Hyperspectral Imaging Technology: Distinguishing and Detecting New and Aged Tea

「背(bèi)景 / Background」

在茶葉(yè)品(pǐn)質(zhì)鑒(jiàn)別領域，如(rú)何(hé)準確(què)區分新(xīn)的茶葉與老茶(chá)葉(yè)一直是個(gè)技(jì)術難題。傳統方(fāng)法(fǎ)主要(yào)依(yī)靠(kào)感(gǎn)官(guān)評定(dìng)和(hé)經驗判斷(duàn)，存(cún)在(zài)主觀性強，標準(zhǔn)不統一等(děng)問(wèn)題(tí)。而隨(suí)著(zhuó)高(gāo)光(guāng)譜(pǔ)成(chéng)像(xiàng)技術的(dí)發展，這一(yī)難(nán)題(tí)正迎來全(quán)新(xīn)的解決方案(àn)。

新(xīn)的茶(chá)葉與老茶(chá)在感官(guān)特(tè)征(zhēng)上存在明顯差異。新(xīn)鮮(xiān)茶葉色(sè)澤翠(cuì)綠(lǜ)鮮亮，葉(yè)麵(miàn)富有光澤，形態(tài)完(wán)整(zhěng)飽滿，特別(bié)是(shì)綠(lǜ)茶(chá)和白(bái)茶這(zhè)類未經發酵(jiào)的茶葉(yè)，其新(xīn)鮮特(tè)征更為突(tū)出。相(xiāng)比(bǐ)之下(xià)，老(lǎo)茶由於存放時(shí)間較長，色(sè)澤會逐漸轉暗，光澤度(dù)降(jiàng)低，葉片(piàn)可能(néng)出(chū)現(xiàn)碎裂或變形(xíng)。

從香氣(qì)特(tè)征來(lái)看(kàn)，新的散(sàn)發(fā)著清(qīng)新(xīn)怡(yí)人(rén)的花草(cǎo)果香，而老茶則呈(chéng)現(xiàn)出沉(chén)穩的(dí)陳香(xiāng)，藥香或木質(zhì)香調，尤其(qí)像普(pǔ)洱茶，白茶這(zhè)類(lèi)適合長(cháng)期(qī)存放的茶(chá)葉，其陳化特征更(gēng)為(wéi)顯著(zhù)。

這些(xiē)感(gǎn)官(guān)差異的(dí)本質(zhì)在(zài)於(yú)茶葉(yè)內(nèi)部(bù)化學(xué)成(chéng)分(fēn)的變(biàn)化(huà)。新的(dí)茶(chá)葉中茶多酚(fēn)，咖(kā)啡鹼(jiǎn)，氨(ān)基酸(suān)等活性物質含(hán)量(liáng)較高，而隨著(zhuó)時(shí)間推移，這(zhè)些成分(fēn)會(huì)逐漸氧化(huà)分解(jiě)，同(tóng)時產生(shēng)新(xīn)的(dí)次(cì)級代謝產(chǎn)物。

高光譜(pǔ)成像技術正是通過(guò)捕捉這些(xiē)細微的化(huà)學(xué)變化，實(shí)現對(duì)茶葉新(xīn)老的(dí)精準鑒(jiàn)別(bié)。該技(jì)術能夠(gòu)檢(jiǎn)測茶(chá)葉在不同波(bō)長下(xià)的光譜(pǔ)特征(zhēng)，通(tōng)過分析反(fǎn)射或(huò)透(tòu)射(shè)光譜的(dí)變(biàn)化，揭(jiē)示茶葉內(nèi)部(bù)的化學(xué)成(chéng)分(fēn)差(chà)異。這種方(fāng)法可(kě)用於茶(chá)葉(yè)品質的無損(sǔn)檢測，輔助茶(chá)葉(yè)的分(fēn)類(lèi)，分(fēn)級(jí)和市場交易。

In the field of tea quality identification, accurately distinguishing new tea from aged tea has always been a technical challenge. Traditional methods primarily rely on sensory evaluation and empirical judgment, which suffer from strong subjectivity and inconsistent standards. With the development of hyperspectral imaging technology, this challenge is now being addressed with a novel solution.

