Candidatus liberobacter asiaticum prediction method for analyzing content of callose in citrus vein conduit through microspectrum

Abstract

The invention provides a Huanglongbing prediction method for analyzing the content of the callose of the citrus vein duct through microspectrum, which comprises the following steps of: detecting the change of the callose by adopting a microspectrum image aiming at the callose of the citrus vein duct through a spectrum image method, and analyzing whether a plant has a defensive reaction of Huanglongbing infection so as to predict the Huanglongbing infection of the plant; according to the invention, the early diagnosis effect is better, and the accuracy is higher.

Description

Technical Field

[0001] The invention relates to the technical field of agricultural plant disease prevention and control, in particular to a method for predicting Huanglongbing by analyzing the callose content in citrus leaf veins using microspectroscopy. Background Art

[0002] Citrus is a globally cultivated and important economic crop. Huanglongbing (Huanglongbing) remains a devastating, incurable threat to the healthy growth of citrus. Early diagnosis of Huanglongbing allows for the timely detection and removal of diseased trees, preventing them from continuing to consume nutrients and occupy space in the later stages of the disease, thereby reducing direct economic losses due to decreased fruit yield and quality. This ensures the number and growth of healthy trees in orchards, stabilizes citrus production and quality, meets market demand, and maintains the stable development of the citrus industry. This prevents the widespread spread of the disease, minimizes damage to the orchard's ecological environment, maintains the ecological balance within the orchard, and mitigates negative impacts on surrounding biodiversity.

[0003] Traditional citrus Huanglongbing disease detection methods rely on manual observation, examining appearance features such as yellowing and mottled leaves and fruit deformities. However, this method is highly subjective, inaccurate, and difficult to detect in its early stages.

[0004] Existing spectral imaging methods use the reflectance and absorption characteristics of citrus leaves at different wavelengths to provide rich, detailed information for analysis. However, these methods, which focus on detecting citrus fruits after they show symptoms, lack detailed information about early-stage citrus fruits before symptoms appear. Summary of the Invention

[0005] The present invention proposes a Huanglongbing prediction method based on microspectral analysis of the callose content in citrus leaf veins, which can achieve better and more accurate early diagnosis.

[0006] The present invention adopts the following technical solutions.

[0007] A method for predicting Huanglongbing by analyzing the callose content in citrus leaf veins using microspectral analysis is characterized in that the method uses microspectral imaging of the callose in citrus leaf veins to detect changes in the callose through spectral imaging methods, analyzes whether the plant has a defense response to Huanglongbing infection, and predicts the plant's Huanglongbing infection.

[0008] The method comprises the following steps:

[0009] Step S0, sample processing, specifically: after slicing the collected citrus leaf samples, microscopic hyperspectral imaging of the leaf vein duct part is collected to obtain spectral image data of the leaves.

[0010] Step S1, image preprocessing: preprocessing the acquired spectral image, including image denoising, spectral correction, and image scale adjustment, to obtain the spectral characteristics of the leaf vein duct area to facilitate subsequent analysis and identification;

[0011] Step S2, vascular region segmentation: in the microscopic hyperspectral image of the citrus leaf vein vessel, the vascular region is separated from the entire leaf image by image processing methods;

[0012] Step S3, callose content prediction;

[0013] Step S4: disease condition prediction and detection.

[0014] In the slicing method of step S0, the leaves are random leaves of citrus plants, and the leaf samples are collected from the 1 / 3 to 2 / 3 area of ​​the main vein of the leaves; the slicing equipment uses a slicer or a blade, and the slice thickness ranges from 30-50um; if some varieties or samples need to reduce the thickness, it must not be less than 10um.

[0015] The leaf spectral data acquisition method in step S0 is as follows: the leaf is placed in a dark box and spectral data is acquired using a spectrometer, with only full-band spectral fill light providing illumination. The optimal number of spectral data bands is 256, but should be at least greater than 32, and the band range is 300 to 1000 nm. The image data is in the form of an image, image stream, video, or video stream, with a microscopic magnification of 10-40x.

