Rapid detection method for citrus huanglongbing
By combining the Fourier transform infrared spectroscopy and convolutional neural network methods, the 1DConvLSTM-SA model was established, which solved the rapid, accurate and economic problems of citrus yellow dragon disease detection, and realized disease monitoring and prevention and control of the citrus industry.
Patent Information
- Application Number
- CN202510594380.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-09
- Publication Date
- 2025-08-19
AI Technical Summary
The prior art is difficult to detect citrus yellow dragon disease quickly, accurately and economically. Traditional methods such as PCR and ELISA are costly and complex in operation, while visual detection is low, and FT-IR technology has insufficient classification accuracy when processing high-dimensional spectral data.
Combining the Fourier transform infrared spectroscopy and convolutional neural network, by establishing a 1DConvLSTM-SA qualitative classification discriminant model, FT-IR data is used to extract and classify citrus leaves, and quickly detect.
It has achieved early, fast and accurate diagnosis of citrus Huanglong disease, with low detection cost, short cycle and high accuracy, and is suitable for large-scale field testing.
Smart Images
Figure CN120507306A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of fruit tree disease diagnosis, and more particularly to a rapid detection method for citrus Huanglongbing disease. Background Art
[0002] Citrus Huanglongbing (HLB), also known as citrus "greening disease," is one of the most serious diseases of citrus crops. Since its global outbreak in the late 20th century, HLB has caused significant economic losses to citrus cultivation worldwide. Research shows that HLB is caused by the bacterium Candidatus Liberibacterasiaticus (CLas), and its damage to plants is primarily manifested in three aspects: First, CLas significantly reduces chlorophyll content, causing yellowing of leaves and decreased photosynthesis capacity; second, infected citrus leaves often show abnormal starch deposition, affecting normal metabolic function; finally, CLas can cause serious obstruction of nutrient transport, leading to poor fruit development, gradual weakening of the tree, and ultimately, the death of the entire tree.
[0003] Early detection of HLB is crucial for blocking the spread of the disease. However, traditional detection methods have certain limitations in practical applications. Molecular biology methods, such as polymerase chain reaction (PCR) and fluorescent quantitative PCR (q-PCR), although with high sensitivity and accuracy, are often difficult to meet the needs of large-scale field testing due to their high requirements for test equipment and operating skills. Enzyme-linked immunosorbent assay (ELISA) is also limited in its application in large-scale screening due to the need for specific antibodies, high reagent costs, and long detection cycles. In addition, traditional visual field detection methods usually rely on manual experience or computer vision software. Although convenient and fast, they have low accuracy, especially in the early stages of the disease when symptoms are not yet obvious, which can easily lead to misdiagnosis and missed diagnosis. Therefore, there is an urgent need for a fast, accurate and inexpensive means for the precise identification of HLB.
[0004] In recent years, Fourier transform infrared spectroscopy (FT-IR), as a highly efficient spectral analysis technique, has demonstrated outstanding advantages in plant disease detection, including high sensitivity, rapidity, ease of operation, and wide applicability. By detecting molecular vibrational information in plant samples, FT-IR can rapidly generate chemical fingerprints that reflect changes in the chemical composition of plant tissues. Studies have shown that FT-IR technology can detect HLB infection and has great potential for early diagnosis of other plant diseases such as wheat fusarium head blight, soybean cyst nematode disease, tomato early blight, and rice blast. However, FT-IR technology also faces challenges in practical application, especially when processing high-dimensional and complex spectral data. Traditional data analysis methods (such as principal component analysis, partial least squares, and linear discriminant analysis) still have certain shortcomings in feature extraction and classification accuracy.
[0005] With the rapid development of deep learning technology, convolutional neural networks (CNNs) have achieved significant breakthroughs in image recognition and are widely used in fields such as medical image analysis, autonomous driving, and industrial inspection. Through multi-layer convolution and pooling operations, CNNs can automatically extract key features from raw data and effectively classify and identify them. In the field of plant diseases, the advantages of CNNs in image recognition, symptom grading, monitoring, and prediction have been widely demonstrated. However, no research has systematically combined CNNs with Fourier transform infrared spectroscopy for the rapid detection of HLB. Summary of the Invention
[0006] In view of this, the object of the present invention is to provide a rapid detection method for citrus Huanglongbing to address the deficiencies in the prior art.
