Application and Analysis of Hyperspectral Technology in Cabbage Freshness Discrimination

食品(pǐn)新鮮(xiān)度檢測是食(shí)品(pǐn)安全(quán)與(yǔ)品質控(kòng)製(zhì)中的(dí)重要環節，直(zhí)接關係(xì)到消(xiāo)費(fèi)者的健(jiàn)康(kāng)體(tǐ)驗與營(yíng)養攝(shè)入。對大眾(zhòng)而言(yán)，食品新鮮(xiān)度不僅影響食(shí)材的口(kǒu)感和營(yíng)養(yǎng)價值(zhí)，更是保障(zhàng)飲食(shí)安全，減(jiǎn)少食源性疾病風險的(dí)關鍵因素。

Food freshness detection is a critical aspect of food safety and quality control, directly impacting consumers' health experiences and nutritional intake. For the public, food freshness not only influences the taste and nutritional value of ingredients but also plays a key role in ensuring dietary safety and reducing the risk of foodborne illnesses.

本(běn)次測試(shì)以不(bù)同新鮮(xiān)程(chéng)度的白(bái)菜為研究對象，利用高(gāo)光譜技(jì)術(shù)實現對(duì)白菜新(xīn)鮮(xiān)度的有(yǒu)效(xiào)區分(fēn)。

This study used cabbages of different freshness levels as research subjects and employed hyperspectral technology to achieve effective discrimination of cabbage freshness.

本次(cì)測試(shì)所(suǒ)采用的高(gāo)光(guāng)譜相(xiāng)機覆蓋(gài)400-1000nm的光譜(pǔ)範(fàn)圍，具備優於(yú)2.8nm的(dí)光(guāng)譜分(fēn)辨率，高達300個光(guāng)譜波段，F/2的(dí)大光圈設(shè)計提(tí)升光(guāng)通(tōng)量(liáng)，480個空間像素確保空間細節表現，采(cǎi)用CMOS探(tàn)測器(qì)並(bìng)結合(hé)USB接口實(shí)現便捷高效的數(shù)據傳輸，12bits的有效(xiào)位(wèi)深保(bǎo)障(zhàng)了圖(tú)像(xiàng)數據(jù)的(dí)豐富層(céng)次(cì)與(yǔ)精度。該設備為農(nóng)，林，食品(pǐn)檢測等應用提(tí)供了有力(lì)工(gōng)具，歡迎(yíng)大(dà)家進(jìn)一(yī)步了解(jiě)。

測試采(cǎi)用(yòng)綫性推掃(sǎo)成像(xiàng)方(fāng)案(àn)，照明(míng)光源(yuán)為鹵素(sù)光源(yuán)。實驗(yàn)在暗(àn)室(shì)環境(jìng)中進行(háng)，樣品被放置(zhì)於(yú)水平位(wèi)移台上以(yǐ)完成圖(tú)像(xiàng)采(cǎi)集。

The hyperspectral camera used in this test covers a spectral range of 400–1000 nm, with a spectral resolution better than 2.8 nm, up to 300 spectral bands, and an F/2 large aperture design that enhances light throughput. With 480 spatial pixels ensuring detailed spatial representation, it utilizes a CMOS detector combined with a USB interface for efficient data transmission. The 12-bit effective bit depth ensures rich image data hierarchy and precision. This device serves as a powerful tool for applications in agriculture, forestry, and food detection. Further inquiries are welcome.

The test adopted a linear push-broom imaging method, with a halogen light source for illumination. The experiment was conducted in a darkroom environment, and samples were placed on a horizontal translation stage for image acquisition.

測試樣(yàng)品(pǐn)及(jí)測試(shì)環境(jìng) / Test Samples and Testing Environment

通(tōng)過獲取不(bù)同新(xīn)鮮度(dù)白(bái)菜(cài)在400-1000nm範(fàn)圍內的光(guāng)譜曲綫，並分(fēn)別選取完(wán)好的莖(jīng)與葉區域以及(jí)幹(gān)枯的莖(jīng)與(yǔ)葉(yè)區(qū)域計(jì)算平(píng)均光(guāng)譜，分析(xī)表明(míng)：

完(wán)好的(dí)葉(yè)片（紅色(sè)曲綫(xiàn)）與幹枯葉(yè)片(piàn)（紫色(sè)曲綫）在(zài)500-700nm和(hé)800-900nm波段的(dí)光譜響(xiǎng)應存(cún)在明顯(xiǎn)差異；

完(wán)好(hǎo)的(dí)莖（綠(lǜ)色曲(qū)綫(xiàn)）與幹枯的(dí)莖(jīng)（黃色(sè)曲綫(xiàn)）則(zé)在(zài)650-850nm範圍內表現出(chū)顯著(zhù)光譜變(biàn)化(huà)。

By obtaining spectral curves of cabbages with different freshness levels within the 400–1000 nm range and calculating average spectra from intact stem and leaf regions as well as dried stem and leaf regions, the analysis revealed:

Significant differences in spectral responses between intact leaves (red curve) and dried leaves (purple curve) in the 500–700 nm and 800–900 nm bands;

intact stems (green curve) and dried stems (yellow curve) exhibited notable spectral variations within the 650–850 nm range.

