When "Mild Spicy" in Sichuan Equals "Extremely Spicy" in Guangdong... Hyperspectral Camera Measures Chili Heat Levels

每(měi)個人(rén)對辣度的接受程(chéng)度(dù)都(dū)不一(yī)樣(yàng)，火(huǒ)鍋底料(liào)的辣度如何(hé)科(kē)學量化？本次(cì)實(shí)驗利用高(gāo)光(guāng)譜(pǔ)相(xiāng)機(jī)，對6種(zhǒng)不同(tóng)辣(là)度的火(huǒ)鍋(guō)底(dǐ)料(liào)進行(háng)測試，探(tàn)索(suǒ)光(guāng)譜(pǔ)數據與辣(là)度(dù)的關聯性(xìng)。

People's tolerance for spiciness varies widely, but how can the heat level of hot pot base be scientifically quantified? This experiment utilized a hyperspectral camera to test six hot pot bases with different spiciness levels, exploring the correlation between spectral data and chili heat intensity.

「樣品介(jiè)紹(shào) / Samples」

測試6種不(bù)同辣度(dù)的火鍋底(dǐ)料(liào)，辣(là)度(dù)分別為
：12°，36°，45°，52°，65°，75°

Six hot pot base samples with varying heat levels were tested: 12°, 36°, 45°, 52°, 65°, and 75°.

「數(shù)據采集(jí) / Data Acquisition」

高(gāo)光譜相機(jī)
：覆蓋(gài)400~1700nm波段(duàn)（可(kě)見(jiàn)光(guāng)+短波紅外）

成像方式
：綫性推掃(sǎo)，確(què)保數據精準(zhǔn)

光源與(yǔ)環(huán)境
：鹵素燈均匀照明，暗室環境減少(shǎo)幹擾

樣(yàng)品(pǐn)擺(bǎi)放
：水平(píng)位移台(tái)固(gù)定(dìng)，保證(zhèng)成像(xiàng)穩定

Hyperspectral Camera: Covered 400–1700 nm (visible light + short-wave infrared).

Imaging Method: Linear push-broom scanning for precise data capture.

Lighting & Environment: Halogen lamp for uniform illumination, darkroom to minimize interference.

Sample Setup: Fixed on a horizontal displacement platform for stable imaging.

400-1000nm

900-1700nm

「分析方(fāng)法 / Analysis Method」

高光譜成像(xiàng)不僅(jǐn)能(néng)拍(pāi)出照片(piàn)，還能記(jì)錄(lù)每個像素點的(dí)光(guāng)譜“指(zhǐ)紋"。 

實驗過(guò)程中(zhōng)，首先使用400-1000nm可(kě)見近紅外和900-1700nm短波紅(hóng)外(wài)兩(liǎng)台(tái)高光(guāng)譜相機采(cǎi)集6種火鍋底(dǐ)料(liào)樣(yàng)品的光譜(pǔ)數(shù)據。

在(zài)數據預處(chǔ)理階段(duàn)，通過(guò)專業(yè)的(dí)高(gāo)光譜分析(xī)軟(ruǎn)件對(duì)原(yuán)始(shǐ)數據進行(háng)降噪處理和反(fǎn)射率計(jì)算，同(tóng)時(shí)消除背景光(guāng)譜幹(gān)擾(rǎo)，確保(bǎo)獲得(dé)純淨的目標物體光(guāng)譜信息，這一過(guò)程通常(cháng)在數據采集時同(tóng)步完成(chéng)。

隨後從處理(lǐ)後(hòu)的高(gāo)光(guāng)譜(pǔ)數據中(zhōng)提(tí)取關(guān)鍵特征，包括(kuò)光譜反射率，吸收(shōu)峰位置及光(guāng)譜形(xíng)態特征(zhēng)等，並運(yùn)用(yòng)主(zhǔ)成分(fēn)分(fēn)析等降(jiàng)維方法篩(shāi)選(xuǎn)出(chū)具有代表(biǎo)性的(dí)特征參(cān)數。

在分(fēn)類(lèi)識(shí)別環節(jié)，利用不同(tóng)物質對特(tè)定波(bō)段反(fǎn)射率的差異(yì)特性(xìng)，分別(bié)采用監(jiān)督學習和(hé)無(wú)監(jiān)督(dū)學習(xí)兩(liǎng)種方法
：前(qián)者通過(guò)標記(jì)數據(jù)集(jí)訓(xùn)練光(guāng)譜(pǔ)角(jiǎo)製(zhì)圖(tú)或(huò)卷(juàn)積神(shén)經(jīng)網(wǎng)絡等分類模型(xíng)，後者則運用K均(jūn)值(zhí)或(huò)層次聚類(lèi)等(děng)算法(fǎ)實現數(shù)據自(zì)動(dòng)分(fēn)類(lèi)。