[0007] In order to achieve the above object, the present invention adopts the following technical solutions:
[0008] A rapid detection method for citrus Huanglongbing disease specifically comprises the following steps:
[0009] (1) Establishment of standard sample spectrum set
[0010] Fourier transform infrared spectrometer was used to detect and read FT-IR data of leaves of plants infected with citrus Huanglongbing fungus and those without Huanglongbing fungus;
[0011] (2) Establishment of qualitative classification and discrimination model
[0012] After correction and preprocessing, the FT-IR data was input into the 1DConvLSTM-SA model for training and verification. After verification on the training set and the verification set, the optimal 1DConvLSTM-SA qualitative classification model was obtained.
[0013] (3) Qualitative identification of unknown samples
[0014] The optimal 1DConvLSTM-SA qualitative classification discriminant model is used to directly distinguish whether the citrus leaf samples to be tested are infected with citrus Huanglongbing.
[0015] Furthermore, in the above step (1), the model of the Fourier transform infrared spectrometer is Nicolet iS10.
[0016] Furthermore, in the above step (1), the spectrum acquisition software used by the Fourier transform infrared spectrometer is OMNIC, and the measurement wavelength range is 4000cm -1 -400cm -1 , with a resolution of 8cm -1 , the number of scans is 32 times.
[0017] Furthermore, in the above step (1), the detection degree of the Fourier transform infrared spectrometer is the sample slices made by crushing the main leaf veins and mixing them with spectral grade KBr, and the leaf age is the autumn mature leaves of the current year or the previous year.
[0018] Furthermore, in the above step (2), in order to enable the model to correctly classify citrus leaf samples infected with citrus Huanglongbing from citrus leaf samples without the infection, the evaluation parameters of the 1DConvLSTM-SA qualitative classification discriminant model include accuracy (Accuracy), precision (Precision, P), recall (Recall, R), F1 score (F1-score) and false negative rate (FNR).
[0019] Furthermore, in the above step (2), the evaluation criteria for establishing the 1DConvLSTM-SA qualitative classification and discrimination model are as follows: (1) Accuracy, which is used to measure the accuracy of the overall classification of the model. The higher the accuracy, the stronger the classification ability of the model; (2) Precision (P), which indicates the proportion of samples predicted as positive by the model that are actually positive, reflecting the reliability of the model for positive samples; (3) Recall (R), which indicates the proportion of all actual positive samples that are correctly classified. A high recall rate indicates that the model has a strong ability to identify positive samples; (4) F1 score (F1-score), which is a balanced indicator of precision and recall, and can more comprehensively evaluate the classification performance of the model when the data is unbalanced; (5) False negative rate (FNR), which measures the proportion of positive samples that cannot be detected. The lower the value, the lower the model's missed judgment rate for positive samples. It is particularly suitable for tasks with strict requirements on false negatives, such as disease detection.
[0020] Furthermore, in the above step (2), the moving window fitting polynomial smoothing method (Savitzky-Golay, SG) is used to eliminate the influence of noise on the spectral signal, and the first-order derivative is performed to eliminate the baseline drift and background influence. The random sample partitioning method is used to divide the training set and the validation set.
[0021] Furthermore, in the above step (2), the citrus leaf spectral data samples in the training set and the validation set are both selected from the standard sample spectral set.
[0022] It can be seen from the above technical solution that compared with the prior art, the beneficial effects of the present invention are as follows:
[0023] 1. Unlike traditional image recognition methods, convolutional neural networks can process complex high-dimensional data and optimize network parameters through backpropagation algorithms to improve classification accuracy.
[0024] 2. By combining convolutional neural networks with Fourier transform infrared spectroscopy, the present invention can utilize the powerful feature learning ability of convolutional neural networks to perform deep feature extraction on spectral data, thereby avoiding the limitations of manual feature selection and further improving the detection accuracy and classification efficiency of HLB.
[0025] 3. The present invention combines the advantages of Fourier transform infrared spectroscopy technology and convolutional neural networks, and proposes a Fourier transform infrared spectroscopy HLB rapid detection method based on an improved convolutional neural network. First, the infrared spectrum data of citrus leaves are obtained using a Fourier transform infrared spectrometer, and then the data are analyzed and classified by the constructed deep learning model. Finally, the data are compared with traditional classification models (PCA, PCA-LDA, PLS-LDA) and q-PCR detection methods in terms of accuracy, cost, and detection time. Through the implementation of this study, it is expected to provide a more efficient, accurate, and economical detection method for disease monitoring and prevention and control in the citrus industry, curb the spread of HLB, ensure the sustainable development of the citrus industry, and provide new ideas for the detection and identification of other agricultural diseases.