反射率(shuài)測(cè)試(shì) / Reflectance Testing

在(zài)數(shù)據處理(lǐ)階段(duàn)，我(wǒ)們利用了兩(liǎng)種(zhǒng)不(bù)同(tóng)的算法(fǎ)：

During data processing, two different algorithms were applied:

算(suàn)法(fǎ)一選(xuǎn)取枯(kū)葉ROI區域(yù)作為(wéi)分類標(biāo)準，能夠(gòu)有效(xiào)識別(bié)部(bù)分白(bái)菜(cài)表(biǎo)麵(miàn)的幹枯區域，但(dàn)對莖部(bù)幹(gān)枯區(qū)域(yù)的(dí)區分(fēn)效果有限。

Algorithm 1 used dried leaf ROI regions as classification criteria, effectively identifying some dried areas on the cabbage surface but demonstrating limited ability to distinguish dried regions on stems.

算法一(yī) / Algorithm 1

算(suàn)法二通過對圖像進(jìn)行(háng)特征(zhēng)提(tí)取，實(shí)現了(liǎo)對(duì)表麵幹枯(kū)區域(yù)的更有效識別，從而(ér)對(duì)不(bù)同新鮮(xiān)度(dù)的(dí)白(bái)菜實現了良好區(qū)分。

Algorithm 2 employed feature extraction from images, achieving more effective identification of surface-dried areas and enabling better discrimination of cabbages with different freshness levels.

算(suàn)法(fǎ)二 / Algorithm 2

實(shí)驗(yàn)結果(guǒ)表(biǎo)明(míng)，基(jī)於(yú)400-1000nm波段(duàn)的高光(guāng)譜相機能夠檢測出(chū)不同新(xīn)鮮程(chéng)度白菜的(dí)光(guāng)譜差異(yì)，且數(shù)據(jù)處理結果(guǒ)與實際狀(zhuàng)態(tài)相(xiāng)符。

Experimental results indicate that the hyperspectral camera based on the 400–1000 nm band can detect spectral differences in cabbages of varying freshness levels, and the data processing outcomes align with actual conditions.

本(běn)實(shí)驗亦(yì)識別(bié)出若幹(gān)實際測(cè)量中的(dí)難(nán)點：白(bái)菜(cài)表(biǎo)麵(miàn)覆蓋(gài)的(dí)保鮮(xiān)膜易引(yǐn)起(qǐ)光綫(xiàn)反射，對信號造(zào)成(chéng)幹擾(rǎo)；同時，白(bái)菜的弧形表麵(miàn)不僅(jǐn)影(yǐng)響光(guāng)綫(xiàn)反(fǎn)射特(tè)性，也對相(xiāng)機(jī)的(dí)對焦(jiāo)精(jīng)度(dù)提出了(liǎo)挑戰。

針(zhēn)對(duì)這(zhè)些問(wèn)題，下一步計(jì)劃(huá)包括(kuò)優化光源結(jié)構(gòu)，引入多角度照明方案(àn)，建立反射(shè)率(shuài)校(xiào)正模型(xíng)以(yǐ)消(xiāo)除(chú)弧麵(miàn)造成的光(guāng)譜強度(dù)偏(piān)差，提(tí)升(shēng)數據可比(bǐ)性(xìng)。此外，還(huán)將擴(kuò)大(dà)樣本數量(liáng)，構建基於(yú)深(shēn)度學習的(dí)新鮮(xiān)度判別(bié)模型，以期(qī)實現(xiàn)更(gēng)精確(què)，可靠(kào)的(dí)白(bái)菜新鮮(xiān)度分(fēn)類(lèi)能(néng)力(lì)，為實(shí)現(xiàn)更(gēng)安全(quán)，更可靠(kào)的生鮮(xiān)食(shí)品(pǐn)供應鏈提(tí)供(gōng)了有(yǒu)效的(dí)技術保障。

This experiment also identified several challenges in practical measurements: The freshness-preserving film covering the cabbage surface easily causes light reflection, interfering with signals; meanwhile, the curved surface of the cabbage not only affects light reflection characteristics but also poses challenges to the camera’s focusing accuracy.

To address these issues, future plans include optimizing the light source structure, introducing multi-angle lighting schemes, and establishing a reflectance correction model to eliminate spectral intensity deviations caused by curved surfaces, thereby improving data comparability. Additionally, the sample size will be expanded to develop a deep learning-based freshness discrimination model, aiming to achieve more accurate and reliable cabbage freshness classification capabilities. This provides effective technical support for building a safer and more reliable fresh food supply chain.