A precise recognition and detection method for citrus huanglongbing near-infrared spectral signals by pls
Patent Information
- Application Number
- CN202610848229.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-12
- Publication Date
- 2026-10-30
AI Technical Summary
[0044]检测准确率高：采用改进的组合预处理方法和组合式特征波长选择方法，有效消除了噪声和散射干扰，筛选出了真正与黄龙病相关的关键特征波段，模型检测准确率达98.5% 以上，误判率低于 1.5%。
Smart Images
Figure CN123017262A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of plant disease detection technology, specifically involving a method for accurate identification and detection of citrus Huanglongbing based on near-infrared spectroscopy and partial least squares regression (PLSR) algorithm. Background Technology
[0002] Citrus Huanglongbing (HLB) is a devastating bacterial disease caused by Candidatus Liberibacter asiaticus (CLAs), often referred to as the "cancer" of the citrus industry. Primarily transmitted by the citrus psyllid, it is characterized by a long incubation period, rapid spread, and severe damage. Infected citrus trees exhibit symptoms such as yellowing leaves, deformed fruit, and decreased quality, ultimately leading to tree decline and even death. Currently, there is no effective treatment for HLB; the only feasible control measure is "early detection, early removal, and early treatment."
[0003] Traditional methods for detecting Huanglongbing (HLB) in citrus mainly include visual diagnosis and molecular biological detection. Visual diagnosis relies on experienced technicians and is prone to misdiagnosis in the early stages of the disease or when symptoms are not obvious. Furthermore, it cannot distinguish HLB from similar conditions such as nutrient deficiencies, resulting in high rates of missed detection and false positives. While molecular biological detection methods, such as polymerase chain reaction (PCR), are accurate and reliable, they suffer from drawbacks such as high cost, complex operation, time-consuming and labor-intensive processes, and the need for specialized laboratories and technicians, making them unsuitable for large-scale, rapid field screening.
[0004] Near-infrared spectroscopy is a rapid, non-destructive, and environmentally friendly analytical technique that can obtain information about the chemical composition and structure of molecules by detecting vibrational and rotational energy level transitions. In recent years, near-infrared spectroscopy combined with chemometric methods has been applied to the detection and research of citrus Huanglongbing (HLB). However, current techniques still have the following limitations:
[0005] Most methods employ a single spectral preprocessing approach, which cannot effectively eliminate noise and scattering interference caused by complex field environments.
[0006] Modeling often uses full-band spectral data, which results in high data redundancy, large computational load, and the potential for introducing irrelevant information that affects model accuracy.
[0007] The method for selecting characteristic wavelengths is too simplistic and makes it difficult to identify the key characteristic bands that are truly related to Huanglongbing (HLB).
[0008] The model has poor generalization ability and its detection results are unstable on citrus samples of different varieties, different growth stages and different geographical environments.
[0009] Most of them can only make qualitative judgments (healthy / infected), and cannot accurately assess the degree of infection, which is not conducive to the graded prevention and control of diseases.
[0010] Therefore, developing a near-infrared spectroscopy detection method for citrus Huanglongbing (HLB) with high accuracy, low false positive rate, strong generalization ability, and the ability to simultaneously achieve qualitative and quantitative assessment has significant practical importance and application value. (Invention Content)
[0011] (I) Technical Problem Solved The purpose of this invention is to address the shortcomings of existing technologies by providing a precise PLSR (Pre-Range Sensor Recognition) method for identifying and detecting citrus Huanglongbing (HLB). This method, through improved spectral preprocessing techniques, a combined characteristic wavelength selection method, and an optimized PLSR modeling algorithm, achieves early, rapid, non-destructive, and accurate detection of HLB, effectively solving problems such as high false positive rates, poor generalization ability, and inability to assess infection severity in existing technologies. (II) Technical Solution
[0012] To achieve the above objectives, the present invention adopts the following technical solution:
[0013] A method for accurate identification and detection of near-infrared spectral signals of citrus Huanglongbing (HLB) using PLSR (Precision Near-Infrared Spectral Signals), comprising the following steps:
[0014] Step 1: Sample Collection and Calibration
[0015] Citrus samples were collected from different varieties (such as navel orange, tangerine, and Wogan) and at different growth stages (young trees, mature trees, and fruit-bearing trees) in different citrus producing areas. For each sample, leaf and bark phloem tissues were collected simultaneously. The leaf samples were collected from mature leaves of the current year's spring shoots, and the bark phloem samples were collected from the phloem of one-year-old branches.