[0026] 4. Based on FT-IR data of citrus leaves collected in the field, the present invention establishes a discrimination model that can qualitatively classify and discriminate citrus Huanglongbing, and establishes a rapid FT-IR detection method for citrus Huanglongbing, so as to achieve early, rapid, and accurate diagnosis of citrus Huanglongbing. Compared with the currently commonly used q-PCR detection technology, the present invention has a low detection cost for each sample and can also greatly shorten the detection cycle of citrus Huanglongbing in production. For a single plant of different varieties, only 3-5 leaves need to be scanned, which takes only 5-7 minutes to achieve rapid diagnosis of citrus Huanglongbing. It is simple to operate, fast, accurate, and cost-effective. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] Figure 1 The FT-IR raw data curves of the pomelo leaf samples infected with citrus Huanglongbing and the pomelo leaf samples without the infection;
[0028] Figure 2 This is the prediction result of the optimal 1DConvLSTM-SA qualitative classification discriminant model. DETAILED DESCRIPTION
[0029] The following is a clear and complete description of the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.
[0030] Example 1
[0031] This embodiment uses a Nicolet iS10 FT-IR instrument to detect and read the FT-IR data of tangerine and pomelo leaves infected with citrus Huanglongbing disease and those without Huanglongbing disease, and simultaneously applies a 1DConvLSTM-SA model for qualitative classification and discrimination to distinguish whether the tangerine and pomelo leaf samples to be tested are infected with citrus Huanglongbing disease.
[0032] The rapid detection method for citrus Huanglongbing specifically includes the following steps:
[0033] (1) Establishment of standard sample spectrum set
[0034] The FT-IR data of the citrus pomelo leaf samples without bacteria and the citrus pomelo leaf samples infected with citrus Huanglongbing were detected and read by FT-IR instrument (NicoletiS10). The spectrum acquisition software used was OMNIC, and the measurement wavelength range was 4000 cm -1 -400cm -1 , with a resolution of 8cm -1 , the number of scans is 32 times;
[0035] (2) Establishment of qualitative classification and discrimination model
[0036] Using Pycharm 2024.3.3 software, based on 1D-CNN, combined with the LSTM module and a custom SA module, a qualitative classification model for citrus Huanglongbing diseased leaves was established. The moving window fitting polynomial smoothing method was used to eliminate the influence of noise on the spectral signal. The first-order derivative was performed to eliminate baseline drift and background effects. The random sample partitioning method was used to divide the training set and the validation set.
[0037] After correction and preprocessing, the FT-IR data was input into the 1DConvLSTM-SA model for training and verification. After verification on the training set and the verification set, the optimal 1DConvLSTM-SA qualitative classification model was obtained.
[0038] (3) Qualitative identification of unknown samples
[0039] The optimal 1DConvLSTM-SA qualitative classification discriminant model is used to directly distinguish whether the tested tangerine and pomelo leaf samples are infected with citrus Huanglongbing.
[0040] Performance Testing
[0041] 1. Rapid detection of citrus Huanglongbing on tangerine and pomelo leaves
[0042] The method of Example 1 was used to detect citrus Huanglongbing disease. 300 samples of infected tangerine and pomelo leaves were selected, including 210 training sets and 90 validation sets; 150 samples of uninfected tangerine and pomelo leaves were selected, including 105 training sets and 45 validation sets.
[0043] Figure 1 The FT-IR raw data curves of tangerine pomelo leaf samples infected with citrus Huanglongbing and tangerine pomelo leaf samples without bacteria. Figure 1It can be seen that the spectral curve of the tangerine and pomelo leaf sample infected with citrus Huanglongbing (yellow curve) and the citrus leaf sample without bacteria (blue curve) are in the range of 800-1000 cm -1 There is an obvious baseline drift in the band, and there are also large differences in the peak values of the absorption peaks.
[0044] Figure 2 is the prediction result of the optimal 1DConvLSTM-SA qualitative classification and discrimination model. Figure 2 As can be seen in the figure, the model correctly classified tangerine pomelo leaf samples infected with citrus greening from those without the disease. The validation set classification accuracy was calculated to be 95.4%.
