Artemisia argyi growth state and leafhopper infringement dynamic correlation monitoring method

Abstract

The invention relates to the technical field of growth state monitoring, in particular to an artemisia argyi growth state and leafhopper infringement dynamic correlation monitoring method, which comprises the following steps of: shooting artemisia argyi regularly by using a multispectral camera, and collecting reflection spectrum information of near-infrared, red light and green light wavebands to obtain preliminary physiological parameter data; according to the preliminary physiological parameter data, the health condition and growth trend of the mugwort are evaluated, and a health evaluation result is generated. Reflection spectrum information of near-infrared, red light and green light wave bands is collected through the multispectral camera, plant physiological parameter data is generated, and multi-dimensional quantitative evaluation of the growth state of mugwort is achieved. The physiological parameters are trained and analyzed through deep learning, the lesion type and leafhopper invasion degree of the plant leaves can be recognized, manual judgment is not needed any more, and therefore the accuracy and stability of detection are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of growth state monitoring, and particularly to a method for dynamically monitoring the correlation between the growth state of Artemisia argyi and the infestation of leafhoppers. Background Art

[0002] The technical field of growth state monitoring mainly focuses on the changes in physiological and ecological parameters of plants at different growth stages. By real-time and continuous data collection and analysis, the growth state, health status and environmental adaptability of plants are evaluated. However, the existing technologies mainly rely on single physiological parameters or spectral data in limited bands in the monitoring of plant growth state, resulting in one-sidedness in the evaluation of plant health status. At the same time, due to the lack of multi-dimensional spectral reflection information, it is difficult to comprehensively reflect the physiological changes of plants at different growth stages, and early health abnormalities are easily missed. Therefore, improvements are needed. Summary of the Invention

[0003] The purpose of the present invention is to solve the deficiencies existing in the prior art, and to propose a method for dynamically monitoring the correlation between the growth state of Artemisia argyi and the infestation of leafhoppers.

[0004] To achieve the above purpose, the present invention adopts the following technical solutions. The method for dynamically monitoring the correlation between the growth state of Artemisia argyi and the infestation of leafhoppers includes the following steps:

[0005] Regularly photograph Artemisia argyi using a multispectral camera, collect the reflected spectral information in the near-infrared, red and green bands, and obtain preliminary physiological parameter data; according to the preliminary physiological parameter data, evaluate the health status and growth trend of Artemisia argyi, and generate a health assessment result;

[0006] Based on the preliminary physiological parameter data, train and analyze through a deep learning algorithm, detect whether there are lesions or infestations by leafhoppers on the plant leaves, determine the type of lesions and the degree of infestation, and generate a lesion infestation diagnosis result; based on the lesion infestation diagnosis result, mark the damaged area, calculate the proportion of the damaged area, and generate an analysis result of the damaged area;

[0007] Based on the analysis result of the damaged area and the health assessment result, formulate a treatment and intervention plan according to the degree of damage and the leaf health index, and generate an intervention plan formulation result;

[0008] According to the intervention plan formulation result, set a sliding time window to capture the short-term fluctuations and long-term trends in the growth process of Artemisia argyi, and generate a multi-scale growth analysis result.

[0009] Preferably, the step of obtaining the preliminary physiological parameter data is as follows:

[0010] Regularly photograph Artemisia argyi, collect the reflected spectral information in the near-infrared, red and green bands, and obtain a spectral image data set;

[0011] Extract spectral intensities, peak and trough points from the spectral image dataset, perform signal enhancement and background noise removal on the spectral intensities, peak and trough points, and obtain a cleaned spectral feature dataset;

[0012] Based on the cleaned spectral feature dataset, apply principal component analysis to identify patterns and trends in the spectral data and obtain preliminary physiological parameter data.

[0013] Preferably, the steps for obtaining the health assessment result are as follows:

[0014] Analyze the preliminary physiological parameter data, identify chlorophyll content, water status, and growth rate, and construct a physiological index dataset;

[0015] Based on the physiological index dataset, calculate the health score of Artemisia argyi, and the calculation formula is:

[0016]

[0017] where L i represents the i-th chlorophyll content value, W i represents the i-th water status value, G i represents the i-th growth rate, n is the number of samples, and HC is the health score;

[0018] Based on the health score, combine the growth data and temperature and light conditions, perform trend analysis, and obtain the health assessment result.

[0019] Preferably, the steps for obtaining the lesion invasion diagnosis result are as follows:

[0020] Based on the preliminary physiological parameter data, configure and train a deep learning model, optimize the recognition accuracy by identifying lesion features and leafhopper invasion signs in the leaf image using a convolutional neural network, and obtain a trained model;

[0021] Use the trained model to analyze newly collected leaf data, identify and classify the types and degrees of leaf lesions by parsing the image data, and generate a lesion invasion diagnosis result.

[0022] Preferably, the steps for obtaining the damaged area analysis result are as follows:

[0023] Based on the lesion invasion diagnosis result, extract the pixel distribution information of the damaged area, identify the boundary of the lesion area on the leaf surface, remove noise and improve the contrast of the damaged area, and generate a binary image of the damaged area;

[0024] According to the binary image of the damaged area, calculate the damaged area ratio, and the calculation formula is:

[0025]

[0026] Among them, X j , Y j are the coordinates of the j-th damaged pixel, X c , Y c are the coordinates of the center of the damaged area, m is the total number of damaged pixels, A is the total leaf area, and R is the damaged area ratio;

[0027] Based on the damaged area ratio, combined with the geometric morphological characteristics of the damaged area, analyze the diffusion trend of the lesion, classify the lesion type, and generate the analysis result of the damaged area.