[0016] The infection status and severity of Huanglongbing (HLB) in all samples were determined using real-time quantitative PCR (qPCR). Based on the qPCR results, the samples were divided into five categories: healthy samples, mildly infected samples, moderately infected samples, severely infected samples, and samples with nutritional deficiencies (including nitrogen, phosphorus, potassium, and magnesium deficiencies). The ratio of each category was 1:1:1:1:1, and the total sample size was no less than 500.
[0017] Step 2: Near-infrared spectral data acquisition
[0018] Near-infrared reflectance spectra of samples were acquired using a portable near-infrared spectrometer within the wavenumber range of 4000-10000 cm⁻¹, with a spectral resolution of 8 cm⁻¹ and 32 scans. Spectra of each sample were acquired three times at different locations, and the average value was used as the raw spectral data for that sample.
[0019] During spectral acquisition, leaf samples were collected from both sides of the central vein, avoiding the veins and lesions; bark phloem samples were collected from the phloem of one-year-old branches, after scraping off the epidermis. The ambient temperature was controlled at 20-25℃ and the relative humidity at 40%-60% to minimize the impact of environmental factors on the spectral data.
[0020] Step 3: Spectral data preprocessing
[0021] The raw spectral data were processed using a combined preprocessing method of "Standard Normal Variable Transform (SNV) + Multivariate Scattering Correction (MSC) + Savitzky-Golay Second Derivative Smoothing".
[0022] First, the original spectrum is subjected to standard normal variable transformation (SNV) to eliminate spectral differences caused by sample particle size and surface scattering;
[0023] Then, multivariate scattering correction (MSC) is performed to further correct for scattering effects and baseline drift;
[0024] Finally, Savitzky-Golay second-derivative smoothing was used with a window size of 15 points and a polynomial degree of 2 to amplify subtle differences in the spectrum, eliminate background interference, and improve spectral resolution.
[0025] Step 4: Selecting the characteristic wavelength
[0026] A method combining Competitive Adaptive Reweighted Sampling (CARS) and Continuous Projection Algorithm (SPA) was used to screen key characteristic wavelengths related to Huanglongbing from the preprocessed full-band spectral data:
[0027] First, the CARS algorithm is used to initially screen 50-100 candidate characteristic wavelengths from the full-band spectrum. The CARS algorithm, through adaptive reweighted sampling technology combined with the regression coefficients of the PLSR model, can effectively screen wavelength variables that contribute significantly to the model.
[0028] Then, the SPA algorithm is used to further filter out 10-20 optimal feature wavelengths from the candidate wavelengths. Through projection analysis, the SPA algorithm can select the wavelength combination with the minimum collinearity, further reducing data dimensionality and improving model efficiency.
[0029] Through the above combination of methods, the optimal characteristic wavelengths finally selected include: 4320 cm⁻¹, 4670 cm⁻¹, 5180 cm⁻¹, 5760 cm⁻¹, 6320 cm⁻¹, 6890 cm⁻¹, 7450 cm⁻¹, 8010 cm⁻¹, 8570 cm⁻¹, and 9130 cm⁻¹. These characteristic wavelengths mainly correspond to the characteristic absorption peaks of substances such as chlorophyll, water, starch, and cellulose, and can effectively reflect the changes in the internal chemical composition of citrus tissues caused by Huanglongbing infection.