New and aged teas exhibit distinct sensory characteristics. Fresh tea leaves are vibrant green in color, with glossy surfaces and intact, plump shapes—especially in unfermented teas like green tea and white tea, where these fresh features are more pronounced. In contrast, aged tea, due to prolonged storage, gradually darkens in color, loses glossiness, and may exhibit leaf fragmentation or deformation.

In terms of aroma, new tea emits a fresh and pleasant floral or fruity fragrance, while aged tea presents a more沉(chén)穩(wěn) (mellow) aged aroma, medicinal or woody notes. This is particularly evident in teas suitable for long-term storage, such as pu-erh and white tea, where aging characteristics are more pronounced.

The essence of these sensory differences lies in changes in the tea's internal chemical composition. New tea contains higher levels of active substances like polyphenols, caffeine, and amino acids. Over time, these components gradually oxidize and decompose, while new secondary metabolites are produced.

Hyperspectral imaging technology captures these subtle chemical changes to achieve precise identification of new and aged tea. By detecting the spectral characteristics of tea leaves at different wavelengths and analyzing variations in reflectance or transmittance spectra, it reveals differences in internal chemical composition. This method enables non-destructive testing of tea quality, assisting in classification, grading, and market transactions.

「設(shè)備介紹 / Equipment Introduction」

在本次(cì)實(shí)驗中(zhōng)，我(wǒ)們采(cǎi)用400-1000nm波(bō)段(duàn)的國產高光譜(pǔ)相機進(jìn)行(háng)數據(jù)采(cǎi)集。

•光(guāng)譜範圍：400-1000nm

•光(guāng)譜分(fēn)辨率：優於2.5nm

•探(tàn)測(cè)器：CMOS

•空(kōng)間維有(yǒu)效(xiào)像(xiàng)元數：1920

•波段數：300

•視(shì)場角（FOV）：32°@f=17mm

•幀頻(pín)：128fps

配(pèi)套(tào)的專(zhuān)業分析(xī)軟件(jiàn)具備*的數據處理能力，包括(kuò)反射(shè)率(shuài)校正，輻射(shè)校正，濾波，降(jiàng)噪(zào)等。

軟件內置高(gāo)光譜(pǔ)數據裁切與拼(pīn)接算(suàn)法；具有(yǒu)光譜角(jiǎo)，監督(dū)分(fēn)類，非監督分(fēn)類等(děng)常(cháng)用算法，支(zhī)持(chí)用(yòng)戶自定(dìng)義波段進行運算，內(nèi)置NDVI，NDWI等(děng)25種以(yǐ)上常見(jiàn)植(zhí)被指數(shù)分析(xī)，為(wéi)數據(jù)解(jiě)析提(tí)供(gōng)多維度支(zhī)持(chí)。

In this experiment, a domestically produced hyperspectral camera with a 400–1000 nm wavelength range was used for data acquisition.

•Spectral range: 400–1000 nm

•Spectral resolution: Better than 2.5 nm

•Detector: CMOS

•Spatial dimension effective pixels: 1920

•Number of bands: 300

•Field of view (FOV): 32°@f=17 mm

•Frame rate: 128 fps

The accompanying professional analysis software features robust data processing capabilities, including reflectance correction, radiometric correction, filtering, and noise reduction.

The software also incorporates built-in algorithms for hyperspectral data cropping and stitching, spectral angle mapping, supervised and unsupervised classification, and supports user-defined band operations. It includes over 25 common vegetation indices (e.g., NDVI, NDWI) for multidimensional data analysis.