[0028] Preferably, the steps for obtaining the intervention plan formulation result are as follows:

[0029] According to the damaged area analysis result and the health assessment result, calculate the intervention priority score, and the calculation formula is:

[0030]

[0031] Among them, R represents the damaged area ratio, H represents the health score, E represents the lesion diffusion rate, C represents the complexity of the boundary of the damaged area, B represents the complexity of the boundary of the healthy area, SG represents the leaf growth rate, F represents the leaf moisture content, and P is the intervention priority score;

[0032] Based on the intervention priority score, combined with the lesion type and the growth environment conditions, screen the matching treatment and intervention measures to generate the intervention plan formulation result.

[0033] Preferably, the steps for obtaining the multi-scale growth analysis result are as follows:

[0034] According to the intervention plan formulation result, set a sliding time window, segment the growth index data during the Artemisia argyi growth process at fixed time intervals, and extract the growth characteristics including the leaf area growth rate, the stem height increment, and the chlorophyll concentration change rate to generate a time window growth feature matrix;

[0035] Based on the time window growth feature matrix, calculate the growth fluctuation intensity, and the expression is:

[0036]

[0037] Among them, I represents the growth fluctuation intensity, XA j is the leaf area of the j-th time window, XH j is the stem height of the j-th time window, XC j is the chlorophyll concentration of the j-th time window, and xm is the number of time windows.

[0038] Preferably, the step of obtaining the multi-scale growth analysis result further includes: based on the growth fluctuation intensity, applying singular spectrum analysis to generate the multi-scale growth analysis result.

[0039] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0040] The present invention collects the reflected spectral information in the near-infrared, red, and green bands through a multi-spectral camera to generate plant physiological parameter data, realizing the multi-dimensional quantitative evaluation of the growth state of Artemisia argyi. Through deep learning training and analysis of these physiological parameters, it can identify the types of leaf lesions and the degree of leafhopper infestation of plant leaves, no longer relying on manual judgment, thereby improving the accuracy and stability of detection. The marking of the lesion area and the quantitative analysis of the damaged area ratio enable the impact of the infestation to be presented in the form of data, providing a clear basis for formulating prevention and control plans and avoiding deviations caused by empirical judgments. The generation of the intervention plan is based on the health assessment result and the analysis result of the damaged area, making the treatment measures no longer limited to a single standard but able to be dynamically adjusted according to the actual situation of individual plants. The setting of the sliding time window separates short-term fluctuations and long-term trends, and combined with the singular spectrum analysis algorithm, it realizes the multi-scale analysis of the changes in growth indicators, enhancing the ability to capture growth dynamics, so that physiological abnormalities can be sensed in the early stage of disease occurrence and early intervention can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a schematic diagram of the steps of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0043] Please refer to Figure 1 , the present invention provides a technical solution, a method for dynamically monitoring the correlation between the growth state of Artemisia argyi and leafhopper infestation, including the following steps:

[0044] Regularly photograph Artemisia argyi using a multi-spectral camera to collect the reflected spectral information in the near-infrared, red, and green bands to obtain preliminary physiological parameter data; based on the preliminary physiological parameter data, evaluate the health status and growth trend of Artemisia argyi to generate a health assessment result;

[0045] Based on the preliminary physiological parameter data, through deep learning algorithms for training and analysis, detect whether there are lesions or leafhopper infestations on plant leaves, determine the types of lesions and the degree of infestation to generate a lesion infestation diagnosis result; based on the lesion infestation diagnosis result, mark the damaged area and calculate the damaged area ratio to generate an analysis result of the damaged area;

[0046] Based on the analysis results of the damaged area and the health assessment results, treatment and intervention plans are formulated according to the degree of damage and leaf health indicators, and intervention plan formulation results are generated;

[0047] According to the results of the intervention plan formulation, a sliding time window was set to capture the short-term fluctuations and long-term trends in the growth process of Artemisia annua and generate multi-scale growth analysis results.

[0048] The steps for obtaining preliminary physiological parameter data are as follows:

[0049] Regularly photograph mugwort to collect reflectance spectrum information in the near-infrared, red, and green bands to obtain a spectral image dataset;

[0050] Extract spectral intensity, peaks and troughs from the spectral image data set, perform signal enhancement and background noise removal on the spectral intensity, peaks and troughs to obtain a cleaned spectral feature data set;

[0051] Based on the cleaned spectral feature dataset, principal component analysis was applied to identify patterns and trends in the spectral data and obtain preliminary physiological parameter data.

