Rapid detection system for Candidatus Liberibacter asiaticum based on infrared imaging

Abstract

The invention discloses a rapid detection system for citrus liberobacter asiaticum based on infrared imaging, which relates to the technical field of citrus disease detection and comprises a data acquisition module, an environment correction module, a leaf extraction module, a feature calculation module and an infection identification module. The data acquisition module is used for deploying infrared thermal imaging equipment at key observation points of citrus tree canopies and acquiring infrared radiation images and environment temperature and humidity; the environment correction module calculates a compensation coefficient through a temperature and humidity correction model, and corrects an image pixel gray value to eliminate environment interference; the blade extraction module extracts a refined blade thermal image; the feature calculation module calculates the pixel gray scale standard deviation of the local region of interest as a transpiration heat dissipation difference feature value; and the infection identification module compares the characteristic value with the thermal physiological baseline, marks a suspected infection site and generates an infection risk thermodynamic diagram. The system can realize rapid and accurate detection of the Candidatus Liberibacter asiaticum and assist disease prevention and control.

Description

Technical Field

[0001] This invention belongs to the field of citrus disease detection technology, specifically a rapid detection system for Huanglongbing fungus in citrus based on infrared imaging. Background Technology

[0002] Citrus Huanglongbing (HLB) is a highly devastating disease in the citrus industry, and rapid and accurate detection of pathogen infection is crucial for controlling its spread. Currently, infrared imaging-based methods for detecting HLB are widely used. These methods acquire infrared radiation images of the citrus canopy using infrared thermal imaging equipment. By observing temperature changes caused by abnormal transpiration after leaf infection, a preliminary assessment of the infection status can be made. Simultaneously, existing technologies record the ambient temperature and relative humidity at the time of image capture to minimize the impact of environmental factors on the detection results.

[0003] Existing infrared imaging detection technologies have limitations in practical applications. Fluctuations in ambient temperature and relative humidity significantly interfere with the temperature measurement accuracy of infrared radiation images. Current technologies do not specifically correct for environmental factors, resulting in the acquired infrared radiation images' grayscale values ​​failing to accurately reflect the actual thermal distribution of the leaves, thus affecting the accuracy of infection assessment. In leaf thermal image analysis, existing technologies often rely on overall grayscale features for infection identification, making it difficult to capture subtle differences in transpiration and heat dissipation caused by pathogens in specific areas of the leaf. This hinders the precise location of early suspected infection sites and fails to visually represent the spatial distribution of infection risk, impeding the accurate implementation of subsequent prevention and control measures. Summary of the Invention

[0004] This invention aims to solve at least one of the technical problems existing in the prior art;

[0005] Therefore, this invention proposes a rapid detection system for Huanglongbing (HLB) of citrus based on infrared imaging, comprising:

[0006] The data acquisition module deploys infrared thermal imaging equipment at key observation points in the canopy of the target citrus trees to collect infrared radiation images of the fruit trees under natural light conditions and record the ambient temperature and relative humidity at the time of shooting.

[0007] The environmental correction module inputs the ambient temperature and relative humidity into a pre-established temperature and humidity correction model, calculates the current infrared radiation compensation coefficient, and corrects the gray value of each pixel in the acquired infrared radiation image based on the infrared radiation compensation coefficient to eliminate the temperature measurement deviation caused by environmental interference and generate a corrected canopy thermal distribution matrix.

[0008] The leaf extraction module performs morphological gradient operations on the corrected canopy thermal distribution matrix to extract the leaf edge contours and remove pixel data from the trunk and thick branch areas, retaining only the refined thermal image containing leaf texture.

[0009] The feature calculation module delineates several non-overlapping local regions of interest in the refined thermal image and calculates the standard deviation of pixel grayscale in each local region of interest, which is used as the feature value of local transpiration heat dissipation difference of the blade.

[0010] The infection identification module compares the transpiration heat dissipation difference feature values ​​with the thermophysiological baseline data, identifies and marks suspected infection sites, integrates the coordinate information and feature values ​​of all suspected infection sites, and generates a heat map of citrus Huanglongbing infection risk.

[0011] Furthermore, the ambient temperature and relative humidity are input into a pre-established temperature and humidity correction model to calculate the current infrared radiation compensation coefficient, specifically including:

[0012] Retrieve the blackbody radiation reference table associated with the factory calibration parameters of the infrared thermal imaging device to obtain the theoretical radiation response curve under the current device gain setting;

[0013] Based on the ambient temperature and relative humidity, consult the lookup table for atmospheric transmittance and humidity attenuation coefficient to calculate the energy loss rate of infrared light in the air under the current meteorological conditions.

[0014] By convolving the theoretical radiation response curve with the energy loss rate, the additional radiation component caused by non-blade body factors under the current environment is derived.

[0015] Dividing the additional radiation component by the theoretical radiation intensity of the blackbody at the current ambient temperature yields a dimensionless offset correction factor.

[0016] The offset correction factor is multiplied by the reference compensation coefficient at the time of device manufacturing to output the final infrared radiation compensation coefficient.

[0017] Furthermore, the correction of the grayscale value of each pixel in the acquired infrared radiation image to eliminate temperature measurement deviations caused by environmental interference specifically includes:

[0018] Read the original metadata of the infrared radiation image and parse out the original grayscale value of each pixel and the integration time parameter at the time of shooting;

[0019] Using the infrared radiation compensation coefficient, a linear grayscale transformation equation is established to map the original grayscale value to the equivalent blackbody temperature value;

[0020] A cosine correction factor for the solar incidence angle at the time of shooting is introduced to perform secondary correction for uneven heating of the leaf surface caused by the angle of illumination.

[0021] The equivalent blackbody temperature values ​​after secondary correction are converted into a unified unit of Celsius and stored in matrix form to form the corrected canopy heat distribution matrix.

[0022] The corrected canopy heat distribution matrix is ​​subjected to median filtering to remove isolated noise pixels and smooth out small temperature fluctuations on the leaf surface.

[0023] Furthermore, morphological gradient operations are performed on the corrected canopy heat distribution matrix to extract the leaf edge contours, specifically including:

[0024] Construct a flat, disc-shaped structural element with a diameter slightly larger than the average width of a citrus leaf;

[0025] Using the disk-shaped structural element as a convolution kernel, erosion and dilation operations are sequentially performed on the corrected canopy heat distribution matrix to calculate the morphological gradient magnitude map of the matrix.