[0045] 2. Verification of the applicability of the qualitative classification model for citrus Huanglongbing
[0046] To validate the applicability of the model developed using the detection method in Example 1, field samples from nine different citrus varieties were collected from orchards in five locations. Each orchard was randomly sampled using a five-point random sampling method, with 10 samples collected at each point, for a total of 90 samples from the five orchards. Real-time fluorescence quantitative PCR results revealed 67 bacterial samples and 23 uninfected samples from this batch of field samples. All citrus leaf samples were preprocessed using an FT-IR instrument according to optimal parameters and then fed into the optimal 1DConvLSTM-SA qualitative classification model for classification prediction. The discrepancies between the model's predicted values and the theoretical values were tested.
[0047] Table 1 summarizes the training set accuracy, validation set accuracy, test set accuracy, and false negative rate of the optimal 1DConvLSTM-SA qualitative classification discriminant model.
[0048] Accuracy:
[0049] Precision (P):
[0050] Recall (R):
[0051] F1-score:
[0052] False Negative Rate (FNR):
[0053] In formulas (1)-(5), TP represents the number of positive samples, N represents the number of negative samples, FP represents the number of false positive samples, and FNR represents the false negative rate.
[0054] Table 1 Prediction results of 90 citrus variety samples using the 1DconvLSTM-SA model
[0055] Training set accuracy / % Validation set accuracy / % Test set accuracy / % False negative rate / % 95.7 93.5 93.4 7.4
[0056] Table 1 shows that during the training and validation process, the model achieved comparable accuracy on both the training and validation sets, indicating that the model did not suffer from overfitting or inaccurate predictions and that the model's predictions were close to the theoretical values. Furthermore, the model achieved an accuracy rate of 93.4% and a false negative rate of only 7.4% for the FT-IR dataset of nine citrus varieties used in the test set. This demonstrates that the model's prediction accuracy is sufficient to meet the requirements for qualitative classification of field samples.
[0057] The verification results of 9 different citrus varieties in orchards in 5 regions showed that the correct recognition rate of the 1DConvLSTM-SA model was above 93.4%, with a maximum of 95.7%, once again verifying the model's accurate prediction ability. The model can be further used for rapid field diagnosis of citrus Huanglongbing disease.
[0058] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A rapid detection method for citrus Huanglongbing, characterized in that: The specific steps include: (1) Establishment of standard sample spectrum set Fourier transform infrared spectrometer was used to detect and read FT-IR data of leaves of plants infected with citrus Huanglongbing fungus and those without Huanglongbing fungus; (2) Establishment of qualitative classification and discrimination model After correction and preprocessing, the FT-IR data was input into the 1DConvLSTM-SA model for training and verification. After verification on the training set and the verification set, the optimal 1DConvLSTM-SA qualitative classification model was obtained. (3) Qualitative identification of unknown samples The optimal 1DConvLSTM-SA qualitative classification discriminant model is used to directly distinguish whether the citrus leaf samples to be tested are infected with citrus Huanglongbing.
2. A rapid detection method for citrus Huanglongbing according to claim 1, characterized in that: In step (1), the model of the Fourier transform infrared spectrometer is Nicoleti S10.
3. A rapid detection method for citrus Huanglongbing according to claim 1, characterized in that: In step (1), the spectrum acquisition software used by the Fourier transform infrared spectrometer is OMNIC, and the measurement wavelength range is 4000cm -1 -400cm -1 , with a resolution of 8cm -1 , the number of scans is 32 times.
4. A rapid detection method for citrus Huanglongbing according to claim 1, characterized in that: In step (1), the detection degree of the Fourier transform infrared spectrometer is a sample slice made by mixing the main leaf veins with spectral grade KBr after crushing, and the leaf age is the autumn mature leaves of the current year or the previous year.
5. The rapid detection method for citrus Huanglongbing according to claim 1, characterized in that: In step (2), the evaluation parameters of the 1DConvLSTM-SA qualitative classification discriminant model include accuracy, precision, recall, F1 score and false negative rate.
6. The rapid detection method for citrus Huanglongbing according to claim 1, characterized in that: In step (2), the moving window fitting polynomial smoothing method is used to eliminate the influence of noise on the spectral signal, and the first-order derivative is performed to eliminate the baseline drift and background influence. The random sample partitioning method is used to divide the training set and the validation set.
7. The rapid detection method for citrus Huanglongbing according to claim 1, characterized in that: In step (2), the citrus leaf spectral data samples in the training set and the validation set are both selected from the standard sample spectral set.