[0052] Specifically, images of mugwort in the same planting area are taken regularly. When taking pictures, first arrange the periodic collection of mugwort according to the time intervals listed in the operation instructions, for example, start shooting one hour after sunrise every 48 hours, obtain at least five original images each time through visible light and near-infrared cameras, and record the ambient temperature and humidity during shooting. Then, set the focal length, exposure time and aperture size based on a pre-compiled list of shooting equipment parameters. The exposure time here can be determined based on equipment performance and empirical measurement values. For example, 1 / 200 second can be set on a sunny day with strong lighting conditions. If the exposure during the test exceeds this range, the focal length or aperture will be fine-tuned. The range of exposure can be determined based on the equipment manufacturer's data and The average value of historical shooting results is comprehensively formulated. For example, if a reasonable exposure is measured between 0.7EV and 1.2EV in several shots, the ambient temperature can be between 0℃ and 50℃ and the relative humidity can be between 0% and 100% during the shooting process. If the temperature or humidity exceeds this range during shooting, it is necessary to first check whether there is condensation or fault on the device casing and lens. After confirming that the photographic equipment is working normally, continue shooting and record the time label and shooting number of each shot. Then, the data of the three reflection bands of near-infrared, red light and green light are captured simultaneously and digitally converted through the photosensitive components inside the camera. The converted raw data are summarized at one time, and finally a spectral image data set that can be used in subsequent links is obtained.

[0053] Extract the spectral intensity, peak and valley point information corresponding to each image from the acquired spectral image dataset. When extracting, first scan the pixel matrix of the image one by one according to the band numbers listed in the operation instruction list, and judge the spectral response position corresponding to each pixel by comparing with the known reference wavelength range. If the pixel intensity value exceeds the standard pixel range of 0 to 255, mark it as abnormal and discard it. Then, preliminarily divide the intensity curve according to the defined threshold. For example, the threshold for peak selection can be set as the empirical average plus 1.5 times the standard deviation, which is calculated from all the image intensity values read. If the intensity of a certain point exceeds this value, it is determined as a candidate peak position. The corresponding valley selection can be identified by subtracting 1.2 times the standard deviation from the average value. Then, filter out the noise for these candidate points. During the noise filtering process, for example, the smoothing filtering method can be used to judge the difference between adjacent pixel points before and after. If the difference is abnormally greater than the amplitude between 2 and 5 determined by testing in practice, this peak or valley is regarded as noise and excluded. Then, use the signal enhancement method to perform unified normalization operation on the remaining effective peak and valley corresponding intensities. When normalizing, the difference between the maximum value and the minimum value can be selected as the denominator, so as to map the intensity distribution between different images to the interval between 0 and 1. After all images have completed the above processing, summarize all the eigenvalues and arrange them in a list according to the previous pixel scan sequence number. In this way, the integrated data without background impurity information is obtained, and finally the cleaned spectral feature dataset is formed.

[0054] Based on the cleaned spectral feature dataset, use principal component analysis to reduce the dimension and summarize the patterns of the values in it. When performing this step, first read the normalized intensity matrix corresponding to all wavelengths and arrange it in the form of a row vector, and then determine the number of principal components to be retained according to the calculation criteria listed in advance. For example, an empirical threshold can be set to require the cumulative contribution rate to reach 90%. When the k-th principal component is added and the total contribution rate has exceeded 0.90, stop adding new principal components. Before formally calculating the principal components, it is necessary to first construct the covariance matrix, and its elements can be defined by the covariance between any two columns of spectral eigenvalues. Subsequently, solve the eigenvectors through numerical decomposition, sort all the eigenvectors according to the eigenvalues from large to small, and the column vectors corresponding to the first k eigenvectors can be used as the main contribution components. When determining the value of k, if the cumulative contribution rate does not reach the previously set 0.90, the value of k can be appropriately increased. If it is still insufficient when the value of k continues to increase, it means that the data dispersion is relatively large, and it is necessary to re-examine the feature extraction link or relax the contribution rate threshold. After meeting the threshold conditions, multiply the selected principal components by the original feature matrix to obtain several new score vectors, and these score vectors can respectively represent the combination coefficients of different patterns and trends. After obtaining all the principal component scores, arrange them in a data table according to the index numbers of the samples, and finally obtain the preliminary physiological parameter data.

[0055] The steps to obtain the health assessment results are as follows:

[0056] Analyze the preliminary physiological parameter data, identify the chlorophyll content, water status, and growth rate, and construct a physiological index dataset;

[0057] Based on the physiological index dataset, calculate the health score of Artemisia argyi, and the calculation formula is:

[0058]

[0059] where L i represents the i-th chlorophyll content value, W i represents the i-th water status value, G i represents the i-th growth rate, n is the number of samples, and HC is the health score;

[0060] Based on the health score, combined with the growth data and temperature and light conditions, conduct trend analysis to obtain the health assessment results.

[0061] Specifically, when analyzing the previously obtained preliminary physiological parameter data, it is necessary to first confirm whether the preliminary physiological parameter data contains the identification number information and corresponding time stamps of each Artemisia argyi plant. If it is found that there are duplicates in the identification number or time stamp, check one by one according to the sorting instructions whether it is a misrecording or a missed shot. After confirming the integrity of the data, set the corresponding effective ranges based on the previously recorded chlorophyll content values, water status values, and growth rate values. For example, compare the chlorophyll content value with the range between 0 mg / g and 5 mg / g, check the water status value with the range between 0% and 100%, and match the growth rate value with the range between 0 cm / day and 5 cm / day. Records outside these ranges are regarded as outliers. If the number of some outliers is relatively large, it is necessary to further analyze whether there is a malfunction in the shooting device or test instrument according to the camera shooting time or the corresponding sampling serial number. After removing the outliers or correcting the suspicious data, combine the chlorophyll content values, water status values, and growth rate values of each Artemisia argyi plant at the same time period, arrange the data points of the three in the same time sequence, and check whether there is a missed sampling problem through multiple comparisons of the same batch of data. After confirming that there is no error, mark and record it in the corresponding observation sequence, and then integrate the data of each Artemisia argyi plant by number into a single physiological index combination entry. To ensure data traceability, record the acquisition time of each group of entries at the same time. After completing the integration of all entries, summarize them into a physiological index dataset. During this summarization stage, if there are duplicate entries or missing entries, it is necessary to query and supplement the data according to the identification number, so as to finally obtain the physiological index dataset.