[0026] A high threshold is set for the morphological gradient magnitude map, and pixels with magnitudes greater than the high threshold are retained as strong edge candidate points;

[0027] The eight-neighbor connected domain analysis method is used to aggregate adjacent strong edge candidate points into continuous contour lines and fit closed polygon boundaries.

[0028] The pixel mask of the region inside the polygon boundary is applied to the corrected canopy thermal distribution matrix to shield the pixel data of the trunk and thick branches outside the polygon, while preserving the refined thermal image of the leaf texture.

[0029] Furthermore, the calculation of the standard deviation of pixel grayscale in each local region of interest, which is used as the feature value of local transpiration heat dissipation difference of the blade, specifically includes:

[0030] The refined thermal image is divided into several grid units with a preset pixel size on each side, and each grid unit is defined as a local region of interest.

[0031] Traverse each region of interest, count the grayscale values ​​of all pixels within the region, and calculate their arithmetic mean as the reference temperature value of the region of interest;

[0032] Calculate the square of the difference between the gray value of each pixel in the region and the reference temperature value, and sum all the squared differences;

[0033] The variance is obtained by dividing the sum of squares by the total number of pixels in the region minus one. The square root of the variance is then taken to obtain the gray standard deviation of the local region of interest.

[0034] The grayscale standard deviation is compared with a preset transpiration activity threshold. If the standard deviation is lower than the transpiration activity threshold, the local region of interest is determined to be in a state of stagnant heat dissipation caused by pore closure.

[0035] Furthermore, the differential transpiration heat dissipation characteristic values ​​are compared with the thermophysiological baseline data to identify and mark suspected infection sites. The coordinate information and characteristic values ​​of all suspected infection sites are integrated to generate a heat map of citrus Huanglongbing infection risk, including:

[0036] The transpiration heat dissipation difference characteristic values ​​were compared with the thermophysiological baseline data of citrus Huanglongbing fungus infection state to identify abnormal heat dissipation patterns that deviate from the baseline distribution and mark suspected infection sites.

[0037] By integrating the coordinate information of all suspected infection sites with the corresponding transpiration and heat dissipation difference feature values, a visualized heat map of citrus Huanglongbing infection risk is generated.

[0038] The differential transpiration heat dissipation characteristic values ​​are compared with the thermophysiological baseline data of citrus Huanglongbing fungus infection state to identify abnormal heat dissipation patterns that deviate from the baseline distribution, specifically including:

[0039] Load the probability density distribution model of the transpiration heat dissipation difference feature values ​​of healthy citrus plants in different seasons stored in the local database;

[0040] Substitute the transpiration heat dissipation difference feature value into the probability density distribution model to calculate the probability value of the transpiration heat dissipation difference feature value.

[0041] A discrimination threshold is set. When the calculated probability value is less than the discrimination threshold, the heat dissipation mode of the local region of interest is identified as a low-probability event.

[0042] Further search the frequency of occurrence of the heat dissipation pattern corresponding to the low probability event in the Huanglongbing infection case database. If the frequency exceeds a set proportion, it is classified as an abnormal heat dissipation pattern.

[0043] In the refined thermal image, the local region of interest belonging to the abnormal heat dissipation mode is marked as a high-risk area in red, and the remaining areas are marked as normal areas in green.

[0044] Furthermore, the process of integrating the coordinate information of all suspected infection sites with their corresponding transpiration and heat dissipation difference feature values ​​to generate a visualized heat map of citrus Huanglongbing infection risk specifically includes:

[0045] Create a blank canvas with the same resolution as the original infrared radiation image as the underlying carrier of the risk heat map;

[0046] Iterate through the coordinates of all suspected infection sites marked in red, and draw a semi-transparent circular mark at the corresponding position on the bottom canvas based on the magnitude of the corresponding transpiration heat dissipation difference characteristic value.

[0047] A linear mapping is made between the transparency of the circular markers and the characteristic value of the difference in evaporative heat dissipation. The larger the characteristic value, the darker the red of the marker and the higher the transparency.

[0048] Gaussian blur is applied to the underlying canvas after all the circular markers are superimposed, so that the color gradient transition between adjacent risk points is created, forming a continuous risk field.

[0049] Latitude and longitude grid lines and orchard zoning markers are superimposed on the continuous risk field to finally output a visualized heat map of citrus Huanglongbing infection risk containing spatial distribution information.

[0050] The linear mapping between the transparency of the circular markers and the difference in evaporative heat dissipation features specifically includes:

[0051] Obtain the maximum and minimum values ​​of the transpiration heat dissipation difference characteristic values ​​of all suspected infection sites in the current global scope;

[0052] Calculate the normalized relative strength of the transpiration heat dissipation difference characteristic value of the current suspected infection site relative to the maximum and minimum values;

[0053] Set the transparency mapping range to completely transparent to completely opaque, and multiply the normalized relative intensity by the maximum transparency coefficient to obtain the transparency value of the current mark;

[0054] At the same time, the normalized relative intensity is mapped onto a predefined chromatographic band, and the RGB color value at the corresponding position on the chromatographic band is selected as the fill color of the circle.

[0055] By applying the fill color and transparency value together to the circular markers, the differences in infection risk levels can be visually reflected in the color and shade on the risk heatmap.

[0056] Furthermore, the frequency of occurrence of the heat dissipation pattern corresponding to the low-probability event in the Huanglongbing infection case database is retrieved, specifically including:

[0057] Extract the spatial morphological features of the currently identified abnormal heat dissipation patterns, including the shape factor and dispersion index of the low-temperature patches;

[0058] The spatial morphological features are encoded into feature vectors and then Euclidean distance is used to measure the feature vectors of existing historical cases in the Huanglongbing infection case database.

[0059] Select the top few historical cases with the smallest Euclidean distance to form a set of nearest neighbor cases;

[0060] The number of positive samples in the nearest neighbor case set is counted and divided by the total number of samples to calculate the current confidence score.

[0061] The confidence score is appended to the coordinate data of the suspected infection site as one of the input parameters for risk assessment.