[0016] In step S1, image denoising is performed using a linear mean filter with a 3×3 neighborhood window. Spectral correction uses a full-band reflector with a reflectance of 0.8. Scaling is performed using bilinear interpolation for upscaling and averaging for downscaling. The optimal scale is 320×480, the maximum scale should be less than or equal to 1080×1080, and the minimum scale should be greater than or equal to 180×180. Scaling is evaluated based on the prominence of callose areas.

[0017] In step S2, the region segmentation process is performed for objects with different band counts. For image data with a scale of less than or equal to 320×480, the segmentation method is to identify the leaf vein region through image texture conversion, edge extraction, and threshold segmentation. Texture conversion uses the gray-level co-occurrence matrix (GLCM) method, which calculates the gray-level co-occurrence of pixel pairs at different orientations (0°, 45°, 90°, and 135°) and at different distances in the image to generate a gray-level co-occurrence matrix. Multiple texture feature parameters, such as contrast, correlation, energy, and entropy, are extracted to enhance the texture image of the vein region. Edges are detected by calculating the gray-level gradient within a pixel neighborhood using the Sobel and Canny operators. The gradient magnitude and direction are then calculated, and edge pixels are identified through non-maximum suppression and double-threshold detection. Segmentation is then achieved using maximum inter-class variance.

[0018] In step S2, the gray-level co-occurrence matrix (GLCM) is used to analyze the gray-level relationship of pixels in the image. It is based on the statistics of the gray-level values ​​of pixel pairs in the image and describes the texture characteristics of the image through statistical information. Its usage methods include: method a - pixel pair statistics, method b - constructing a gray-level co-occurrence matrix, and method c - calculating gray-level co-occurrence evidence results;

[0019] The statistical method of pixel pairs is that a pixel pair refers to a pair of pixels in an image at a specified distance and direction. Let the distance be d and the direction be r, which are parameters. For example, d = 1, r = 45; then, traverse the image to count pixel pairs and count the frequency of occurrence of the grayscale values ​​of these two pixels; and normalize the statistics according to the frequency of occurrence; constructing the grayscale co-occurrence matrix means that after the statistics of method a, the pixel pairs are arranged according to the level difference to form a matrix G, then G is the grayscale co-occurrence matrix, and at this time the matrix G = {g i，j} in the element g i，j It represents the frequency of occurrence of pixel pairs of levels i and j;

[0020] In method c, the results of the gray-level co-occurrence matrix include image contrast, correlation, energy, and entropy. The formula for image contrast is expressed as:

[0021]

[0022] Where L is the level, such as 16 or 32;

[0023] Correlation: Measures the linear correlation of pixel grayscale values ​​in an image. The calculation formula is

[0024]

[0025] In the formula, u is the mean and σ is the standard deviation. The greater the correlation, the more similar the local structures of the image textures are. Energy: reflects the uniformity of the image grayscale distribution and the coarseness of the texture. The calculation formula is

[0026] The larger the energy value, the more uniform the image grayscale distribution and the finer the texture; the smaller the energy value, the coarser the image texture. Entropy: measures the richness of image information and is calculated as

[0027]

[0028] The larger the entropy value, the more complex the grayscale distribution in the image and the richer the texture information; the smaller the entropy value, the simpler the image texture.

[0029] In step S2, the region segmentation process is implemented for objects with different numbers of bands. The segmentation method for image data with a scale greater than 320×480 is: segmentation is achieved through a convolutional neural network (CNN) model. First, the training data and test data are labeled. Then, a deep neural network (CNN) is established through multiple convolutional layers, pooling layers, and fully connected layers. The CNN is trained to automatically learn the characteristic patterns of the catheter area in the image to achieve region segmentation.

[0030] In step S3, the callose content prediction is achieved through three steps: spectral information extraction, characteristic band extraction, and callose content prediction, specifically:

[0031] Spectral information is extracted from the area after segmenting the catheter: the band curve is extracted using the regional average statistical method with error. The error is implemented using the standard deviation. When the variance is greater than a threshold (preferably 10%), a binary method is used to reduce the area to ensure the purity of the spectral information extracted.