[0062] The advantage of the formula is that by simultaneously considering the geometric combination of chlorophyll content and water status and the reciprocal addition of growth rate, the impact of growth rate reduction on the health score can be more intuitively observed, and the physiological indicators of plants can be comprehensively measured as a whole. In subsequent monitoring and analysis, the growth status of Artemisia argyi at different stages can be compared and tracked based on the score result.

[0063] L i The steps for obtaining the parameter are as follows: First, extract chlorophyll samples from Artemisia argyi leaves. This operation requires the use of chemical extraction methods. The specific method is to place an appropriate mass of leaves in an organic solvent and ensure that the ratio of the solvent volume to the leaf sample mass is constant. Then, measure the absorbance of the extract using a photometer in an environment of 20°C to 25°C, convert the absorbance value to chlorophyll content, and then match it according to the sampling number of each Artemisia argyi plant. Record the obtained chlorophyll content in mg / g as L i , if there are multiple measurements during the same period, take the average value. The average value is obtained by adding all the measurement results and then dividing by the number of measurements. For example, if the leaves of the same Artemisia argyi plant are measured three times continuously within three days, and the measured chlorophyll contents are 1.6 mg / g, 1.4 mg / g, and 1.5 mg / g respectively, then L i can be recorded as (1.6 + 1.4 + 1.5) / 3 = 1.5 mg / g. After all the leaves are measured, sort out the sequence entries of L i .

[0064] W i The steps for obtaining the parameter are as follows: Use a moisture analyzer to detect the moisture content on the collected Artemisia argyi leaves, compare the mass of each leaf with the mass after drying, and the ratio of the mass difference to the mass after drying is the leaf water status value. During the detection, it is necessary to first perform a constant-temperature drying treatment on the leaves, place them in a drying oven set at 105°C and continuously dry for 2 to 4 hours, record the mass before and after drying in g, and convert the ratio of the difference between the two to the mass after drying into a percentage form and record it as W i , if there are multiple leaf samples in one measurement, calculate them separately according to the same principle. For multiple measurements, the average value can be taken or necessary checks can be made for extreme values. For example, if the water status values obtained from three measurements are 72%, 68%, and 70% respectively, then W i is sorted out as (72% + 68% + 70%) / 3 = 70%. After processing all Artemisia argyi samples in the same way, obtain the W i sequence.

[0065] G iThe steps to obtain the parameters are as follows: Confirm the growth rate by observing the change in the length of Artemisia argyi within a unit time. Fixed-point measurements can be carried out every 48 hours. Use a ruler to measure the length from the root to the topmost tip and record it in centimeters. Divide the difference between two consecutive measurements by the time interval to obtain the single growth rate. Multiple measurements can be accumulated and statistically analyzed. When measuring, the same Artemisia argyi plant needs to be repeatedly compared at the same position and ensure that the ruler is vertically aligned with the root to obtain as accurate data as possible. If a plant is measured three times within five days and the length values obtained are 22.0 cm, 24.5 cm, and 27.0 cm respectively, then the growth rates within two 48-hour periods can be calculated as (24.5 - 22.0) / 2 = 1.25 cm / day and (27.0 - 24.5) / 2 = 1.25 cm / day. Since the two results are the same, both being 1.25 cm / day, then G i can be uniformly recorded as 1.25 cm / day.

[0066] The steps to obtain the n parameter are as follows: When calculating the health score, it is necessary to confirm the number of valid samples collected, that is, the total number of Artemisia argyi samples participating in the operation. If 50 Artemisia argyi plants with complete chlorophyll content, moisture status, and growth rate values are recorded in the previous collection link, then n is determined to be 50. If individual plants are found to have missing or abnormal data and cannot participate in the operation during subsequent monitoring, they will be removed from the statistics and n will be updated. In specific operations, first summarize the numbers of all eligible Artemisia argyi plants and check one by one whether the three indicators are valid. If all are valid, they will be included in the total of n. For example, if after confirmation, the three indicators of 45 plants are available for use, then n = 45.

[0067] Calculation process:

[0068] If 3 valid samples (n = 3) are taken for example, let L 1 = 1.5 mg / g, W 1 = 70%, G 1

[0069] = 1.25 cm / day, L 2 = 1.6 mg / g, W 2 = 68%, G 2 = 1.30 cm / day, L 3 = 1.4 mg / g, W 3 = 72%, G 3 = 1.20 cm / day. First, convert W 1 , W 2 , W 3 into a ratio between 0 and 1. For example, 70% is recorded as 0.70, 68% is recorded as 0.68, and 72% is recorded as 0.72. Calculate under the same dimension For the first sample, Meanwhile, 1 / G 1 = 1 / 1.25 = 0.80. Therefore, the bracketed term for the first sample is 1.655 + 0.80 = 2.455. The value of the second sample can be calculated in a similar manner as The value of the third sample is

[0070] Adding the three together gives

[0071] 2.455 + 2.5092 + 2.4073 = 7.3715. Then dividing by n = 3, the result HC = 7.3715 / 3 = 2.45717. Rounding it gives 2.4572 which can be recorded as such.