[0062] Furthermore, the method of employing eight-neighbor connected component analysis to aggregate adjacent strong edge candidate points into continuous contour segments specifically includes:

[0063] Create a Boolean access marker matrix with the same size as the morphological gradient magnitude map, and initialize all elements to an unvisited state;

[0064] Scan the morphological gradient magnitude map from left to right and from top to bottom. When an unvisited pixel with a positive gray value is encountered, initiate a connected component tracing.

[0065] During the tracking process, the neighboring pixels of the pixel in eight directions are recursively retrieved, and all neighboring pixels with positive gray values ​​are added to the current contour chain list.

[0066] After each recursive call, the pixel is marked as visited in the access mark matrix to prevent duplicate counting of different contours;

[0067] When the contour list is closed or there are no more new adjacent pixels, stop tracking and save the current contour list as an independent leaf edge contour.

[0068] Furthermore, the construction method of the pre-established temperature and humidity correction model includes:

[0069] In a standard experimental environment, a set of constant temperature and humidity test chambers covering the expected working temperature and humidity range are configured.

[0070] Under each preset temperature and humidity combination, an infrared thermal imaging device to be calibrated is used to acquire images of a set of blackbody targets with calibrated surface temperatures, and obtain multiple sets of infrared radiation grayscale value data under different environmental parameters.

[0071] From the infrared radiation grayscale data, extract the grayscale value of each pixel under a specific ambient temperature and relative humidity, and form a data pair with the corresponding blackbody target real temperature value.

[0072] Using ambient temperature and relative humidity as input features, and the deviation between the grayscale value and the theoretical grayscale value calculated based on the actual temperature value and the device response curve as the output target, a multivariate regression dataset is constructed.

[0073] The least squares method was used to fit the multivariate regression dataset, and a set of regression coefficients were obtained. The regression coefficients define the quantitative relationship between the combined effect of ambient temperature and relative humidity on the systematic shift of infrared radiation measurements.

[0074] The regression coefficients are encapsulated into a mathematical function, which takes the real-time collected ambient temperature and relative humidity as input and directly outputs the infrared radiation compensation coefficient for correcting the current measurement data, thus completing the construction of the temperature and humidity correction model.

[0075] Compared with the prior art, the beneficial effects of the present invention are:

[0076] The ambient temperature and relative humidity, collected in real time during the shooting process, are input into a pre-established temperature and humidity correction model to calculate the infrared radiation compensation coefficient. Based on this coefficient, the grayscale value of each pixel in the infrared radiation image is corrected, eliminating temperature measurement deviations caused by environmental interference and generating a corrected canopy thermal distribution matrix. This technology avoids the interference of ambient temperature and humidity fluctuations on infrared thermometry, ensuring that the corrected canopy thermal distribution matrix accurately reflects the actual thermophysiological state of the leaves. It avoids temperature measurement deviations caused by environmental factors, making infection identification based on thermal distribution more reliable and reducing misjudgments caused by environmental interference. Compared to conventional infrared detection technologies without environmental correction, it effectively improves detection stability.

[0077] In a refined thermal image containing only leaf texture, several non-overlapping regions of interest (ROIs) are delineated. The standard deviation of pixel grayscale within each ROI is calculated and used as a feature value for localized transpiration heat dissipation differences in the leaf. This value is then compared with baseline thermophysiological data to identify suspected infection sites. By integrating the coordinate information and feature values ​​of all suspected infection sites, a heat map of citrus Huanglongbing (HLB) infection risk is generated. This technology can accurately capture subtle differences in transpiration heat dissipation caused by pathogen infection in localized areas of the leaf, enabling the identification of early and mild suspected infection sites. Compared to conventional identification methods that use overall grayscale features, it effectively avoids missed and false detections. Furthermore, the infection risk heat map clearly presents the spatial distribution of suspected infection sites, facilitating rapid identification of potential infection areas and providing precise guidance for disease control. Attached Figure Description

[0078] Figure 1 This is a timing diagram of the rapid detection system for citrus Huanglongbing fungus based on infrared imaging as described in this invention;

[0079] Figure 2A flowchart for pixel correction in infrared radiation images;

[0080] Figure 3 Probability density analysis diagram of the identification stage of citrus Huanglongbing infection;

[0081] Figure 4 An edge detection analysis diagram for the citrus leaf extraction stage;

[0082] Figure 5 A graph showing the morphological gradient amplitude analysis during the extraction stage of citrus leaves. Detailed Implementation

[0083] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments.

[0084] See Figure 1 The workflow begins with deploying infrared thermal imaging equipment at key observation points in the canopy of the target citrus trees. This equipment acquires infrared radiation images of the trees under natural light conditions, simultaneously recording ambient temperature and relative humidity data at the time of capture. The acquired infrared radiation images and environmental parameters are then transmitted to the environmental correction module. This module inputs the ambient temperature and relative humidity into a pre-established temperature and humidity correction model to calculate the current infrared radiation compensation coefficient. Based on this coefficient, the grayscale value of each pixel in the acquired infrared radiation image is corrected to eliminate temperature measurement deviations caused by environmental interference, thereby generating a corrected canopy thermal distribution matrix. The leaf extraction module performs morphological gradient operations on the corrected canopy thermal distribution matrix to extract the leaf edge contours. Based on this, pixel data from the trunk and thick branches are removed, ultimately retaining a refined thermal image containing only leaf texture. The feature calculation module delineates several non-overlapping regions of interest (ROIs) in the obtained refined thermal image and calculates the standard deviation of pixel grayscale values ​​within each ROI. This standard deviation is defined as the characteristic value of local transpiration heat dissipation difference in the leaves. The infection identification module compares the calculated transpiration heat loss difference feature values ​​with the pre-stored thermophysiological baseline data, identifies and marks suspected infection sites, and integrates the coordinate information and feature values ​​of all suspected infection sites to generate a visualized heat map of citrus Huanglongbing infection risk, thus completing a rapid, non-contact assessment of plant health status.