[0032] Feature band extraction: PCA or SPA is used to implement it. The number of extracted features should preferably be 32 and should be greater than or equal to 5 to ensure that key information is not lost;

[0033] Prediction of callose content: Multiple linear regression (less than or equal to 10) or support vector machine regression (greater than 10) is used according to the number of features. The multiple linear regression is expressed as:

[0034] y=Σβ i x i ,

[0035] where x i is the parameter, β i is a parameter corresponding to the characteristic band. Support vector machine regression is expressed as f(x)=w T x+b, where w is the parameter matrix, x corresponds to the characteristic band, and b is the offset.

[0036] In step S4, the disease detection method is as follows: based on the callose content, the prediction is made in combination with the month. The detection categories of the disease severity of Huanglongbing include: infection and health;

[0037] The formula is expressed as

[0038] in is the standard callose content, x is the month, k and b are parameters. When , it is infected, and less than , it is healthy.

[0039] In actual testing applications, the accuracy of the present invention is greater than 93%.

[0040] The advantages of the present invention are:

[0041] 1. High Accuracy: The early symptoms of Huanglongbing are subtle and difficult to detect using traditional observation methods. However, callose in citrus leaf veins begins to increase in content and shift in location even in the early stages of infection. Using spectral imaging technology to detect changes in callose and related characteristics allows for more effective and accurate early diagnosis.

[0042] 2. This invention is more targeted: Existing methods rely on complex, comprehensive assessments of leaf appearance to infer disease conditions. However, spectral imaging technology for callose detection focuses directly on callose within the venous vessels of citrus leaves. Callose is a plant's direct defense against Huanglongbing infection. In the early stages of infection, the interaction between the pathogen and plant cells directly triggers the synthesis and deposition of callose as a defensive response. Directly detecting changes in callose through spectral imaging is like directly detecting the plant's most direct physiological response to Huanglongbing infection. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0044] Attachment Figure 1 1 is a schematic diagram of the extraction result of the duct area of ​​citrus leaf veins in an embodiment of the present invention;

[0045] Attachment Figure 2 FIG. 1 is a schematic diagram of a disease condition detection method based on callus content in an embodiment of the present invention. DETAILED DESCRIPTION

[0046] As shown in the figure, the method for predicting Huanglongbing by microspectral analysis of the callose content in the venous vessels of citrus leaves is characterized by: the method uses microspectral imaging of the callose content in the venous vessels of citrus leaves, detects changes in the callose through spectral imaging methods, analyzes whether the plant has a defense response to Huanglongbing infection, and predicts the Huanglongbing infection of the plant.

[0047] The method comprises the following steps:

[0048] Step S0, sample processing, specifically: after slicing the collected citrus leaf samples, microscopic hyperspectral imaging of the leaf vein duct part is collected to obtain spectral image data of the leaves.

[0049] Step S1, image preprocessing: preprocessing the acquired spectral image, including image denoising, spectral correction, and image scale adjustment, to obtain the spectral characteristics of the leaf vein duct area to facilitate subsequent analysis and identification;

[0050] Step S2, vascular region segmentation: in the microscopic hyperspectral image of the citrus leaf vein vessel, the vascular region is separated from the entire leaf image by image processing methods;

[0051] Step S3, callose content prediction;

[0052] Step S4: disease condition prediction and detection.

[0053] In the slicing method of step S0, the leaves are random leaves of citrus plants, and the leaf samples are collected from the 1 / 3 to 2 / 3 area of ​​the main vein of the leaves; the slicing equipment uses a slicer or a blade, and the slice thickness ranges from 30-50um; if some varieties or samples need to reduce the thickness, it must not be less than 10um.

[0054] The leaf spectral data acquisition method in step S0 is as follows: the leaf is placed in a dark box and spectral data is acquired using a spectrometer, with only full-band spectral fill light providing illumination. The optimal number of spectral data bands is 256, but should be at least greater than 32, and the band range is 300 to 1000 nm. The image data is in the form of an image, image stream, video, or video stream, with a microscopic magnification of 10-40x.