[0072] This result indicates that the value of HC is 2.4572, representing the health score obtained through average calculation based on the geometric combination of chlorophyll content, moisture status, and the reciprocal addition of growth rates of the above three Artemisia argyi plants during the current collection period. In subsequent monitoring, the HC values of different periods can be continuously compared to determine whether it is rising or falling, thereby obtaining a quantitative reflection of the overall health level of the samples.

[0073] Based on the previously calculated health score, compare it with the previously integrated growth data and the temperature and light conditions recorded during each shooting. During the execution process, the temperature values at the same time node need to be uniformly converted to Celsius for inspection between 0°C and 50°C. If it exceeds this range, it is necessary to confirm the calibration status of the temperature sensor or check on-site whether the measurement is inaccurate due to high or low temperature. The light conditions can be checked according to the light intensity recorded by the on-site illuminometer in the range of 0 lx to 200,000 lx. If the illuminance is not in this range, it is necessary to check whether it corresponds to night collection or collection in a dark environment. After completing the above data merger, arrange the three groups of values of temperature, light, and health score in a time series. Taking a continuous number of observation periods as a statistical unit, check the fluctuations of the health score within each day or hour according to the continuous values of the health score within the observation period. For example, to count the health score within 24 hours of a day, it can be divided into 24 observation points and combined with temperature and light data to form three columns of records respectively. Then, summarize these records in the order of the acquisition time. Compare the data of the daily observation period with the previously set growth trend range. For example, if the growth trend range is set to a health score in the range of 2.0 to 3.0, a temperature in the range of 15°C to 35°C, and a light intensity in the range of 10,000 lx to 150,000 lx, then the three indicators at different time points can be statistically analyzed in segments to check whether they meet the above range. After confirming that each indicator is within the above range, this period can be recorded as normal. If there is a situation where it exceeds this range, it is necessary to mark the corresponding time period and retrieve more details for investigation. Finally, after integrating the health score with the temperature and light records for each period, perform linear or polynomial fitting to obtain the data change trend in the corresponding time series. The operation results of these steps are combined to obtain the health assessment result.

[0074] The steps to obtain the lesion invasion diagnosis result are as follows:

[0075] Based on the preliminary physiological parameter data, configure and train a deep learning model. By identifying the lesion characteristics and leafhopper invasion signs in the leaf images and applying a convolutional neural network to optimize the recognition accuracy, a trained model is obtained.

[0076] Use the trained model to analyze the newly collected leaf data. By parsing the image data, identify and classify the types and invasion degrees of leaf lesions, and generate the lesion invasion diagnosis result.

[0077] Specifically, based on the preliminary physiological parameter data obtained previously and the corresponding labeled leaf image information, first refer to a control list containing descriptions of lesion types and descriptions of leafhopper infestation signs to group and number the images in the required training set and validation set. Subsequently, during the image preprocessing stage, uniformly adjust the input resolution to 256×256 pixels and convert it to the standard RGB channel format. Then, in the image annotation file, check the lesion types and infestation feature labels of each picture one by one according to the pre-established type labels. If a situation inconsistent with the descriptions in the list is found, query the original shooting record and update the annotation. After that, prepare the training process of the convolutional neural network. First, set the batch size to 32 and the learning rate to 0.001. The selection of these parameters is based on the numerical range set by combining the training error curve situation and the video memory capacity test results collected in previous similar plant pathology image classification tasks. At the same time, randomly shuffle the training set in each training loop and send it into the network in batches. If the accuracy of the validation set no longer improves after a certain number of training rounds, terminate the training in advance. The threshold of the number of terminated rounds can be defined as 5 by referring to the phenomenon that the accuracy of the validation set does not improve for 5 consecutive epochs determined previously. After the training is completed, check whether abnormal image samples need to be supplemented or removed according to the final accuracy and loss value of the validation set. If it is found that some images have errors during the annotation stage, re-check the labels. If no abnormalities are found, confirm that the training is completed and save all the weight parameters of the current neural network. Then, record the network together with the corresponding structural parameters to form a trained model.

[0078] Use the trained model to load the newly collected leaf image data and perform scaling processing with a size of 256×256 pixels. Then, convert the pixel values of the image to the range between 0 and 1 and standardize them according to the recorded mean and standard deviation of the three channels. Subsequently, input the processed image into the above-mentioned trained network. After the network extracts image features in the convolutional layer and pooling layer, it will enter the classification layer. The classification layer scores according to the lesion types and leafhopper infestation signs corresponding to the previous training, so as to output the probability values of various lesions and the degree of infestation. If the probability value of a certain class exceeds the threshold of 0.8 set based on the comprehensive F1 score of the validation set, determine the current image as this lesion type and record the degree of infestation. The threshold 0.8 is calculated from the compromise result of the optimal accuracy and recall rate on the validation set. If the predicted probabilities of all classes are lower than 0.8, it is determined that the image has an uncertain situation and can be transferred for manual inspection or re-confirmation of the label. After the inference of all image data is completed, summarize the lesion types and infestation degree labels corresponding to each image to generate the lesion infestation diagnosis result.