[0085] In one embodiment of the present invention, the model is constructed in a standard experimental environment. This standard experimental environment is configured with a set of constant temperature and humidity test chambers covering the expected operating temperature and humidity range. For example, five sets of constant temperature and humidity test chambers are set, with a temperature range covering 15°C to 35°C and a humidity range covering 40% to 80% relative humidity. Under each preset temperature and humidity combination, such as a temperature of 20°C and a relative humidity of 60%, an infrared thermal imaging device to be calibrated is used to acquire images of a set of blackbody targets with calibrated surface temperatures. The surface temperatures of the blackbody targets are set to multiple known gradients, thereby obtaining multiple sets of infrared radiation grayscale value data under different environmental parameters. From the infrared radiation grayscale data, the grayscale value of each pixel under specific ambient temperature and relative humidity is extracted. Taking the grayscale value 126 of the center pixel of an image acquired from a blackbody target with a surface temperature of 25 degrees Celsius under an environment of 20 degrees Celsius and 60% relative humidity as an example, the grayscale value 126 and its corresponding actual blackbody target temperature of 25 degrees Celsius constitute a data pair. Using ambient temperature and relative humidity as input features, and the deviation between the grayscale value and the theoretical grayscale value calculated based on the actual temperature value and the device response curve as the output target, a multivariate regression dataset is constructed. The least squares method is used to fit the multivariate regression dataset to obtain a set of regression coefficients. The regression coefficients define the quantitative relationship of the systematic shift in infrared radiation measurement values ​​caused by the combined effect of ambient temperature and relative humidity. It can be understood that the regression coefficients are encapsulated as a mathematical function. The mathematical function takes the real-time acquired ambient temperature and relative humidity as input and directly outputs the infrared radiation compensation coefficients used to correct the current measurement data, thus completing the construction of the temperature and humidity correction model. The model function can be expressed as follows:

[0086] in: This is the infrared radiation compensation coefficient. For ambient temperature, Relative humidity, These are the regression coefficients obtained from the fitting.

[0087] In some embodiments, when the infrared thermal imaging device is deployed in an orchard for actual measurement, the calculation of the infrared radiation compensation coefficient is performed according to the temperature and humidity correction model and specific steps. In a specific implementation, when calculating the infrared radiation compensation coefficient, the blackbody radiation reference table associated with the factory calibration parameters of the infrared thermal imaging device is retrieved to obtain the theoretical radiation response curve under the current device gain setting. Based on the real-time collected ambient temperature and relative humidity, for example, an ambient temperature of 28 degrees Celsius and a relative humidity of 65%, the corresponding lookup table of atmospheric transmittance and humidity attenuation coefficient is consulted to calculate the energy loss rate of infrared light in the air under the current meteorological conditions. The theoretical radiation response curve is convolved with the energy loss rate to derive the additional radiation component caused by non-leaf body factors under the current environment. The additional radiation component is divided by the theoretical radiation intensity of the blackbody at the current ambient temperature to obtain a dimensionless offset correction factor. Optionally, the offset correction factor is finally multiplied by the reference compensation coefficient at the time of device factory calibration to output the final infrared radiation compensation coefficient used for image correction. For example, under the conditions of an ambient temperature of 28 degrees Celsius and a relative humidity of 65%, the calculated additional radiation component is... Watts per square meter per steradian, the theoretical radiation intensity of a blackbody at 28 degrees Celsius is Watts per square meter per steradian, then the offset correction factor is: If the equipment's baseline compensation coefficient is 1.02, then the final infrared radiation compensation coefficient is: .

[0088] In one embodiment of the present invention, see [reference] Figure 2 The system reads the original metadata of the infrared radiation image and parses the original grayscale value and integration time parameter of each pixel at the time of capture. For example, a pixel located at image coordinates (100, 150) has an original grayscale value of 85 and an integration time parameter of 8 milliseconds. Using the infrared radiation compensation coefficient obtained from the temperature and humidity correction model, a linear grayscale transformation equation is established to map the original grayscale value to the equivalent blackbody temperature value. The linear grayscale transformation equation is expressed as:

[0089] in: This represents the equivalent blackbody temperature. Represents the original grayscale value. and This is a mapping coefficient determined by the infrared radiation compensation coefficient and equipment parameters. A cosine correction factor for the solar incidence angle at the time of shooting is introduced to perform a secondary correction for uneven heating of the blade surface caused by the illumination angle. Assuming the solar incidence angle is 30 degrees, the cosine correction factor is... The equivalent blackbody temperature values, after secondary correction, are converted into a unified unit of Celsius and stored in matrix form to form the corrected canopy heat distribution matrix. Optionally, the corrected canopy heat distribution matrix is ​​subjected to median filtering, for example, using a 3x3 filter window, to remove isolated noise pixels and smooth out small temperature fluctuations on the leaf surface.

[0090] In some embodiments, the leaf extraction process performs morphological gradient operations on the corrected canopy heat distribution matrix. A flat, disk-shaped structuring element is constructed, with a diameter slightly larger than the average width of a citrus leaf, for example, a diameter of 15 pixels. Using this disk-shaped structuring element as a convolution kernel, erosion and dilation operations are sequentially performed on the corrected canopy heat distribution matrix to calculate the morphological gradient magnitude map. A high threshold is set for the morphological gradient magnitude map, for example, pixels with a gradient magnitude greater than 25 are retained as strong edge candidates. An eight-neighbor connected component analysis method is used to aggregate adjacent strong edge candidates into continuous contour segments. Specifically, the aggregation process involves creating a Boolean access marker matrix with the same size as the morphological gradient magnitude map and initializing all elements to an unvisited state. The morphological gradient magnitude map is scanned from left to right and from top to bottom. When an unvisited pixel with a positive grayscale value is encountered, a connected component tracing is initiated. During the tracking process, neighboring pixels in eight directions of the pixel are recursively retrieved, and all neighboring pixels with positive grayscale values ​​are added to the current contour list. After each recursive call, the pixel is marked as visited in the access marker matrix. Tracking stops when the contour list is closed or there are no more new neighboring pixels, and the current contour list is saved as an independent leaf edge contour. This can be understood as fitting closed polygonal boundaries to multiple independent leaf edge contours, and applying a pixel mask of the region inside the polygonal boundary to the corrected canopy thermal distribution matrix, thus masking the pixel data of the trunk and thick branches outside the polygon, thereby preserving a refined thermal image containing only leaf texture.

