每年4月22日的世界地球日，都提醒著我們關(guān)注環(huán)境問題，而塑料污染無(wú)疑是其中一個(gè)棘手的挑戰(zhàn)。大量的塑料垃圾難以有效回收，傳統(tǒng)的分選流程難以實(shí)現(xiàn)塑料的高純度分類，這不僅是資源的巨大浪費(fèi)，也限制了再生塑料的應(yīng)用價(jià)值，最終只能堆積或焚燒，對(duì)地球造成沉重負(fù)擔(dān)。如何才能讓這些廢棄的塑料“變廢為寶”，以更高的純度回歸生產(chǎn)循環(huán)，成為我們共同探索的方向。

傳統(tǒng)的塑料分選方法，無(wú)論是人工操作還是視覺相機(jī)，都難以應(yīng)對(duì)日益復(fù)雜的塑料混合物。想象一下，面對(duì)五花八門的塑料制品，想要一一辨別它們的“真身”并精確分類，無(wú)疑是一項(xiàng)艱巨的任務(wù)。然而，科技的進(jìn)步正在打破這一瓶頸。

一種被譽(yù)為“材料成分鑒定官”的技術(shù)——高光譜成像，正在已更高的精度改變著塑料回收的格局。這項(xiàng)技術(shù)通過(guò)捕捉物體在連續(xù)光譜范圍內(nèi)的反射或輻射信息，構(gòu)建一個(gè)包含豐富化學(xué)成分信息的“光譜指紋庫(kù)”。它能夠識(shí)別出不同塑料分子的吸收特征，實(shí)現(xiàn)對(duì)微小光譜差異的高度敏感識(shí)別，對(duì)可降解塑料也能精準(zhǔn)區(qū)分，并且這一切都可以在產(chǎn)線上高速、無(wú)損地完成。

Every year, Earth Day on April 22nd serves as a crucial reminder to address environmental challenges, and plastic pollution stands out as a particularly thorny one. The sheer volume of plastic waste is difficult to recycle effectively. Traditional sorting methods struggle to achieve high purity levels, leading to significant resource waste and limiting the value of recycled plastic. Ultimately, much of this ends up in landfills or incinerators, placing a heavy burden on our planet. Finding ways to turn this discarded plastic into valuable resources, returning it to the production cycle with higher purity, is a shared goal we are actively pursuing.

Traditional plastic sorting methods, whether manual or standard visual cameras, are ill-equipped to handle increasingly complex plastic mixtures. Imagine trying to identify the "true identity" of a bewildering array of plastic products and sort them accurately – it's truly a tall order. However, technological advancements are breaking through this bottleneck.

Hyperspectral imaging, often dubbed a "material composition detective," is revolutionizing plastic recycling with unprecedented accuracy. This technology captures the reflectance or emission information of objects across a continuous spectral range, creating a "spectral fingerprint database" rich in chemical composition data. It can identify the absorption characteristics of different plastic molecules, enabling highly sensitive recognition of even subtle spectral differences. Biodegradable plastics can also be precisely distinguished. Crucially, all of this can be done at high speed and without damaging the material on a production line.

高光譜可以獲得連續(xù)的光譜曲線

Hyperspectral imaging provides continuous spectral curves

(Upper left: grayscale image, Upper right: RGB image, Lower left: multispectral image, Lower right: hyperspectral image)

這項(xiàng)技術(shù)已經(jīng)在多個(gè)應(yīng)用場(chǎng)景中大顯身手。例如，英國(guó)倫敦大學(xué)的研究人員采用近紅外高光譜技術(shù)對(duì)不同尺寸、不同材料的塑料樣本進(jìn)行檢測(cè)，包括可堆肥材料（甘蔗衍生和棕櫚葉衍生）、可堆肥塑料（PLA、PBAT）和傳統(tǒng)塑料（PP、PET和LDPE），重點(diǎn)關(guān)注950~1730nm波段，使用了主成分分析（PCA）和偏最小二乘判別分析（PLS-DA）。實(shí)驗(yàn)結(jié)果顯示，對(duì)于尺寸大于10毫米×10毫米的樣品，分類準(zhǔn)確率達(dá)到100%，而對(duì)于較小碎片，準(zhǔn)確率略有下降。該結(jié)果充分證明了高光譜技術(shù)在實(shí)際塑料分選中的高效性。

This technology has already proven its mettle in various applications. For instance, researchers at University College London in the UK utilized Near-Infrared Hyperspectral Imaging (NIR-HSI) to analyze plastic samples of different sizes and materials, including compostable materials (sugarcane- and palm leaf-derived), compostable plastics (PLA, PBAT), and conventional plastics (PP, PET, and LDPE). Focusing on the 950~1730nm range and employing Principal Component Analysis (PCA) and Partial Least Squares Discriminant Analysis (PLS-DA), their experiments showed a 100% classification accuracy for samples larger than 10mm x 10mm, with a slight decrease for smaller fragments. This clearly demonstrates the high efficiency of hyperspectral technology in practical plastic sorting.

通過(guò)高光譜相機(jī)獲取的不同塑料的原始吸收光譜

Raw absorbance spectra of sugarcane derived packaging, PP, PLA, PET, LDPE, PBAT and palm leaf derived packaging acquired by hyperspectral camera

意大利的一個(gè)研究團(tuán)隊(duì)通過(guò)檢測(cè)PET、PS和PLA的主要吸收峰（分別出現(xiàn)在1150nm至1660nm區(qū)間），成功區(qū)分了不同塑料類型。論文指出，這種方法能夠定量評(píng)估分選過(guò)程的準(zhǔn)確性，為工業(yè)應(yīng)用提供了可靠依據(jù)。

An Italian research team successfully differentiated various plastic types, including PET, PS, and PLA, by detecting their main absorption peaks (occurring between 1150nm and 1660nm). Their paper highlights that this method allows for quantitative evaluation of sorting accuracy, providing a reliable basis for industrial applications.