[0079] The steps to obtain the analysis result of the damaged area are as follows:

[0080] Based on the diagnosis result of lesion invasion, extract the pixel distribution information of the damaged area, identify the boundary of the lesion area on the leaf surface, remove noise and enhance the contrast of the damaged area, and generate a binary image of the damaged area;

[0081] According to the binary image of the damaged area, calculate the damaged area ratio. The calculation formula is:

[0082]

[0083] where X j , Y j is the coordinate of the j-th damaged pixel, X c , Y c is the coordinate of the center of the damaged area, m is the total number of damaged pixels, A is the total leaf area, and R is the damaged area ratio;

[0084] Based on the damaged area ratio, combined with the geometric morphological characteristics of the damaged area, analyze the spread trend of the lesion, classify the lesion type, and generate the analysis result of the damaged area.

[0085] Specifically, based on the lesion invasion diagnosis result obtained previously, extract the pixel distribution information of the damaged area in the corresponding image. First, retrieve all pixels presenting lesion characteristics or cicada leafhopper invasion traces in the image data, and screen these pixels according to the chromaticity range of lesion pixels and the texture feature threshold listed in the annotation manual. When screening, the values of the pixels in the RGB channels are read one by one and compared with the standard range between 0 and 255. When it is detected that the value exceeds the upper limit of the chromaticity threshold calculated from the historical image statistics in advance, such as exceeding the value of 220, or being less than the value of 30, it is listed as a suspicious pixel for subsequent comparison. If the texture feature also has an effective matching degree with the reference texture data exceeding the set 80%, it is marked as a damaged area pixel. After extracting a series of damaged pixels obtained into a unified data list, identify the lesion boundary by referring to the coordinate distribution of this data list. For the convenience of boundary tracking, iterative calculation can be performed along the outer contour of this coordinate set. If the distance between two consecutive contour points is greater than the range of 2 to 5 pixels determined according to experience, it is regarded as noise or detection deviation and excluded. Subsequently, according to the distribution of the actual imaging brightness in the range of 0 lx to 200000 lx, enhance the brightness of the lesion edge pixels, map the brightness values uniformly to the higher end segment between 0 and 255, form a damaged area that is easier to identify after enhancement, and finally mark all valid damaged pixels as 1 in a binary manner and mark other areas as 0 to obtain the binary image of the damaged area.

[0086] The advantage of the formula is that it can characterize the overall range of the damaged distribution by combining the cumulative distance between damaged pixels and the central coordinates. Compared with the simple count of damaged pixels, this formula takes into account the degree of discreteness of the lesion morphology, thus more comprehensively reflecting the expansion characteristics of the damaged area in subsequent judgments.

[0087] X j The steps for obtaining the parameter are as follows. When retrieving the j-th damaged pixel in the binary image, read its coordinate value in the horizontal direction. For each image, the horizontal direction is defined as the incremental value from left to right through the coordinate axis normalization process, and the range is usually between 0 and the image width minus 1. To obtain the specific value of X j it is necessary to first map according to the row and column positions of each pixel in the image. The row number corresponds to the vertical direction, and the column number corresponds to the horizontal direction. After reading the column number, record it as X j , in practice, the image width can be determined with reference to the shooting resolution. For example, when the image resolution is 1024×768 pixels, X j can be distributed between 0 and 1023. To exclude invalid coordinates, all X j should be controlled within the range of 0 to 1023. If the read value exceeds this range, it is necessary to further check whether there is image distortion or pixel offset. After completion of the confirmation, it can be summarized into the X j sequence. The following lists an example of obtaining: In a 768×768 image, a certain lesion pixel is located in the 400th column, record X j = 400.

[0088] Y j The steps for obtaining the parameter are as follows. For the j-th damaged pixel in the same binary image, read its coordinate value in the vertical direction. Here, the vertical direction is defined as increasing row by row from top to bottom, and the range is usually between 0 and the image height minus 1. The specific obtaining method is similar to that of X j , except that it corresponds to the pixel row number instead of the column number. If the image resolution is 1024×768, then Y j can take values between 0 and 767. When recording, it is necessary to keep the row and column information of the pixel points in one-to-one correspondence, and uniformly organize them into the form of (X j , Y j ). In actual use, if it is found that Y j is too close to the upper and lower edges of the image, it is also necessary to check whether there are misclassifications at these extreme positions in the lesion detection to avoid affecting subsequent analysis. The following lists an example of obtaining: In a 768×768 image, if a certain lesion pixel is located in the 250th row, then Y j = 250.

[0089] X cThe steps to obtain the parameter are as follows. After confirming all the damaged pixel sets, it is necessary to calculate the geometric center of the lesion area, and the abscissa X of this center c can be obtained by first averaging the X of the damaged pixels j Collect all the X j Then add them up and divide by the number m of damaged pixels. This method is usually called centroid calculation in the field of image processing and is used to represent the average position of the distribution. When reading X c it is necessary to ensure that the image coordinates are stable. If the image has been transformed beforehand, it needs to be recalculated after the transformation is completed. The following is an example of obtaining: When the lesion area contains 200 damaged pixels, and the recorded X j are 120, 121, 122, ……, 180 respectively. Add these values and divide by 200 to obtain X c , if the sum is 30500, then X c = 30500 / 200 = 152.5.