[0091] In one embodiment of the present invention, the refined thermal image is a thermal data matrix containing only the leaf texture, obtained after environmental correction and leaf extraction. In a specific implementation, the standard deviation of pixel grayscale values ​​within each local region of interest (ROI) is calculated as the characteristic value of local transpiration heat dissipation difference of the leaf. First, the refined thermal image is divided into several grid units with a preset pixel size, for example, 20 pixels. Each square grid unit with a side length of 20 pixels is defined as a ROI. Each ROI is traversed, and the grayscale values ​​of all pixels within the region are counted. For example, a ROI located in the upper left corner of the image contains 400 pixels, and its grayscale values, after normalization, range from 0.3 to 0.7. The arithmetic mean of these 400 grayscale values ​​is calculated as the reference temperature value of the ROI. Calculate the square of the difference between the gray value of each pixel within the region and the reference temperature value, sum all squared differences, divide the sum of squares by the total number of pixels in the region minus one to obtain the variance, and then take the square root of the variance to obtain the gray standard deviation of the region of interest. The calculation formula is as follows:

[0092] in: This represents the grayscale standard deviation of the local region of interest, i.e., the characteristic value of transpiration heat dissipation difference. This represents the total number of pixels within the local region of interest. Indicates the first grayscale value of each pixel. The reference temperature value represents the arithmetic mean of the grayscale values ​​of all pixels within the local region of interest. In a specific example, the average grayscale value of pixels in one local region of interest is 0.52, and the calculated standard deviation of grayscale is 0.15, while the average grayscale value of pixels in another region is 0.49, and the calculated standard deviation of grayscale is 0.05, indicating that the pixel grayscale value distribution is more concentrated in the latter region.

[0093] In some embodiments, the preset pixel size can be adjusted according to the actual image size of the leaf. For example, when the image resolution is high, the size of the region of interest can be set to 10x10 pixels to capture more subtle temperature differences. When the image resolution is low or to obtain a wider range of uniformity features, the size can be set to 30x30 pixels. It is understood that the calculated grayscale standard deviation is compared with a preset transpiration activity threshold. For example, the preset transpiration activity threshold is the grayscale standard deviation corresponding to 0.8 degrees Celsius. If the grayscale standard deviation calculated for a region of interest is 0.5, which is lower than the transpiration activity threshold of 0.8, then the region of interest is determined to be in a state of stagnant heat dissipation due to stomatal closure. Optionally, the transpiration activity threshold can be set based on the thermal image statistical characteristics of healthy leaves of different citrus varieties under standard light conditions.

[0094] In specific implementations, after traversing and calculating all local regions of interest (ROIs) in the refined thermal image, a series of transpiration heat dissipation difference feature values ​​corresponding to spatial locations are obtained. These feature values ​​constitute the basic data for subsequent infection identification. In some embodiments, the grayscale standard deviations calculated for different ROIs differ. For example, the grayscale standard deviation in the region near the leaf midrib may be 0.12, while the grayscale standard deviation in the ROI at the leaf margin may be 0.25. This spatial difference itself reflects the non-uniformity of transpiration heat dissipation on the leaf surface. Optionally, for ROIs located at the leaf edge or at the junction with the background, some pixels may belong to non-leaf areas. Pixel selection can be performed based on the leaf mask before calculating the grayscale standard deviation, using only pixels identified as leaf tissue for calculation. It is understood that the magnitude of the transpiration heat dissipation difference feature values ​​is directly related to the physiological activity level of the local leaf tissue. A lower grayscale standard deviation means that the pixel temperature in that area is more uniform, which may indicate a decrease in stomatal conductance and a weakening of transpiration.

[0095] See Figure 3This is a probability density analysis chart of the citrus Huanglongbing (HLB) infection identification stage, used to compare the distribution of transpiration heat dissipation differences between healthy and suspected infected leaves. The baseline distribution of healthy leaves represents the typical distribution pattern of transpiration heat dissipation differences in healthy leaves, with a mean of approximately 0.18 and a concentrated distribution. The distribution of measured values ​​reflects the characteristics of the actual collected leaf thermal data, with a mean of approximately 0.14, shifted to the left overall and a more gradual distribution, indicating the existence of many low standard deviation areas. The abnormal / suspected infection areas deviate from the healthy baseline in terms of transpiration heat dissipation differences, representing low-probability events, and are marked as suspected infection sites. The discrimination threshold is the area below which is judged as a state of stagnant heat dissipation due to stomatal closure, a typical thermophysiological characteristic of HLB infection. The overall leftward shift in the measured value distribution indicates that the gray-level standard deviation of many local areas is lower than the healthy baseline, corresponding to weakened transpiration heat dissipation. The red-filled areas are the core suspected infection areas, where the heat dissipation patterns differ significantly from those of healthy leaves.

[0096] In one embodiment of the present invention, the infection identification module loads a probability density distribution model of the transpiration heat dissipation difference feature values ​​of healthy citrus plants under different seasons, stored in a local database. This probability density distribution model describes the statistical distribution characteristics of the transpiration heat dissipation difference feature values ​​of healthy leaves. The calculated transpiration heat dissipation difference feature values ​​are substituted into the probability density distribution model to calculate the probability value of the occurrence of each feature value. For example, if the transpiration heat dissipation difference feature value of a local area of ​​interest is 0.10, the calculated probability value after substituting it into the spring health model is 0.02. A discrimination threshold is set. When the calculated probability value is less than the discrimination threshold, the heat dissipation pattern of the local area of ​​interest is considered a low-probability event. For example, if the discrimination threshold is set to 0.05, then the probability value 0.02 is less than 0.05, and the region is determined to be a low-probability event.

[0097] In some embodiments, the frequency of occurrence of the heat dissipation pattern corresponding to the low-probability event in the Huanglongbing (HLB) infection case database is further retrieved. Spatial morphological features of the currently identified abnormal heat dissipation pattern are extracted, including the shape factor and dispersion index of the low-temperature patch. The shape factor describes the proximity of the patch shape to a circle, and the dispersion index describes the temperature dispersion of pixels within the patch. The spatial morphological features are encoded into feature vectors, and Euclidean distance is measured between them and the feature vectors of existing historical cases in the HLB infection case database. The top few historical cases with the smallest Euclidean distance are selected to form a nearest neighbor case set. The number of samples marked as positive in the nearest neighbor case set is counted, and divided by the total number of samples to calculate the current confidence score. The confidence score calculation formula is:

[0098] in: Indicates the confidence score. This represents the number of samples marked as positive in the nearest neighbor case set. This represents the total number of samples in the nearest neighbor case set. The confidence score is appended to the coordinate data of the suspected infection site as one of the input parameters for risk assessment. It is understood that if the frequency exceeds a set proportion, for example, if the proportion of positive samples in the nearest neighbor case set exceeds 70%, it is classified as an abnormal heat dissipation pattern. In the refined thermal image, the local region of interest belonging to the abnormal heat dissipation pattern is marked as a red high-risk area, and the remaining areas are marked as green normal areas. Optionally, refer to Table 1 for the confidence score and nearest neighbor cases.

