Citrus Huanglongbing Terminator: How Hyperspectral Technology Achieves Precise Prevention and Control in Orchards

當超市貨架上飽滿的臍橙閃爍著誘人的橙紅色，很少有人知道這份甜蜜背后潛伏著一場持續(xù)百年的“柑橘世界大戰(zhàn)"。黃龍病，柑橘產(chǎn)業(yè)的頭號威脅，被稱為柑橘的“癌癥"，一旦感染只能砍樹。該病癥狀最早在18世紀被發(fā)現(xiàn)，我國1919年開始報告此病。自2021年起，廣西柑桔產(chǎn)區(qū)連續(xù)兩年遭受木虱疫害，果園普遍感染黃龍病，導致葉片黃化和果實畸變。

面對葉片黃化、果實畸變的染病果樹，傳統(tǒng)防控如同“盲人摸象"——人工巡查工作量大效率低，常錯過防治時機，而噴灑農(nóng)藥和砍伐病樹也未能解決問題。如今，通過高光譜遙感技術(shù)，在葉片尚未泛黃時識別黃龍病，提前把握防控窗口期。這串隱藏在光譜波段里的“柑橘密碼"，正在重新定義人類與病害的博弈規(guī)則。

When plump navel oranges flash their enticing orange-red color on supermarket shelves, few realize that this sweetness is overshadowed by a century-long "World War of Citrus." Huanglongbing (HLB), the number one threat to the citrus industry, is often referred to as the “cancer" of citrus; once infected, the trees must be cut down. The symptoms of this disease were first discovered in the 18th century, and it was first reported in China in 1919. Since 2021, the citrus-growing regions of Guangxi have suffered from severe infestations by the Asian citrus psyllid for two consecutive years, leading to widespread HLB infection that results in yellowing leaves and deformed fruit.

Faced with infected trees exhibiting yellowing leaves and deformities, traditional control methods resemble "blind men trying to touch an elephant"—manual inspections are labor-intensive and inefficient, often missing critical prevention opportunities, while pesticide spraying and tree removal have not effectively resolved the issue. Nowadays, using hyperspectral remote sensing technology, HLB can be identified before the leaves turn yellow, allowing for proactive control measures during the crucial prevention window. This hidden “citrus code" embedded in the spectral bands is redefining the rules of engagement between humans and disease.

當無人機搭載400~1000nm波段的高光譜相機掠過果園，每片葉子都會留下光譜“指紋"。由于患病葉片的光合作用受到抑制，并且含水量降低，其在可見光波段的葉綠素反射區(qū)和O-H伸縮振動區(qū)與健康葉片之間存在顯著差異。

在一些前期驗證實驗中，某科研團隊采集了健康葉片及不同病害程度的柑橘葉片，使用可見-近紅外光譜波段進行反射率測量，并重點關(guān)注450~800nm區(qū)間。經(jīng)過有效數(shù)據(jù)篩選，該研究通過最小二乘支持向量機（LS-SVM）和隨機森林（RF）算法建立了多種快速分類模型，發(fā)現(xiàn)模型的分類準確率分別在92.5%~95%、92.5%~97%。這樣的高效檢測手段大大提高了病害的早期識別率，為果農(nóng)提供了及時防治的依據(jù)。

When drones equipped with hyperspectral cameras ranging from 400 to 1000 nm fly over orchards, each leaf leaves behind a unique spectral "fingerprint." Infected leaves show significant differences in chlorophyll reflectance in the visible light range and O-H stretching vibration zones compared to healthy leaves due to suppressed photosynthesis and reduced moisture content.

In a series of preliminary validation experiments, a research team collected healthy leaves and leaves with varying levels of disease severity, using the visible-near-infrared spectral range to measure reflectance, focusing primarily on the 450-800 nm range. Following effective data filtering, the study established multiple rapid classification models through least squares support vector machine (LS-SVM) and random forest (RF) algorithms, achieving classification accuracies ranging from 92.5% to 95% and 92.5% to 97%. Such efficient detection methods significantly improved the early identification rate of the disease, providing timely preventive measures for farmers.

健康葉片和患病葉片（不同患病程度）的光譜曲線

Spectral curves of healthy leaves and infected leaves (with varying degrees of disease severity)

華南一個科研團隊使用了無人機載高光譜成像系統(tǒng)收集高光譜圖像，該高光譜成像儀波長范圍是450~950nm，通道數(shù)為125。經(jīng)過光譜預處理和特征工程，應(yīng)用了連續(xù)投影算法（SPA）來提取對柑橘患病植株分類影響最大的特征波長組合，從125個波段中鎖定了10個關(guān)鍵波段。這些波段如同破解黃龍病的“摩爾斯電碼"，高效地傳達了柑橘植株病蟲害的信息。

在模型構(gòu)建方面，研究人員基于全波段數(shù)據(jù)，應(yīng)用了BP神經(jīng)網(wǎng)絡(luò)和XgBoost算法進行分類評估，同時基于特征波段使用邏輯回歸和支持向量機（SVM）算法建立分類模型。