[0055] In step S1, image denoising is performed using a linear mean filter with a 3×3 neighborhood window. Spectral correction uses a full-band reflector with a reflectance of 0.8. Scaling is performed using bilinear interpolation for upscaling and averaging for downscaling. The optimal scale is 320×480, the maximum scale should be less than or equal to 1080×1080, and the minimum scale should be greater than or equal to 180×180. Scaling is evaluated based on the prominence of callose areas.

[0056] In step S2, the region segmentation process is performed for objects with different band counts. For image data with a scale of less than or equal to 320×480, the segmentation method is to identify the leaf vein region through image texture conversion, edge extraction, and threshold segmentation. Texture conversion uses the gray-level co-occurrence matrix (GLCM) method, which calculates the gray-level co-occurrence of pixel pairs at different orientations (0°, 45°, 90°, and 135°) and at different distances in the image to generate a gray-level co-occurrence matrix. Multiple texture feature parameters, such as contrast, correlation, energy, and entropy, are extracted to enhance the texture image of the vein region. Edges are detected by calculating the gray-level gradient within a pixel neighborhood using the Sobel and Canny operators. The gradient magnitude and direction are then calculated, and edge pixels are identified using non-maximum suppression and double-threshold detection. Segmentation is then achieved using maximum inter-class variance.

[0057] In step S2, the gray-level co-occurrence matrix (GLCM) is used to analyze the gray-level relationship of pixels in the image. It is based on the statistics of the gray-level values ​​of pixel pairs in the image and describes the texture characteristics of the image through statistical information. Its usage methods include: method a - pixel pair statistics, method b - constructing a gray-level co-occurrence matrix, and method c - calculating gray-level co-occurrence evidence results;

[0058] The statistical method of pixel pairs is that a pixel pair refers to a pair of pixels in an image at a specified distance and direction. Let the distance be d and the direction be r, which are parameters. For example, d = 1, r = 45; then, traverse the image to count pixel pairs and count the frequency of occurrence of the grayscale values ​​of these two pixels; and normalize the statistics according to the frequency of occurrence; constructing the grayscale co-occurrence matrix means that after the statistics of method a, the pixel pairs are arranged according to the level difference to form a matrix G, then G is the grayscale co-occurrence matrix, and at this time the matrix G = {g i，j} in the element g i，j It represents the frequency of occurrence of pixel pairs of levels i and j;

[0059] In method c, the results of the gray-level co-occurrence matrix include image contrast, correlation, energy, and entropy. The formula for image contrast is expressed as:

[0060]

[0061] Where L is the level, such as 16 or 32;

[0062] Correlation: Measures the linear correlation of pixel grayscale values ​​in an image. The calculation formula is

[0063]

[0064] In the formula, u is the mean and σ is the standard deviation. The greater the correlation, the more similar the local structures of the image textures are.

[0065] Energy: reflects the uniformity of the image grayscale distribution and the coarseness of the texture. The calculation formula is:

[0066] The larger the energy value, the more uniform the image grayscale distribution and the finer the texture; the smaller the energy value, the coarser the image texture.

[0067] Entropy: measures the richness of image information, and the calculation formula is

[0068]

[0069] The larger the entropy value, the more complex the grayscale distribution in the image and the richer the texture information; the smaller the entropy value, the simpler the image texture.

[0070] In step S2, the region segmentation process is implemented for objects with different numbers of bands. The segmentation method for image data with a scale greater than 320×480 is: segmentation is achieved through a convolutional neural network (CNN) model. First, the training data and test data are labeled. Then, a deep neural network (CNN) is established through multiple convolutional layers, pooling layers, and fully connected layers. The CNN is trained to automatically learn the characteristic patterns of the catheter area in the image to achieve region segmentation.

[0071] In step S3, the callose content prediction is achieved through three steps: spectral information extraction, characteristic band extraction, and callose content prediction, specifically:

[0072] Spectral information is extracted from the area after segmenting the catheter: the band curve is extracted using the regional average statistical method with error. The error is implemented using the standard deviation. When the variance is greater than a threshold (preferably 10%), a binary method is used to reduce the area to ensure the purity of the spectral information extracted.

