識(shí)。北京建筑大學(xué)與山西省考古研究所聯(lián)合團(tuán)隊(duì)采用高光譜成像系統(tǒng)（波長(zhǎng)范圍400-1000nm，光譜分辨率0.6nm）對(duì)這一考古難題展開研究。

The bottom of an ancient tomb in southern Shanxi contained numerous black residues—faint traces suspected to be ancient writing or patterns, rendered nearly invisible to the naked eye due to age and environmental erosion. A joint team from Beijing University of Civil Engineering and Architecture and the Shanxi Provincial Institute of Archaeology employed a hyperspectral imaging system (wavelength range: 400–1000 nm, spectral resolution: 0.6 nm) to study this archaeological challenge.

古墓底部的照片，及實(shí)驗(yàn)區(qū)域的高光譜圖像 / Photos of the tomb floor and hyperspectral images of the experimental area

研究團(tuán)隊(duì)開發(fā)了完整的高光譜分析流程：首先通過(guò)輻射校正公式轉(zhuǎn)換數(shù)據(jù)，有效消除了儀器噪聲和光照干擾；然后篩選950個(gè)有效波段，利用主成分分析（PCA）濃縮信息；創(chuàng)新提出兩個(gè)特征指數(shù)公式優(yōu)化目標(biāo)識(shí)別；最后運(yùn)用多種形態(tài)學(xué)算法組合，生成15種處理方案，為考古人員提供了豐富的參考選擇。

從考古價(jià)值來(lái)看，研究提供的多版本形態(tài)學(xué)處理結(jié)果成功重建了古墓底部疑似文字的連貫圖案，這些圖案在傳統(tǒng)檢測(cè)手段下無(wú)法辨識(shí)。這一發(fā)現(xiàn)不僅為解讀該墓葬的文化內(nèi)涵提供了全新實(shí)物證據(jù)，更展示了高光譜技術(shù)在非接觸考古檢測(cè)中的巨大潛力。

The team developed a complete hyperspectral analysis workflow: first converting data via radiometric correction formulas to eliminate instrument noise and lighting interference; then selecting 950 effective bands and condensing information using principal component analysis (PCA); innovatively proposing two feature index formulas to optimize target identification; and finally applying a combination of morphological algorithms to generate 15 processing schemes, providing archaeologists with rich reference options.

From an archaeological perspective, the multi-version morphological processing results successfully reconstructed coherent patterns of suspected writing on the tomb floor—patterns entirely undetectable using traditional methods. This discovery not only offers new physical evidence for interpreting the tomb’s cultural significance but also highlights the immense potential of hyperspectral technology in non-contact archaeological detection.

數(shù)據(jù)處理流程 / Data Processing Workflow

光譜曲線 / Spectral Curve

「結(jié)語(yǔ)  / Conclusion」

高光譜成像技術(shù)正在為文物考古提供了新的技術(shù)路徑，幫助我們揭開歷史塵封的面紗。這項(xiàng)技術(shù)不僅守護(hù)著文物安全，更讓千年文明得以延續(xù)傳承。

Hyperspectral imaging is forging a new technological path for cultural relics archaeology, helping lift the veil of history. This technology not only safeguards artifacts but also ensures the enduring legacy of millennia-old civilizations.

案例來(lái)源 / Source：

1. Qiu, S., Zhang, P., Tang, X., Zeng, Z., Zhang, M. et al. (2023). Sanxingdui Cultural Relics Recognition Algorithm Based on Hyperspectral Multi-Network Fusion. Computers, Materials & Continua, 77(3), 3783–3800.

2. Yang, X., Hou1, M., Lyu, S., Ma, S., Gao, Z., Bai, S., Gu, M., Liu, Y., & Hou, M. (2018). INFORMATION EXTRACTION IN TOMB PIT USING HYPERSPECTRAL DATA. ISPRS - International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 2107-2111.