[0099] Table 1: Example Table of Nearest Neighbor Case Set and Confidence Score Calculation

[0100]

[0101] In practical implementation, the coordinate information of all suspected infection sites and their corresponding transpiration heat dissipation difference feature values ​​are integrated to generate a visualized heat map of citrus Huanglongbing (HLB) infection risk. In practical implementation, the infection identification module outputs a list containing the coordinates of high-risk areas, their corresponding transpiration heat dissipation difference feature values, and confidence scores. In some embodiments, a local region of interest located at image coordinates (x=155, y=230) with a transpiration heat dissipation difference feature value of 0.08 is classified as an abnormal heat dissipation pattern, with a confidence score of 0.75, and this coordinate point is recorded as a suspected infection site. Optionally, the system iterates through all recorded suspected infection sites, transmitting their coordinates and feature information to the risk heat map generation module. It is understood that the probability density distribution model, discrimination threshold, number of nearest neighbor cases K, and frequency setting ratio can all be parameterized according to the specific circumstances of different citrus varieties or planting areas.

[0102] See Figure 4 This is an edge detection analysis diagram from the citrus leaf extraction stage. It primarily showcases the comparison between the grayscale values ​​of the original infrared thermal image and the morphological edge extraction results, providing a crucial visualization for separating leaf texture from a complex background. The original thermal image reflects the natural temperature distribution of the citrus canopy leaves. Peak areas represent the main leaf body, with grayscale values ​​around 50-60, while valley areas represent leaf texture details or areas of uneven heating. The overall grayscale distribution reflects the spatial differences in transpiration heat dissipation on the leaf surface. The edge extraction results accurately locate the leaf edges. The red vertical line segments correspond to the pixel coordinates of the leaf edges, representing strong edge candidate points extracted by morphological gradient operations. The horizontal line segments serve as mask markers for the internal areas of the leaf, used to subsequently mask the tree trunk / branch background. The curves clearly mark the physical boundaries of the leaves, achieving the core objective of separating leaf tissue from the background.

[0103] In one embodiment of the present invention, a blank canvas with the same resolution as the original infrared radiation image is established as the underlying carrier of the risk heat map. For example, if the original infrared image resolution is 640 pixels wide and 480 pixels high, then the size of the established blank canvas is also 640x480 pixels. The coordinates of all suspected infection sites marked in red are traversed. Based on the magnitude of the corresponding transpiration heat dissipation difference feature value, a semi-transparent circular marker is drawn at the corresponding position on the underlying canvas. The radius of the circular marker can be set to a fixed value, such as 5 pixels. The transparency mapping process obtains the maximum and minimum values ​​of the transpiration heat dissipation difference feature values ​​of all suspected infection sites in the current global range. It is assumed that the maximum value is 0.85 and the minimum value is 0.10. The normalized relative intensity of the transpiration heat dissipation difference feature value of the current suspected infection site relative to the maximum and minimum values ​​is calculated. The formula for calculating the normalized relative intensity is: ;

[0104] in: Indicates the normalized relative intensity. This represents the characteristic value of the difference in transpiration heat dissipation at the current suspected infection site. This represents the global maximum value of the feature values ​​for all suspected infection sites. This represents the global minimum value of all suspected infection site feature values. The transparency mapping range is set from completely transparent to completely opaque. The normalized relative intensity is multiplied by the maximum transparency coefficient to obtain the transparency value of the current marker. This can be understood as simultaneously mapping the normalized relative intensity to a predefined color band, selecting the RGB color value at the corresponding position on the color band as the fill color of the circle. For example, using a gradient from light yellow to dark red, a normalized relative intensity of 0.0 corresponds to light yellow (RGB:255,255,200), and a normalized relative intensity of 1.0 corresponds to dark red (RGB:200,0,0). The fill color and transparency value are applied together to the circular marker, visually reflecting the differences in infection risk levels on the risk heatmap through color and depth. The larger the feature value, the deeper the red of the marker, and the higher the transparency.

[0105] In some embodiments, the underlying canvas after overlaying all circular markers is Gaussian blurred to create a color gradient transition between adjacent risk points, forming a continuous risk field. The radius of the Gaussian blur convolution kernel can be set to 1.5 times the marker radius. Latitude and longitude grid lines and orchard zoning markers are then overlaid on this continuous risk field, ultimately outputting a visualized heatmap of citrus Huanglongbing infection risk containing spatial distribution information. The orchard zoning markers may include pre-entered row and column numbers of fruit trees. Optionally, when traversing and drawing markers, for a suspected infection site with coordinates (300, 200) and a transpiration heat dissipation difference characteristic value of 0.60, its normalized relative intensity is calculated as follows: If the maximum transparency factor is set to 0.8, then its transparency value is... Simultaneously, the color value corresponding to a relative intensity of 0.667 is obtained from the chromatographic bands as the fill color for the circle. In specific implementations, the radius of the circular marker can also be variable, positively correlated with the magnitude of the feature value, but the mapping logic between transparency and color remains unchanged. It can be understood that areas with larger feature values ​​will appear as a deeper red and with higher opacity on the heatmap, while areas with smaller feature values ​​will appear as a lighter color and with higher transparency, thus forming an intuitive visualization effect of risk gradient.

[0106] See Figure 5 This is a morphological gradient magnitude analysis chart from the citrus leaf extraction stage, used to demonstrate the edge detection effect after morphological gradient calculation. The morphological gradient magnitude reflects the temperature difference between pixels. High-magnitude regions (>0.5) correspond to the leaf edge contour and are strong edge candidate points; low-magnitude regions (<0.5) correspond to the interior of the leaf or background areas, where the temperature distribution is more uniform. A strong edge filtering threshold is used to filter strong edge candidate points, retaining only pixels with magnitudes greater than this threshold as the basis for subsequent contour aggregation. The leaf edge region visually marks the physical boundary of the leaf, corresponding to the input region of the "eight-neighbor connected domain analysis" in the embodiment, used for subsequent fitting of closed polygon boundaries. High-magnitude regions are concentrated in the pixel x-coordinate range of 35-75, clearly outlining the left and right edges of the leaf, consistent with the logic of "extracting leaf edges using disk-shaped structural elements" in the embodiment. Gradient magnitudes in the 0-35 and 75-160 ranges are both below the threshold, indicating that the background area has been effectively masked, retaining only the leaf texture.