[0073] Feature band extraction: PCA or SPA is used to implement it. The number of extracted features should preferably be 32 and should be greater than or equal to 5 to ensure that key information is not lost;

[0074] Prediction of callose content: Multiple linear regression (less than or equal to 10) or support vector machine regression (greater than 10) is used according to the number of features. The multiple linear regression is expressed as:

[0075] y=∑β i x i ,

[0076] where x i is the parameter, β i is a parameter corresponding to the characteristic band. Support vector machine regression is expressed as f(x)=w T x+b, where w is the parameter matrix, x corresponds to the characteristic band, and b is the offset.

[0077] In step S4, the disease detection method is as follows: based on the callose content, the prediction is made in combination with the month. The detection categories of the disease severity of Huanglongbing include: infection and health;

[0078] The formula is expressed as

[0079] in is the standard callose content, x is the month, k and b are parameters. When , it is infected, and less than , it is healthy.

[0080] In actual test applications, the accuracy of this example is greater than 93%.

Claims

1. A method for predicting Huanglongbing by analyzing the callose content in citrus leaf veins using microspectroscopy, characterized by: The method uses microscopic spectral imaging of callose in citrus leaf vein vessels, detects changes in callose through spectral imaging methods, analyzes whether the plant has a defense response to Huanglongbing infection, and predicts the plant's Huanglongbing infection.

2. The method for predicting Huanglongbing by analyzing the callose content in citrus leaf veins using microspectroscopy according to claim 1, wherein: The method comprises the following steps: Step S0, sample processing, specifically: after slicing the collected citrus leaf samples, microscopic hyperspectral imaging of the leaf vein duct part is collected to obtain spectral image data of the leaves. Step S1, image preprocessing: preprocessing the acquired spectral image, including image denoising, spectral correction, and image scale adjustment, to obtain the spectral characteristics of the leaf vein duct area to facilitate subsequent analysis and identification; Step S2, vascular region segmentation: in the microscopic hyperspectral image of the citrus leaf vein vessel, the vascular region is separated from the entire leaf image by image processing methods; Step S3, callose content prediction; Step S4: disease condition prediction and detection.

3. The method for predicting Huanglongbing by analyzing the callose content in citrus leaf veins using microspectroscopy according to claim 2, wherein: In the slicing method of step S0, the leaf is a random leaf of a citrus plant, and the leaf sample is collected from the 1 / 3 to 2 / 3 area of ​​the main vein of the leaf; The slicing equipment uses a microtome or blade, and the slice thickness range is 30-50um; if some varieties or samples need to be reduced in thickness, it must not be less than 10um.

4. The method for predicting Huanglongbing by analyzing the callose content in citrus leaf veins using microspectroscopy according to claim 2, wherein: The leaf spectral data acquisition method in step S0 is as follows: the leaf is placed in a dark box and spectral data is acquired using a spectrometer, with only full-band spectral fill light providing illumination. The spectral data must contain at least 32 bands, ranging from 300 to 1000 nm. The image data is in the form of an image, image stream, video, or video stream, with a magnification of 10-40x.

5. The method for predicting Huanglongbing by analyzing the callose content in citrus leaf veins using microspectroscopy according to claim 2, wherein: In step S1, image denoising is performed using a linear mean filter with a 3×3 neighborhood window. Spectral correction uses a full-band reflector with a reflectivity of 0.

8. The scale adjustment uses bilinear interpolation for enlargement and averaging for reduction. The maximum scale should be less than or equal to 1080×1080, and the minimum scale should be greater than or equal to 180×180. The scale evaluation standard is to highlight the callosal area.

6. The method for predicting Huanglongbing by analyzing the callose content in citrus leaf veins using microspectroscopy according to claim 2, wherein: In step S2, the region segmentation process is implemented for objects with different numbers of bands; the segmentation method for image data with a scale less than or equal to 320×480 is: to realize the leaf vein duct area through image texture conversion, edge extraction and threshold segmentation; wherein, the texture conversion uses the gray level co-occurrence matrix GLCM method, by calculating the gray level co-occurrence of pixel pairs in different directions and at different distances in the image, generating a gray level co-occurrence matrix, extracting multiple texture feature parameters such as contrast, correlation, energy, and entropy, and enhancing the texture image of the duct area texture; using the Sobel operator and the Canny operator to calculate the gray level gradient in the pixel neighborhood to detect the edge, then calculating the gradient amplitude and direction, and determining the edge pixels through non-maximum suppression and double threshold detection; and achieving segmentation through maximum inter-class variance.