「反(fǎn)射(shè)率光譜(pǔ)曲(qū)綫(xiàn) / Reflectance Spectral Curve」

使用(yòng)50%反(fǎn)射(shè)率板標定(dìng)後，選(xuǎn)取(qǔ)新葉與老(lǎo)葉的(dí)特(tè)征區(qū)域(yù)進(jìn)行ROI分析，計(jì)算得出(chū)平(píng)均(jūn)反射率曲(qū)綫(xiàn),可以看到(dào)老茶(chá)葉的(dí)整體反射率整體(tǐ)低於新(xīn)的茶葉(yè)。

After calibration with a 50% reflectance panel, regions of interest (ROIs) were selected from characteristic areas of new and aged leaves to calculate average reflectance curves. The results show that aged tea exhibits overall lower reflectance compared to new tea.

「不同算法的茶葉區分(fēn) / Tea Differentiation Using Different Algorithms」

本實驗測試了歸一化差值植(zhí)被(bèi)指(zhǐ)數（NDVI）和(hé)監督分類(lèi)兩種(zhǒng)方(fāng)法。

This experiment tested two methods: the normalized difference vegetation index (NDVI) and supervised classification.

NDVI通過分析(xī)紅光和近(jìn)紅外(wài)波段(duàn)的(dí)反射(shè)特征，能夠有(yǒu)效(xiào)反映(yìng)茶葉的生理狀態變化。

NDVI effectively reflects changes in the physiological state of tea leaves by analyzing reflectance characteristics in the red and near-infrared bands.

監(jiān)督分(fēn)類則(zé)基於統(tǒng)計識別原理(lǐ)，通(tōng)過典型樣(yàng)本訓(xùn)練(liàn)建立(lì)分(fēn)類模型。實驗(yàn)結果(guǒ)顯(xiǎn)示，區分準確率均達到(dào)80%以(yǐ)上(shàng)，雖然葉(yè)片邊(biān)緣和(hé)莖(jīng)部(bù)區域還存在(zài)少量誤判，但整體效(xiào)果令人滿意。

Supervised classification is based on statistical recognition principles, establishing classification models through training with typical samples. Experimental results show that both methods achieved accuracy rates above 80%, with minor misjudgments remaining at leaf edges and stem regions. Overall, the performance was satisfactory.

「展望 / Outlook」

展(zhǎn)望未(wèi)來，將從三個方(fāng)麵持續(xù)優化技術(shù)方(fāng)案(àn)：

首先(xiān)，收集(jí)更多標(biāo)記(jì)清晰(xī)的樣品(pǐn)數(shù)據，擴充樣本(běn)庫規模，優化算(suàn)法參數(shù)設置；其次(cì)，改進光照(zhào)環境(jìng)設(shè)計(jì)，采用專(zhuān)用(yòng)綫(xiàn)光(guāng)源提(tí)升(shēng)信噪(zào)比，降低(dī)環境光幹(gān)擾(rǎo)；隨(suí)著定(dìng)性分(fēn)析(xī)達(dá)的(dí)準(zhǔn)確(què)率逐步提高(gāo)後(hòu)，可開展茶葉(yè)陳化程(chéng)度的定量反演研究(jiū)。

這套技術方案不僅適用(yòng)於茶葉(yè)新(xīn)老鑒別(bié)，還(huán)可拓展(zhǎn)應用(yòng)於(yú)茶葉分級，品質(zhì)檢測(cè)等多個領域，為(wéi)茶(chá)葉(yè)產(chǎn)業高質量發展(zhǎn)提供有(yǒu)力(lì)的技術(shù)支(zhī)撐。

Looking ahead, the technical solution will be optimized in three aspects:

1) Collect more clearly labeled sample data to expand the sample library and optimize algorithm parameters.

2) Improve lighting environment design by adopting dedicated line光(guāng)源 (light sources) to enhance signal-to-noise ratio and reduce ambient light interference.

3) As qualitative analysis accuracy improves, quantitative inversion research on tea aging degree can be conducted.

This technical solution is not only applicable to distinguishing new and aged tea but can also be extended to tea grading, quality testing, and other fields, providing robust support for the high-quality development of the tea industry.