PET樣品光譜特征 / Spectral signatures of PET samples in the NIR region

PLA樣品光譜特征 / Spectral signatures of PLA samples in the NIR region

PS樣品光譜特征 / Spectral signatures of PS samples in the NIR region

國(guó)內(nèi)的研究人員同樣進(jìn)行了分析，結(jié)合RGB和高光譜成像數(shù)據(jù)，開發(fā)了一種多尺度特征融合算法，實(shí)現(xiàn)了對(duì)透明PET、藍(lán)色PET和透明PP瓶的高效辨識(shí)，整體分類準(zhǔn)確率達(dá)到95.55%，而藍(lán)色PET的準(zhǔn)確率高達(dá)97.5%。這表明采用多傳感器融合方法能夠進(jìn)一步提高分選系統(tǒng)的穩(wěn)定性和準(zhǔn)確率。

Chinses researchers have also contributed significantly. By analyzing combined RGB and hyperspectral imaging data, they developed a multi-scale feature fusion algorithm to achieve efficient identification of transparent PET, blue PET, and transparent PP bottles. This resulted in an overall classification accuracy of 95.55%, with blue PET reaching an impressive 97.5%. This work indicates that integrating multiple sensors can further enhance the stability and accuracy of sorting systems.

塑料瓶的平均光譜曲線 / Mean spectral curve of waste plastic bottles

雜亂瓶子的分類圖。（a）RGB 圖像。（b）基本實(shí)況。（c-h）分別預(yù)測(cè)了不同特征融合方法的分類圖。

Classification maps for cluttered bottles. (a) RGB image. (b) Ground truth. (c-h) show the classification maps predicted by different feature fusion methods, respectively.

值得一提的是，我們也為塑料回收提供了成熟的高光譜解決方案。高光譜塑料識(shí)別系統(tǒng)集成到各類塑料分選機(jī)中，無(wú)論是針對(duì)整瓶還是碎片化塑料，都能通過(guò)數(shù)據(jù)接口將精準(zhǔn)的識(shí)別結(jié)果反饋給控制系統(tǒng)，進(jìn)而通過(guò)氣閥實(shí)現(xiàn)自動(dòng)化的高效分選。目前，工業(yè)高光譜相機(jī)已經(jīng)推出，憑借其高幀頻的特點(diǎn)，能夠滿足產(chǎn)線上快速、連續(xù)分選的要求。

更進(jìn)一步，工程師利用900~1700nm近紅外高光譜相機(jī)對(duì)土壤中的微塑料顆粒進(jìn)行了識(shí)別研究，這為解決更復(fù)雜、更貼近實(shí)際環(huán)境的塑料回收提供了重要的實(shí)驗(yàn)基礎(chǔ)和技術(shù)支持。

We also offer mature hyperspectral solutions for plastic recycling. Our hyperspectral plastic identification system can be integrated into various plastic sorting machines, whether for whole bottles or plastic flakes. Through data interfaces, precise identification results are relayed to the control system, enabling automated, high-efficiency sorting via air jets. Industrial hyperspectral cameras are now available, and their high frame rates are well-suited for fast, continuous sorting on production lines.

Furthermore, engineers have conducted research on identifying microplastic particles in soil using 900–1700nm near-infrared hyperspectral cameras. This provides a vital experimental foundation and technical support for tackling more complex, real-world plastic recycling challenges.

塑料分選實(shí)驗(yàn)裝置及分選結(jié)果

Experimental setup and sorting results for plastic separation

綜上所述，高光譜成像技術(shù)憑借其精準(zhǔn)的材料識(shí)別能力和高效的在線檢測(cè)特性，正為塑料回收行業(yè)注入新的活力。這項(xiàng)“材料成分鑒定官”不僅能夠顯著提升回收效率和材料純度，更能為循環(huán)經(jīng)濟(jì)注入強(qiáng)勁動(dòng)力，助力我們實(shí)現(xiàn)更綠色、更可持續(xù)的未來(lái)。讓科技賦能回收，共同開啟塑料“點(diǎn)石成金”的新篇章！

In conclusion, hyperspectral imaging technology, with its precise material identification capabilities and efficient online inspection features, is breathing new life into the plastic recycling industry. This "material composition detective" not only significantly boosts recycling efficiency and material purity but also injects strong momentum into the circular economy. It's a game-changer that helps us move towards a greener, more sustainable future. By empowering recycling with technology, we can truly turn trash into treasure and open a new chapter in plastic recycling.

案例來(lái)源 / Source：

1. Taneepanichskul N, Hailes HC and Miodownik M (2023) Automatic identification and classification of compostable and biodegradable plastics using hyperspectral imaging. Front. Sustain. 4:1125954.

2. Moroni, M.; Mei, A. Characterization and Separation of Traditional and Bio-Plastics by Hyperspectral Devices. Appl. Sci. 2020, 10, 2800.

3. Cai, Z.; Yang, J.; Fang, H.; Ji, T.; Hu, Y.; Wang, X. Research on Waste Plastics Classification Method Based on Multi-Scale Feature Fusion. Sensors 2022, 22, 7974.