7. The method for predicting Huanglongbing by analyzing the callose content in citrus leaf veins using microspectroscopy according to claim 6, wherein: In step S2, the gray-level co-occurrence matrix (GLCM) is used to analyze the gray-level relationship of pixels in the image. It is based on the statistics of the gray-level values ​​of pixel pairs in the image and describes the texture characteristics of the image through statistical information. Its usage methods include: method a - pixel pair statistics, method b - constructing a gray-level co-occurrence matrix, and method c - calculating gray-level co-occurrence evidence results; The pixel pair statistics method is as follows: a pixel pair refers to a pair of pixels in an image at a specified distance and direction; let the distance be d and the direction be r, which are parameters; then, the pixel pairs are counted by traversing the image, and the frequency of occurrence of the grayscale values ​​of these two pixels is counted; and the statistical results are normalized according to the frequency of occurrence; Constructing the gray-level co-occurrence matrix means arranging the pixel pairs according to the level difference to form a matrix G after the statistics of method a, then G is the gray-level co-occurrence matrix, at this time matrix G={g i，j } in the element g i，j It represents the frequency of occurrence of pixel pairs of levels i and j; In method c, the results of the gray-level co-occurrence matrix include image contrast, correlation, energy, and entropy. The formula for image contrast is expressed as: Where L is the level, such as 16 or 32; Correlation: Measures the linear correlation of pixel grayscale values ​​in an image. The calculation formula is In the formula, u is the mean and σ is the standard deviation. The greater the correlation, the more similar the local structures of the image textures are. Energy: reflects the uniformity of the image grayscale distribution and the coarseness of the texture. The calculation formula is The larger the energy value, the more uniform the image grayscale distribution and the finer the texture; the smaller the energy value, the coarser the image texture. Entropy: measures the richness of image information and is calculated as The larger the entropy value, the more complex the grayscale distribution in the image and the richer the texture information; the smaller the entropy value, the simpler the image texture.

8. The method for predicting Huanglongbing by analyzing the callose content in citrus leaf veins using microspectroscopy according to claim 2, wherein: In step S2, the region segmentation process is implemented for objects with different numbers of bands. The segmentation method for image data with a scale greater than 320×480 is: segmentation is achieved through a convolutional neural network (CNN) model. First, the training data and test data are labeled. Then, a deep neural network (CNN) is established through multiple convolutional layers, pooling layers, and fully connected layers. The CNN is trained to automatically learn the characteristic patterns of the catheter area in the image to achieve region segmentation.

9. The method for predicting Huanglongbing by analyzing the callose content in citrus leaf veins using microspectroscopy according to claim 2, wherein: In step S3, the callose content prediction is achieved through three steps: spectral information extraction, characteristic band extraction, and callose content prediction, specifically: Spectral information is extracted from the area after segmenting the catheter: the band curve is extracted using the regional average statistical method with error. The error is implemented using the standard deviation. When the variance is greater than the threshold, the binary method is used to reduce the area to ensure the purity of the spectral information extracted. Feature band extraction: PCA or SPA is used to implement it, and the number of extracted features is greater than or equal to 5 to ensure that key information is not lost; Prediction of callose content: Multiple linear regression or support vector machine regression is selected according to the number of feature numbers, where multiple linear regression is expressed as: and=∑β i x i , where x i is the parameter, β i is a parameter corresponding to the characteristic band. Support vector machine regression is expressed as f(x)=w T x+b, where w is the parameter matrix, x corresponds to the characteristic band, and b is the offset.

10. The method for predicting Huanglongbing by analyzing the callose content in citrus leaf veins using microspectroscopy according to claim 2, characterized in that: In step S4, the disease detection method is as follows: based on the callose content, the prediction is made in combination with the month. The detection categories of the disease severity of Huanglongbing include: infection and health; The formula is expressed as in is the standard callose content, x is the month, k and b are parameters. When , it is infected, and less than , it is healthy.

Images