[0030] Step 5: Establishing the PLSR discriminant model
[0031] Using the selected characteristic wavelength data as independent variables and the infection status and degree of infection identified by qPCR as dependent variables, a partial least squares regression (PLSR) discriminant model was established.
[0032] The PLSR algorithm finds the direction of maximum covariance between independent and dependent variables by simultaneously projecting them into a new low-dimensional space. It can effectively solve the problem of multicollinearity among independent variables and is particularly suitable for modeling and analyzing high-dimensional spectral data.
[0033] The number of principal component factors in the model is determined by leave-one-out cross-validation. The number of principal component factors corresponding to the minimum root mean square error of cross-validation (RMSECV) is selected, which is usually 8-12.
[0034] Step 6: Model Validation and Optimization
[0035] All samples were randomly divided into a calibration set and a validation set in a 3:1 ratio. The PLSR model was trained using the calibration set data, and the model was independently validated using the validation set data.
[0036] The following metrics are used to evaluate model performance: Corrected coefficient of determination (Rc²), Predicted coefficient of determination (Rp²), Corrected root mean square error (RMSEC), Predicted root mean square error (RMSEP), Accuracy, and Misclassification rate.
[0037] The model performance was further optimized by adjusting parameters such as the number of principal component factors and the number of feature wavelengths, ensuring that the model accuracy reached over 98.5% and the misclassification rate was less than 1.5%.
[0038] Step 7: Detection of Unknown Samples
[0039] Near-infrared spectral data of the citrus samples to be tested were collected, preprocessed in the same way as in step 3, and characteristic wavelength data were extracted in the same way as in step 4. Then, the data were input into the optimized PLSR model for discrimination.
[0040] The model simultaneously outputs the infection status (healthy / infected) and infection severity (mild / moderate / severe) of the samples. Infection severity is categorized by the predicted value of the PLSR model: predicted value < 0.2 indicates healthy, 0.2 ≤ predicted value < 0.5 indicates mild infection, 0.5 ≤ predicted value < 0.8 indicates moderate infection, and predicted value ≥ 0.8 indicates severe infection.
[0041] Step 8: Model Update
[0042] New sample data is collected regularly to incrementally update the existing PLSR model, with an update cycle of 3-6 months. During model updates, the core parameters of the original model are retained, and the model is fine-tuned using only the new sample data to adapt to changes in different seasons, varieties, and geographical environments, thereby improving the model's generalization ability and long-term stability. (III) Beneficial Effects
[0043] Compared with the prior art, the present invention has the following advantages:
[0044] High detection accuracy: By adopting an improved combined preprocessing method and a combined characteristic wavelength selection method, noise and scattering interference are effectively eliminated, and key characteristic bands that are truly related to Huanglongbing are screened out. The model detection accuracy reaches over 98.5%, and the false judgment rate is less than 1.5%.
[0045] It can distinguish similar diseases: By collecting nutritional deficiency samples and incorporating them into model training, it can effectively distinguish Huanglongbing from similar diseases such as nitrogen deficiency, phosphorus deficiency, potassium deficiency, and magnesium deficiency, thus solving the problem of easy misjudgment in traditional visual diagnosis.
[0046] The established PLSR model can not only make qualitative judgments (healthy / infected), but also accurately assess the degree of infection (mild / moderate / severe), providing a scientific basis for the graded prevention and control of the disease.
[0047] Fast detection speed: The spectral acquisition and data analysis time for a single sample is no more than 1 minute, making it suitable for large-scale rapid field screening.
[0048] Non-destructive testing: It adopts a non-contact spectral acquisition method, which will not cause any damage to the citrus trees and can be repeated multiple times.
[0049] Strong generalization ability: The model update mechanism is introduced, which can adapt to changes in different varieties, different growth stages and different geographical environments, and the model has good long-term stability.