[0090] Y c The steps to obtain the parameter are the same as those for X c . It is necessary to centrally count the vertical coordinates Y of all the damaged pixels j and add them up and divide by m. In this way, the coordinate value of the center of the lesion area in the vertical direction can be obtained. If the image has been cropped or scaled, the value of Y j also needs to be transformed accordingly before performing the operation to avoid inconsistent coordinate systems. To ensure the accuracy of Y c , the calculation must be carried out after all the pixels in the lesion area have been confirmed. The following is an example of obtaining: When the cumulative sum of the Y of the damaged pixels in an image j is 41000 and a total of 250 damaged pixels are counted, then Y c = 41000 / 250 = 164.

[0091] The steps to obtain the m parameter are as follows. Statistically count the total number of damaged pixels in the lesion area, that is, the number of all pixels marked as 1 in the binary image. After summing, m is obtained. Usually, the acquisition method will first distinguish the lesion from the normal area through threshold segmentation or morphological methods, and then record the pixels of the lesion part into a searchable list. m is the length of the list. If 500 damaged pixels are counted in a certain monitoring, then m = 500.

[0092] The steps to obtain the A parameter are as follows. The total leaf area refers to the sum of the number of all pixels in the entire leaf area in the binarized or corresponding segmentation result. Or, under the condition that the physical size is known and the image ratio is accurate, the number of pixels can be converted into square millimeters. However, in this formula, it is also feasible to directly use the form of the number of pixels. The key is to ensure the measurement consistency of R. When obtaining A, the overall contour of the leaf can be located in the system, and all pixels within the contour are counted as the leaf area. If the total number of pixels in the leaf is statistically 10,000, then A = 10,000.

[0093] Calculation process:

[0094] When the image resolution is 768×768, the total leaf pixel area A = 10,000. It is calculated that the lesion area contains m = 5 pixels, and the coordinates are (120, 130), (121, 131), (122, 129),

[0095] (118, 132), (119, 129). First, add all the X j values (120, 121, 122, 118, 119) to get 600 and divide by 5 to obtain X c = 120. Similarly, add the Y j values (130, 131, 129, 132, 129) to get 651 and divide by 5 to obtain Y c = 130.2. Subsequently, calculate and sum for each pixel according to For example, for the first pixel (120, 130), we can get After calculating all five pixels and adding the results, the example calculation totals 1. Then substitute it into the formula:

[0096]

[0097] This result indicates that when the calculated R = 0.0001, it means that the current lesion has a small degree of spatial dispersion, and the total distance from the lesion pixels to the center coordinates is very weak relative to the total leaf area. If R increases significantly, it represents that the range of the lesion area shows a more extensive distribution.

[0098] Based on the damaged area ratio calculated previously, compare it with the geometric shape information of the damaged area identified previously. First, read the lesion contour identified in the same image and extract the sequence of polygon vertices of the contour. Subsequently, the major axis and minor axis of the polygon can be calculated to determine whether the lesion area is elliptical or irregular in shape. Then, refer to the time series record stored in the image, and arrange and compare the area, perimeter of the current lesion area, and the previously obtained damaged area ratio with the records of the same position observed several times before. If the difference between the major axis and minor axis of the current contour increases by more than the empirical value of 5 to 10 pixels sorted out by pathologists compared with the previous record, the current lesion area can be regarded as having a tendency to expand outward. If the lesion area shows an increase in area and a deformation from a circular shape to a more irregular shape several times in a row, a suitable category can be selected in the lesion type identification file for classification. If it is found that the edge features of the lesion contour are very similar to the previous record and the damaged area ratio does not increase significantly during several observations, it can be judged that the lesion is in a relatively stable state. Combine all these statistics on area and geometric shape and organize them into a complete classification conclusion to generate the analysis result of the damaged area.

[0099] The steps to obtain the intervention plan formulation result are as follows:

[0100] According to the damaged area analysis result and the health assessment result, calculate the intervention priority score. The calculation formula is:

[0101]

[0102] Among them, R represents the damaged area ratio, H represents the health score, E represents the lesion spread rate, C represents the complexity of the damaged area boundary, B represents the complexity of the healthy area boundary, SG represents the leaf growth rate, F represents the leaf moisture content, and P is the intervention priority score;

[0103] Based on the intervention priority score, combined with the lesion type and growth environment conditions, screen and match the treatment and intervention measures to generate the intervention plan formulation result.

[0104] Specifically, the advantage of the formula lies in that it comprehensively integrates various information such as the damaged area ratio, health score, lesion spread rate, the difference between the complexities of the damaged area and healthy area boundaries, growth rate, and moisture content in the same expression. Compared with using a single indicator for evaluation, it can better highlight the balance relationship between the degree of disease development and the overall growth state of Artemisia argyi, thus providing a more explicit reference in the subsequent intervention process.

[0105] The steps to obtain the R parameter are to determine the value of R through the previously obtained damaged area ratio.

[0106] The steps to obtain the H parameter are to determine the actual value of H with reference to the previously obtained health score result.