最終將(jiāng)分析結(jié)果以偽彩色(sè)圖像形(xíng)式(shì)直(zhí)觀(guān)呈現，展(zhǎn)示不(bù)同物質的(dí)空間(jiān)分(fēn)布情(qíng)況(kuàng)，並基於光譜(pǔ)特征開展定量和定(dìng)性分(fēn)析，計(jì)算得出各類物(wù)質(zhì)的濃度(dù)或(huò)類(lèi)別參(cān)數。

Hyperspectral imaging not only captures photos but also records the spectral "fingerprint" of each pixel.

During the experiment, two hyperspectral cameras (400–1000 nm visible-NIR and 900–1700 nm SWIR) were used to collect spectral data from the six samples.

In the preprocessing stage, raw data underwent noise reduction and reflectance calibration via specialized software, while background interference was eliminated to ensure clean spectral data. This process was synchronized with data acquisition.

Key features were then extracted from the processed data, including spectral reflectance, absorption peak positions, and spectral shape characteristics. Dimensionality reduction methods like PCA were applied to identify the most representative parameters.

For classification, both supervised and unsupervised learning were employed:

Supervised methods (e.g., spectral angle mapper or CNN) used labeled datasets to train models.

Unsupervised methods (e.g., K-means or hierarchical clustering) automated data grouping based on reflectance differences in specific bands.

Results were visualized as pseudo-color images to display spatial distributions of materials, followed by quantitative/qualitative analysis to calculate concentrations or categories.

「光譜(pǔ)曲(qū)綫 / Spectral Curves」

在400-1700nm波(bō)長(cháng)範(fàn)圍內，六(liù)種(zhǒng)不同辣(là)度(dù)的(dí)火(huǒ)鍋底(dǐ)料樣本(běn)在(zài)a麵(miàn)和(hé)b麵的反(fǎn)射率(shuài)曲綫(xiàn)呈(chéng)現出(chū)相似的(dí)光(guāng)譜波(bō)形(xíng)，但反(fǎn)射(shè)率數(shù)值(zhí)隨辣(là)度變(biàn)化而(ér)存在顯著差異。具(jù)體表現為(wéi)辣(là)度越高，反射率(shuài)越低，這一趨(qū)勢在a麵(miàn)和b麵(miàn)均(jūn)保持一致(zhì)。

值得注意的是，在b麵(miàn)的860-930nm波段(duàn)範圍內，反射率(shuài)曲(qū)綫對辣(là)度(dù)的區分(fēn)效果(guǒ)尤(yóu)為(wéi)明(míng)顯，能夠更清晰地反映(yìng)辣(là)度差(chà)異。

Within 400–1700 nm, reflectance curves of the six samples (A-side and B-side) showed similar waveforms but significant reflectance variations correlated with spiciness. Higher heat levels consistently exhibited lower reflectance on both sides.

Notably, the 860–930 nm range on the B-side provided the clearest distinction between heat levels.

a麵反射率(shuài)(400-1000nm)

a麵(miàn)反(fǎn)射率(900-1700nm)

b麵反(fǎn)射(shè)率(shuài)(400-1000nm)

「建立CNN模型 / CNN Modeling」

為了進一步分析辣度(dù)分(fēn)類(lèi)的(dí)可行性，研(yán)究采(cǎi)用卷積(jī)神(shén)經(jīng)網(wǎng)絡(luò)（CNN）對(duì)高光(guāng)譜(pǔ)數據(jù)進(jìn)行(háng)建模。

To further assess classification feasibility, a CNN model was applied to hyperspectral data.

建立CNN模(mó)型（400-1000nm  a/b麵） / CNN Model (400–1000 nm, A/B-sides)

在(zài)400-1000nm波(bō)段的(dí)a麵(miàn)數(shù)據(jù)分類中，模型整體準確(què)率介於75%-85%之(zhī)間，其(qí)中辣度(dù)45°和(hé)75°的(dí)分(fēn)類(lèi)效果(guǒ)較好，而(ér)辣(là)度(dù)12°和52°由於數(shù)據采(cǎi)集時受容(róng)器遮擋(dǎng)影(yǐng)響，部分(fēn)區域出現誤(wù)判(pàn)。此外(wài)，辣度(dù)36°因(yīn)樣(yàng)品表麵凹(āo)陷導致數(shù)據質(zhì)量(liáng)下降(jiàng)，而(ér)辣度65°的部(bù)分區域(yù)被(bèi)錯誤(wù)歸(guī)類(lèi)為75°。

相(xiāng)比之下(xià)，b麵的(dí)分類表現更(gēng)為穩定(dìng)，整體(tǐ)準確率約(yuē)為85%，僅辣(là)度(dù)45°的少量區域被誤判為(wéi)36°。

A-side: Overall accuracy ranged 75%–85%. Samples at 45° and 75° were classified best, while 12° and 52° suffered partial misclassification due to container obstruction during imaging. The 36° sample had uneven surfaces, and 65° was occasionally mislabeled as 75°.