[0050] Easy to operate: Using a portable near-infrared spectrometer, it is easy to operate and does not require a professional laboratory or technicians. Ordinary growers can use it after simple training. Attached Figure Description
[0051] To provide a more detailed description of the technical solutions of the embodiments of this invention, the accompanying drawings used in the following description of the embodiments are briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on such drawings without creative effort.
[0052] Figure 1 is a flowchart of the near-infrared spectroscopy detection method for citrus Huanglongbing of the present invention;
[0053] Figure 2 shows the original near-infrared spectra of healthy and infected citrus leaves;
[0054] Figure 3 shows the near-infrared spectrum after combined preprocessing;
[0055] Figure 4 illustrates the process of the CARS algorithm in selecting feature wavelengths.
[0056] Figure 5 is a scatter plot of the predicted and actual values of the PLSR model. Detailed Implementation
[0057] The present invention will be further described in detail below with reference to specific embodiments.
[0058] Example 1
[0059] This embodiment provides a method for accurate identification and detection of near-infrared spectral signals of citrus Huanglongbing (HLB) using PLSR (Pressure Near-Infrared Spectral Signals). The specific steps are as follows:
[0060] Sample collection and calibration
[0061] Navel orange samples were collected from multiple citrus orchards in Ganzhou City, Jiangxi Province, totaling 500 samples. Among them, 100 were healthy, 100 were mildly infected, 100 were moderately infected, 100 were severely infected, and 100 were nutritionally deficient (including 25 samples deficient in nitrogen, 25 in phosphorus, 25 in potassium, and 25 in magnesium).
[0062] All samples were labeled using real-time quantitative PCR (qPCR). Total DNA was extracted from the phloem tissue of the samples, and qPCR detection was performed using CLas-specific primers and probes. Infection status and severity were determined based on Ct values: Ct value > 36 indicated healthy individuals, 30 ≤ Ct value ≤ 36 indicated mild infection, 24 ≤ Ct value < 30 indicated moderate infection, and Ct value < 24 indicated severe infection.
[0063] Near-infrared spectral data acquisition
[0064] Near-infrared reflectance spectral data of the samples were acquired using a Thermo Fisher Anastar II portable near-infrared spectrometer. The spectral acquisition range was 4000–10000 cm⁻¹, the spectral resolution was 8 cm⁻¹, and the number of scans was 32.
[0065] Spectra were collected from three different locations on either side of the midrib of each leaf sample, and from three different locations on each one-year-old branch of each bark phloem sample. The average value was taken as the raw spectral data for that sample. The ambient temperature was 23℃ and the relative humidity was 50%.
[0066] Spectral data preprocessing
[0067] The raw spectral data were processed using a combined preprocessing method of "Standard Normal Transform (SNV) + Multivariate Scattering Correction (MSC) + Savitzky-Golay Second Derivative Smoothing". The Savitzky-Golay second derivative smoothing used a 15-point window and a polynomial degree of 2.
[0068] Figure 2 shows the original near-infrared spectra of healthy and infected citrus leaves. It can be seen that the overall trends of the spectral curves of healthy and infected leaves are similar, but there are subtle differences in some bands. Figure 3 shows the near-infrared spectra after combined preprocessing. It can be seen that the baseline drift of the spectrum was effectively corrected after preprocessing, and the subtle differences in the spectrum are more obvious.
[0069] Characteristic wavelength selection
[0070] First, the CARS algorithm was used to initially screen 75 candidate characteristic wavelengths from the full-band spectrum. Then, the SPA algorithm was used to further screen the candidate wavelengths to select 10 optimal characteristic wavelengths, which are: 4320cm⁻¹, 4670cm⁻¹, 5180cm⁻¹, 5760cm⁻¹, 6320cm⁻¹, 6890cm⁻¹, 7450cm⁻¹, 8010cm⁻¹, 8570cm⁻¹, and 9130cm⁻¹.