Claims

1. A rapid detection system for Huanglongbing (HLB) of citrus based on infrared imaging, characterized in that, include: The data acquisition module deploys infrared thermal imaging equipment at key observation points in the canopy of the target citrus trees to collect infrared radiation images of the fruit trees under natural light conditions and record the ambient temperature and relative humidity at the time of shooting. The environmental correction module inputs the ambient temperature and relative humidity into a pre-established temperature and humidity correction model, calculates the current infrared radiation compensation coefficient, and corrects the gray value of each pixel in the acquired infrared radiation image based on the infrared radiation compensation coefficient to eliminate the temperature measurement deviation caused by environmental interference and generate a corrected canopy thermal distribution matrix. The leaf extraction module performs morphological gradient operations on the corrected canopy thermal distribution matrix to extract the leaf edge contours and remove pixel data from the trunk and thick branch areas, retaining only the refined thermal image containing leaf texture. The feature calculation module delineates several non-overlapping local regions of interest in the refined thermal image and calculates the standard deviation of pixel grayscale in each local region of interest, which is used as the feature value of local transpiration heat dissipation difference of the blade. The infection identification module compares the transpiration heat dissipation difference feature values ​​with the thermophysiological baseline data, identifies and marks suspected infection sites, integrates the coordinate information and feature values ​​of all suspected infection sites, and generates a heat map of citrus Huanglongbing infection risk.

2. The rapid detection system for citrus Huanglongbing (HLB) based on infrared imaging according to claim 1, characterized in that, The ambient temperature and relative humidity are input into a pre-established temperature and humidity correction model to calculate the current infrared radiation compensation coefficient, specifically including: Retrieve the blackbody radiation reference table associated with the factory calibration parameters of the infrared thermal imaging device to obtain the theoretical radiation response curve under the current device gain setting; Based on the ambient temperature and relative humidity, consult the lookup table for atmospheric transmittance and humidity attenuation coefficient to calculate the energy loss rate of infrared light in the air under the current meteorological conditions. By convolving the theoretical radiation response curve with the energy loss rate, the additional radiation component caused by non-blade body factors under the current environment is derived. Dividing the additional radiation component by the theoretical radiation intensity of the blackbody at the current ambient temperature yields a dimensionless offset correction factor. The offset correction factor is multiplied by the reference compensation coefficient at the time of device manufacturing to output the final infrared radiation compensation coefficient.

3. The rapid detection system for citrus Huanglongbing (HLB) based on infrared imaging according to claim 2, characterized in that, The process of correcting the grayscale value of each pixel in the acquired infrared radiation image to eliminate temperature measurement deviations caused by environmental interference specifically includes: Read the original metadata of the infrared radiation image and parse out the original grayscale value of each pixel and the integration time parameter at the time of shooting; Using the infrared radiation compensation coefficient, a linear grayscale transformation equation is established to map the original grayscale value to the equivalent blackbody temperature value; A cosine correction factor for the solar incidence angle at the time of shooting is introduced to perform secondary correction for uneven heating of the leaf surface caused by the angle of illumination. The equivalent blackbody temperature values ​​after secondary correction are converted into a unified unit of Celsius and stored in matrix form to form the corrected canopy heat distribution matrix. The corrected canopy heat distribution matrix is ​​subjected to median filtering to remove isolated noise pixels and smooth out small temperature fluctuations on the leaf surface.

4. The rapid detection system for citrus Huanglongbing (HLB) based on infrared imaging according to claim 3, characterized in that, Perform morphological gradient operations on the corrected canopy heat distribution matrix to extract the leaf edge contours, specifically including: Construct a flat, disc-shaped structural element with a diameter slightly larger than the average width of a citrus leaf; Using the disk-shaped structural element as a convolution kernel, erosion and dilation operations are sequentially performed on the corrected canopy heat distribution matrix to calculate the morphological gradient magnitude map of the matrix. A high threshold is set for the morphological gradient magnitude map, and pixels with magnitudes greater than the high threshold are retained as strong edge candidate points; The eight-neighbor connected domain analysis method is used to aggregate adjacent strong edge candidate points into continuous contour lines and fit closed polygon boundaries. The pixel mask of the region inside the polygon boundary is applied to the corrected canopy thermal distribution matrix to shield the pixel data of the trunk and thick branches outside the polygon, while preserving the refined thermal image of the leaf texture.

5. The rapid detection system for citrus Huanglongbing pathogen based on infrared imaging according to claim 4, characterized in that, The calculation of the standard deviation of pixel grayscale in each local region of interest, which is used as the feature value of local transpiration heat dissipation difference of the blade, specifically includes: The refined thermal image is divided into several grid units with a preset pixel size on each side, and each grid unit is defined as a local region of interest. Traverse each region of interest, count the grayscale values ​​of all pixels within the region, and calculate their arithmetic mean as the reference temperature value of the region of interest; Calculate the square of the difference between the gray value of each pixel in the region and the reference temperature value, and sum all the squared differences; The variance is obtained by dividing the sum of squares by the total number of pixels in the region minus one. The square root of the variance is then taken to obtain the gray standard deviation of the local region of interest. The grayscale standard deviation is compared with a preset transpiration activity threshold. If the standard deviation is lower than the transpiration activity threshold, the local region of interest is determined to be in a state of stagnant heat dissipation caused by pore closure.