B-side: Performance was more stable (~85% accuracy), with only minor misclassification (45° vs. 36°).

a麵(miàn)結果(guǒ) (400-1000nm)

a麵(miàn)結(jié)果(guǒ) (400-1000nm)

建立(lì)CNN模型（900-1700nm  a/b麵(miàn)） / CNN Model (900–1700 nm, A/B-sides)

在900-1700nm波段(duàn)的分析中(zhōng)，a麵(miàn)數(shù)據(jù)的分(fēn)類準確率(shuài)在(zài)70%-80%之(zhī)間(jiān)，其(qí)中辣度12°和36°因表麵凹凸(tū)不平或(huò)凹陷導(dǎo)致數(shù)據質(zhì)量較差(chà)，誤判率(shuài)較高(gāo)，而(ér)辣(là)度45°，52°，65°和75°的(dí)分類效果較好(hǎo)。

相比(bǐ)之(zhī)下，b麵數(shù)據(jù)由(yóu)於表麵更平滑，且無(wú)幹(gān)辣椒(jiāo)等固體(tǐ)遮(zhē)擋(dǎng)，分(fēn)類表(biǎo)現顯(xiǎn)著優於a麵，整體準確率超(chāo)過(guò)90%，僅有(yǒu)少量區域出現(xiàn)誤(wù)判。

這一結果表(biǎo)明(míng)，900-1700nm波段(duàn)可(kě)能更適(shì)合用(yòng)於(yú)火(huǒ)鍋底(dǐ)料辣度的精準檢測(cè)，尤其是結合b麵(miàn)數據時(shí)，分(fēn)類效果(guǒ)更佳(jiā)。

A-side: Accuracy was 70%–80%. Samples at 12° and 36° showed higher misclassification due to surface irregularities, while 45°–75° performed better.

B-side: Superior accuracy (>90%) was achieved thanks to smoother surfaces and absence of solid obstructions (e.g., dried chilies).

These results suggest that 900–1700 nm SWIR, especially with B-side data, is more suitable for precise heat-level detection.

a麵結(jié)果 (900-1700nm)

b麵(miàn)結(jié)果 (900-1700nm)

「總(zǒng)結(jié) / Conclusion」

基(jī)於高(gāo)光(guāng)譜視(shì)覺技術(shù)的研究(jiū)表明(míng)，通(tōng)過(guò)對六(liù)種不(bù)同辣度(dù)的火鍋底(dǐ)料樣(yàng)品(pǐn)進(jìn)行(háng)高光譜(pǔ)數據采(cǎi)集，並經(jīng)過數據預處(chǔ)理和(hé)算法分析，能夠(gòu)有效區(qū)分樣品的辣度等級。

實驗數據顯示，雖(suī)然(rán)樣(yàng)品a麵(miàn)和b麵的光(guāng)譜(pǔ)曲綫(xiàn)均能(néng)反映(yìng)辣(là)度變化，但b麵的區分(fēn)效果更(gēng)為顯著。在光譜(pǔ)波段(duàn)選擇方麵(miàn)，相比(bǐ)400-1000nm的可見(jiàn)近紅(hóng)外譜段，900-1700nm的短波紅外譜段展現出(chū)更(gēng)高的識(shí)別準確率(shuài)和(hé)檢測精度(dù)。

為進一步(bù)提升(shēng)研究結果的(dí)可靠(kào)性，後(hòu)續工(gōng)作將重(zhòng)點(diǎn)擴大樣(yàng)本數據(jù)量，通過(guò)增加樣本多樣(yàng)性(xìng)來(lái)持續(xù)優化識別準確(què)率。

Hyperspectral imaging effectively differentiated the six heat levels of hot pot base samples after data preprocessing and algorithmic analysis.

While both A-side and B-side spectral curves reflected spiciness trends, the B-side provided clearer distinctions. Compared to 400–1000 nm visible-NIR, the 900–1700 nm SWIR band demonstrated higher accuracy and precision.

To enhance reliability, future work will expand sample diversity and dataset size for further optimization.