[0071] Figure 4 shows the process of the CARS algorithm in selecting characteristic wavelengths. It can be seen that as the number of sampling times increases, the number of wavelength variables retained gradually decreases, and RMSECV first decreases and then increases, reaching its minimum value when the number of sampling times is 30.
[0072] PLSR discriminant model establishment
[0073] Using the selected 10 characteristic wavelength data as independent variables and the infection degree values calibrated by qPCR (0 for healthy, 0.33 for mild infection, 0.67 for moderate infection, and 1 for severe infection) as dependent variables, a PLSR discriminant model was established.
[0074] The number of principal component factors in the model was determined to be 10 using leave-one-out cross-validation.
[0075] Model Validation and Optimization
[0076] The 500 samples were randomly divided into a calibration set (375 samples) and a validation set (125 samples) in a 3:1 ratio. The PLSR model was trained using the calibration set data, and the model was independently validated using the validation set data.
[0077] The model performance evaluation results are as follows:
[0078] Corrected coefficient of determination (Rc²): 0.987
[0079] Predictional determination coefficient (Rp²): 0.982
[0080] Corrected root mean square error (RMSEC): 0.042
[0081] Root mean square error of prediction (RMSEP): 0.051
[0082] Accuracy: 98.4%
[0083] False positive rate: 1.6%
[0084] Figure 5 is a scatter plot of the predicted and actual values of the PLSR model. It can be seen that the predicted and actual values have a good correlation, and most of the sample points are distributed near the diagonal.
[0085] Unknown sample detection
[0086] Two hundred navel orange leaf samples of unknown condition were randomly collected. Spectral data were acquired and preprocessed using the method described above, followed by characteristic wavelength extraction. These data were then input into the optimized PLSR model for detection. Simultaneously, qPCR was used to validate these samples.
[0087] The results showed that the consistency between the PLSR model and the qPCR model was 98.5%, with 2 samples being misclassified, resulting in a misclassification rate of 1.5%.
[0088] Example 2
[0089] This embodiment verifies the applicability of the method of the present invention to different citrus varieties.
[0090] A total of 300 mandarin orange samples were collected in Zhaoqing City, Guangdong Province, including 75 healthy samples, 75 mildly infected samples, 75 moderately infected samples, and 75 severely infected samples. Spectral data were collected and preprocessed and characteristic wavelengths extracted according to the method in Example 1, and then the PLSR model established in Example 1 was used for detection.
[0091] Meanwhile, these samples were validated using qPCR. The results showed that the detection accuracy of the PLSR model was 97.7%, and the false positive rate was 2.3%, indicating that the method of the present invention has good applicability to different citrus varieties.
[0092] Example 3
[0093] This embodiment verifies the ability of the method of the present invention to detect early-stage Huanglongbing (HLB).
[0094] Samples were collected from navel orange seedlings artificially inoculated with CLas pathogen. Leaf samples were collected at 1, 2, and 3 months post-inoculation, for a total of 150 samples. Spectral data were collected and analyzed according to the method described in Example 1, and validation was performed using qPCR.
[0095] The results show:
[0096] One month post-vaccination: The accuracy rate of the PLSR model detection was 92.0%.
[0097] Two months post-vaccination: The accuracy rate of the PLSR model detection was 96.0%.
[0098] Three months post-vaccination: The accuracy rate of the PLSR model detection was 98.0%.
[0099] This indicates that the method of the present invention can achieve early detection before the symptoms of Huanglongbing appear, thus gaining valuable time for the early prevention and control of the disease.
[0100] The above description is only a preferred embodiment of the present invention. All equivalent changes and modifications made within the scope of the claims of the present invention should be included in the scope of the present invention.