結(jié)果顯示，基于全波段的BP神經(jīng)網(wǎng)絡(luò)和XgBoost算法的分類模型分類準確率均超過95%。基于特征波段（698nm和762nm）的邏輯回歸和SVM建立的模型實現(xiàn)了93.00%和96.00%的患病樣品分類準確率。這證明了特征波長組合的有效性，為柑橘種植園的病蟲害監(jiān)測和精準防治提供了一定的數(shù)據(jù)和理論支撐。

A research team in South China utilized a drone-mounted hyperspectral imaging system to collect hyperspectral images, with the hyperspectral imager covering a wavelength range of 450-950 nm and comprising 125 channels. After spectral preprocessing and feature engineering, they applied a successive projection algorithm (SPA) to extract the most influential combinations of spectral wavelengths for classifying infected citrus plants, narrowing down to 10 key bands from 125. These bands resemble the "Morse code" for decoding HLB, efficiently conveying information about the disease and pests affecting citrus plants.

In model construction, researchers applied BP neural networks and XgBoost algorithms for classification assessment based on full-band data, while also using logistic regression and support vector machine (SVM) algorithms to establish classification models based on feature bands.

The results showed that classification models based on the full-band BP neural networks and XgBoost algorithms exceeded 95% classification accuracy. Models based on feature bands (698 nm and 762 nm), developed using logistic regression and SVM, achieved classification accuracies of 93.00% and 96.00% for infected samples. This confirms the effectiveness of the feature wavelength combinations, providing data and theoretical support for pest monitoring and precise prevention in citrus plantations.

華南科研團隊的試驗區(qū)域及樣本標注：粉、紅、藍、黃、白圓圈標記分別代表不同患病程度和患病未定級的黃龍病植株，三角形標記為缺素植株；沒有標記的植株為健康植株

Experimental area and sample labeling by the research team in South China: The pink, red, blue, yellow, and white circular markers represent citrus plants with different degrees of Huanglongbing severity and those whose condition has not yet been classified; the triangular markers indicate nutrient-deficient plants; unmarked plants are healthy.

中美兩所高校展開合作，采用無人機搭載高光譜和多光譜成像系統(tǒng)，獲得光譜遙感數(shù)據(jù)，快速識別感染黃龍病的柑橘植株。該團隊將航拍獲取的光譜數(shù)據(jù)與農(nóng)田和實驗室的地面驗證結(jié)果相結(jié)合，顯示出航拍的光譜數(shù)據(jù)能夠有效地區(qū)分健康植株與感染黃龍病的植株，準確率最高可達90%。

Two universities from China and the United States collaborated using drones equipped with hyperspectral and multispectral imaging systems to obtain spectral remote sensing data for rapid identification of citrus plants infected with HLB. This team combined aerial spectral data with ground-truth validation results from fields and laboratories, demonstrating that aerial spectral data effectively distinguish between healthy plants and those infected with HLB, achieving accuracies of up to 90%.

左圖：美國佛羅里達的黃龍病感染區(qū)，右圖：研究中使用的高光譜與多光譜傳感器

Left figure: Huanglongbing infection area in Florida, USA; Right figure: Hyperspectral and multispectral sensors used in the study.

從古詩詞中描繪的“柑橘正熟，金果盈枝"的豐收畫面，到今天光譜儀中的圖譜曲線，人類與作物的對話從未停止。那些舞動的光譜曲線，不僅是科技對抗病害的利劍，更承載著我們對土地最深的敬畏。或許在不遠的未來，每顆柑橘都將擁有自己的“光譜證件"——而這，正是科技賦予農(nóng)業(yè)的詩意浪漫。

From ancient poetry depicting the bountiful harvest of “ripe citrus, golden fruits hanging from branches" to the spectral curves represented in spectrometers today, the dialogue between humans and crops has never ceased. Those dancing spectral curves are not only the sword of technology against diseases but also embody our deep respect for the land. Perhaps soon, each citrus fruit will have its own "spectral ID"—a poetic romance of agriculture bestowed by technology.

案例來源 / Source

1.Bai Ziqin, Zhou Changyong,  The Research Progress of Citrus Huanglongbing on Pathogen Diversity and Epidemiology, Chinese Agricultural Science Bulletin, 2012,28(1):133-137.

2.QIU Hong-lin, LIU Tian-yuan, KONG Li-li, YU Xin-na, WANG Xian-da, HUANG Mei-zhen. Rapid Detection of Citrus Huanglongbing Based on Extraction of Characteristic Wavelength of Visible Spectrum and Classification Algorithm[J]. Spectroscopy and Spectral Analysis, 2024,44(6): 1518-1525.

3.Li X, Lee WS, Li M, et al. Spectral difference analysis and airborne imaging classification for citrus greening infected trees Computers and Electronics in Agriculture.. 2012 Apr;83:32-46.

4.DENG Xiaoling, ZENG Guoliang, ZHU Zihao, et al. Classification and feature band extraction of diseased citrus plants based on UAV hyperspectral remote sensing[J]. Journal of South China Agricultural University, 2020, 41(6): 100-108.

5.Xiuhua Li, Won Suk Lee, Minzan Li, Reza Ehsani, Ashish Ratn Mishra, Chenghai Yang, Robert L. Mangan, Spectral difference analysis and airborne imaging classification for citrus greening infected trees, Computers and Electronics in Agriculture, Volume 83,2012, Pages 32-46.