6. The rapid detection system for citrus Huanglongbing (HLB) based on infrared imaging according to claim 5, characterized in that, The differential transpiration heat dissipation characteristic values ​​are compared with the thermophysiological baseline data to identify and mark suspected infection sites. The coordinate information and characteristic values ​​of all suspected infection sites are integrated to generate a heat map of citrus Huanglongbing infection risk, including: The transpiration heat dissipation difference characteristic values ​​were compared with the thermophysiological baseline data of citrus Huanglongbing fungus infection state to identify abnormal heat dissipation patterns that deviate from the baseline distribution and mark suspected infection sites. By integrating the coordinate information of all suspected infection sites with the corresponding transpiration and heat dissipation difference feature values, a visualized heat map of citrus Huanglongbing infection risk is generated. The differential transpiration heat dissipation characteristic values ​​are compared with the thermophysiological baseline data of citrus Huanglongbing fungus infection state to identify abnormal heat dissipation patterns that deviate from the baseline distribution, specifically including: Load the probability density distribution model of the transpiration heat dissipation difference feature values ​​of healthy citrus plants in different seasons stored in the local database; Substitute the transpiration heat dissipation difference feature value into the probability density distribution model to calculate the probability value of the transpiration heat dissipation difference feature value. A discrimination threshold is set. When the calculated probability value is less than the discrimination threshold, the heat dissipation mode of the local region of interest is identified as a low-probability event. Further search the frequency of occurrence of the heat dissipation pattern corresponding to the low probability event in the Huanglongbing infection case database. If the frequency exceeds a set proportion, it is classified as an abnormal heat dissipation pattern. In the refined thermal image, the local region of interest belonging to the abnormal heat dissipation mode is marked as a high-risk area in red, and the remaining areas are marked as normal areas in green.

7. The rapid detection system for citrus Huanglongbing pathogen based on infrared imaging according to claim 6, characterized in that, The process integrates the coordinate information of all suspected infection sites with their corresponding transpiration and heat dissipation difference feature values ​​to generate a visualized heat map of citrus Huanglongbing infection risk, specifically including: Create a blank canvas with the same resolution as the original infrared radiation image as the underlying carrier of the risk heat map; Iterate through the coordinates of all suspected infection sites marked in red, and draw a semi-transparent circular mark at the corresponding position on the bottom canvas based on the magnitude of the corresponding transpiration heat dissipation difference characteristic value. A linear mapping is made between the transparency of the circular markers and the characteristic value of the difference in evaporative heat dissipation. The larger the characteristic value, the darker the red of the marker and the higher the transparency. Gaussian blur is applied to the underlying canvas after all the circular markers are superimposed, so that the color gradient transition between adjacent risk points is created, forming a continuous risk field. Latitude and longitude grid lines and orchard zoning markers are superimposed on the continuous risk field to finally output a visualized heat map of citrus Huanglongbing infection risk containing spatial distribution information. The linear mapping between the transparency of the circular markers and the difference in evaporative heat dissipation features specifically includes: Obtain the maximum and minimum values ​​of the transpiration heat dissipation difference characteristic values ​​of all suspected infection sites in the current global scope; Calculate the normalized relative strength of the transpiration heat dissipation difference characteristic value of the current suspected infection site relative to the maximum and minimum values; Set the transparency mapping range to completely transparent to completely opaque, and multiply the normalized relative intensity by the maximum transparency coefficient to obtain the transparency value of the current mark; At the same time, the normalized relative intensity is mapped onto a predefined chromatographic band, and the RGB color value at the corresponding position on the chromatographic band is selected as the fill color of the circle. By applying the fill color and transparency value together to the circular markers, the differences in infection risk levels can be visually reflected in the color and shade on the risk heatmap.

8. The rapid detection system for citrus Huanglongbing (HLB) based on infrared imaging according to claim 7, characterized in that, The frequency of occurrence of the heat loss patterns corresponding to the low-probability events in the Huanglongbing (HLB) infection case database was retrieved, specifically including: Extract the spatial morphological features of the currently identified abnormal heat dissipation patterns, including the shape factor and dispersion index of the low-temperature patches; The spatial morphological features are encoded into feature vectors and then Euclidean distance is used to measure the feature vectors of existing historical cases in the Huanglongbing infection case database. Select the top few historical cases with the smallest Euclidean distance to form a set of nearest neighbor cases; The number of positive samples in the nearest neighbor case set is counted and divided by the total number of samples to calculate the current confidence score. The confidence score is appended to the coordinate data of the suspected infection site as one of the input parameters for risk assessment.

9. The rapid detection system for citrus Huanglongbing pathogen based on infrared imaging according to claim 8, characterized in that, The method employing eight-neighbor connected component analysis aggregates adjacent strong edge candidate points into continuous contour segments, specifically including: Create a Boolean access marker matrix with the same size as the morphological gradient magnitude map, and initialize all elements to an unvisited state; Scan the morphological gradient magnitude map from left to right and from top to bottom. When an unvisited pixel with a positive gray value is encountered, initiate a connected component tracing. During the tracking process, the neighboring pixels of the pixel in eight directions are recursively retrieved, and all neighboring pixels with positive gray values ​​are added to the current contour chain list. After each recursive call, the pixel is marked as visited in the access mark matrix to prevent duplicate counting of different contours; When the contour list is closed or there are no more new adjacent pixels, stop tracking and save the current contour list as an independent leaf edge contour.

10. The rapid detection system for citrus Huanglongbing pathogen based on infrared imaging according to claim 9, characterized in that, The construction method of the pre-established temperature and humidity correction model includes: In a standard experimental environment, a set of constant temperature and humidity test chambers covering the expected working temperature and humidity range are configured. Under each preset temperature and humidity combination, an infrared thermal imaging device to be calibrated is used to acquire images of a set of blackbody targets with calibrated surface temperatures, and obtain multiple sets of infrared radiation grayscale value data under different environmental parameters. From the infrared radiation grayscale data, extract the grayscale value of each pixel under a specific ambient temperature and relative humidity, and form a data pair with the corresponding blackbody target real temperature value. Using ambient temperature and relative humidity as input features, and the deviation between the grayscale value and the theoretical grayscale value calculated based on the actual temperature value and the device response curve as the output target, a multivariate regression dataset is constructed. The least squares method was used to fit the multivariate regression dataset, and a set of regression coefficients were obtained. The regression coefficients define the quantitative relationship between the combined effect of ambient temperature and relative humidity on the systematic shift of infrared radiation measurements. The regression coefficients are encapsulated into a mathematical function, which takes the real-time collected ambient temperature and relative humidity as input and directly outputs the infrared radiation compensation coefficient for correcting the current measurement data, thus completing the construction of the temperature and humidity correction model.

Images