Claims
1. Claims A method for accurate identification and detection of near-infrared spectral signals of citrus Huanglongbing (HLB) using PLSR (Precision Near-Infrared Spectral Signal), characterized in that, Includes the following steps: Sample collection and labeling: Citrus leaf and bark phloem samples from different growth stages and varieties were collected, and the Huanglongbing infection status and degree were labeled using real-time quantitative PCR (qPCR). Near-infrared spectral data acquisition: Near-infrared reflectance spectral data of samples were acquired using a portable near-infrared spectrometer in the wavenumber range of 4000-10000 cm⁻¹. Each sample was acquired 3 times and the average value was taken. Spectral data preprocessing: The raw spectral data were processed using a combined preprocessing method of "Standard Normal Variable Transform (SNV) + Multivariate Scattering Correction (MSC) + Savitzky-Golay Second Derivative Smoothing". Feature wavelength selection: A combination of competitive adaptive reweighted sampling (CARS) and continuous projection algorithm (SPA) was used to screen out key feature wavelengths related to Huanglongbing from the preprocessed full-band spectral data; PLSR discriminant model establishment: Using the selected characteristic wavelength data as independent variables and the infection status and degree of infection identified by qPCR as dependent variables, a partial least squares regression (PLSR) discriminant model was established. Model validation and optimization: The model was validated using cross-validation and independent validation sets, and its performance was optimized by adjusting the number of principal component factors. Unknown sample detection: Near-infrared spectral data of the citrus sample to be tested are collected, and after the same preprocessing and characteristic wavelength selection, they are input into the optimized PLSR model for discrimination and the detection results are output.
2. The detection method according to claim 1, characterized in that, The sample collection in step (1) includes healthy samples, mildly infected samples, moderately infected samples, severely infected samples, and samples with nutritional deficiencies such as nitrogen, phosphorus, potassium, and magnesium deficiency. The ratio of the number of each type of sample is 1:1:1:1:
1.
3. The detection method according to claim 1, characterized in that, When collecting spectral data in step (2), the leaf sample collection site is the middle part of the leaf veins on both sides, and the bark phloem sample collection site is the phloem of one-year-old branches. The ambient temperature is controlled at 20-25℃ and the relative humidity is controlled at 40%-60%.
4. The detection method according to claim 1, characterized in that, The window size for Savitzky-Golay second derivative smoothing in step (3) is 15 points, and the polynomial degree is 2.
5. The detection method according to claim 1, characterized in that, The specific method for selecting characteristic wavelengths in step (4) is as follows: First, the CARS algorithm is used to initially screen 50-100 candidate characteristic wavelengths from the full-band spectrum, and then the SPA algorithm is used to further screen 10-20 optimal characteristic wavelengths from the candidate wavelengths.
6. The detection method according to claim 1 or 5, characterized in that, The optimal characteristic wavelengths selected in step (4) include: 4320cm⁻¹, 4670cm⁻¹, 5180cm⁻¹, 5760cm⁻¹, 6320cm⁻¹, 6890cm⁻¹, 7450cm⁻¹, 8010cm⁻¹, 8570cm⁻¹, and 9130cm⁻¹.
7. The detection method according to claim 1, characterized in that, The number of principal component factors in the PLSR discriminant model described in step (5) is determined by leave-one-out cross-validation, and the number of principal component factors is 8-12.
8. The detection method according to claim 1, characterized in that, The PLSR discrimination model described in step (5) simultaneously outputs the infection status (healthy / infected) and infection degree (mild / moderate / severe) of the sample. The infection degree is divided by the predicted value of the PLSR model: predicted value < 0.2 is healthy, 0.2≤predicted value < 0.5 is mild infection, 0.5≤predicted value < 0.8 is moderate infection, and predicted value ≥ 0.8 is severe infection.
9. The detection method according to claim 1, characterized in that, It also includes a model update step: regularly collecting new sample data and incrementally updating the existing PLSR model, with an update cycle of 3-6 months.
10. The application of the detection method according to claim 1 in rapid field screening of citrus Huanglongbing (HLB).