[0107] The steps to obtain the E parameter are as follows: determine the specific value of E according to the lesion diffusion rate statistically obtained from multiple shootings and monitorings of the lesion area in the previous text. The lesion diffusion rate can be obtained by comparing the growth of the lesion area or the boundary range of the same leaf within consecutive shooting intervals (such as 48 hours). If the detected lesion ratio is 3% during one shooting and reaches 5% during the next shooting 48 hours later, then the change amount of the two lesion ranges can be divided by the shooting interval time and expressed in the form of percentage per day or pixel area per day. At the same time, attention should also be paid to whether the change of the dispersion degree or the number of broken points is stable. During the acquisition process, it is necessary to ensure that the imaging angle and lighting conditions are consistent to avoid excessive errors. For example, if the final statistical result shows that the diffusion rate of a certain leaf within 48 hours is 1.0% / day, that is, E = 1.0 (if using pure numerical values, it can be regarded as 1.0).

[0108] The steps to obtain the C parameter are as follows: measure the complexity of the boundary by reading the geometric structure of the damaged area boundary. First, the pixel set of the outer contour of the lesion will be extracted from the binary image, and then the number of broken line segments, the number of concave and convex points, and the average curvature value of the contour on the coordinate plane will be calculated. After summing up or performing weighted combination of these indicators, a numerical result can be obtained. This value can reflect whether the outer edge of the lesion is smooth or has complex features such as fine fragmentation. If there are a large number of small concave points or a high degree of curvature on the edge, it will lead to an increase in complexity. To ensure the coherence of the calculation, the same resolution and binary threshold strategy need to be used for unified processing. When recording, the complexity value can be controlled between 0 and 10 for subsequent comparison. For example, when it is statistically found that there are many broken line segments on the lesion contour line and the average curvature value reaches 2.0, then C can be comprehensively defined as 6.5.

[0109] The steps to obtain the B parameter are as follows: determine the complexity of the healthy area boundary according to the shape characteristics of the healthy area boundary. The healthy area boundary can be statistically obtained from the contour formed by the remaining outer edge of the leaf after removing the lesion pixels in the same leaf image. Weighted counting is performed on the curvature fluctuation and the number of inflection points of the outer edge. If the overall shape of the leaf is a relatively regular ellipse or an approximate ellipse, the complexity of the healthy area boundary will be relatively low. If the shape of the leaf itself is irregular or there are redundant lobes, this value will increase. In order to form a comparable unified measurement, the dimension can be kept consistent with the same method as C, and it is necessary to ensure that the angles and resolutions used in the evaluation process are the same. For example, when it is detected that the curvature fluctuation is very small and the number of broken line segments is extremely small in a relatively regular leaf, it can be recorded as B = 3.2.

[0110] The steps to obtain the SG parameter are as follows: Determine SG by measuring the growth rate values of the leaves over several consecutive observation periods. The growth rate is generally measured by the increment of the stem or leaf length per unit time (such as per day). Combining with the data collected in multiple batches in the early stage, the growth rate can be obtained by dividing the length difference measured at intervals of 48 hours or 24 hours by the specific time interval for each batch, and then averaging the multiple measurement results of the leaves in the same leaf or the same area to obtain a relatively stable growth rate value. To ensure that it is in a similar order of magnitude or easy to compare with parameters such as the lesion spread rate E, the growth rate can also be subjected to a certain normalization process. For example, if the length of a certain Artemisia argyi plant grows from 25.0 cm to 27.5 cm within 48 hours, that is, the daily increment is 1.25 cm / day, then SG = 1.25.

[0111] The steps to obtain the F parameter are as follows: Take the value corresponding to the current leaf moisture content. The moisture content can be measured by the previous constant-temperature drying method or a professional moisture meter and recorded as a percentage or unit mass ratio, and can be matched with the corresponding plant serial number. For example, first dry it in an environment of 105 °C for two hours and then measure the mass reduction value, and then compare it with the original mass to obtain the moisture content percentage. If the multiple observation results fluctuate, the short-term average value or the representative value obtained by weighting within the interval set according to the detection guidelines can be selected, and it is characterized in the range of 0 to 100%. For the convenience of formula operation, this value can be converted between 0 and 10. For example, mapping 70% moisture content to 7.0, that is, F = 7.0.

[0112] Calculation process:

[0113] In a complete example, let R = 0.018, H = 2.46, E = 1.0, C = 6.5, B = 3.2, SG = 1.25, F = 7.0. First, expand the numerator part:

[0114] R×H = 0.018×2.46 = 0.04428

[0115] E 2 =(1.0) 2 =1.0

[0116] |C - B| = |6.5 - 3.2| = 3.3

[0117] Adding the three together, the numerator can be obtained as:

[0118] 0.04428 + 1.0 + 3.3 = 4.34428

[0119] The denominator part is:

[0120] SG + F = 1.25 + 7.0 = 8.25

[0121] Then perform the overall division:

[0122]

[0123] Finally, take the square root:

[0124]

[0125] The results showed that the intervention priority score P = 0.7264, which means that in the comparison between the current lesion damage and the physiological state of Artemisia argyi itself, the larger the value, the more attention and faster intervention are needed. If the subsequent measured value is greater than 1.0, it often indicates that the imbalance between the lesion area and the growth state is increasing. When P is lower than 0.5, it generally shows that the problem is not serious or can be self-regulated in the short term.

[0126] Based on the intervention priority score calculated previously and in association with the previously identified lesion types and current growth environment conditions for comprehensive comparison, first read the pathology files recorded in the system and the data of the test environment this time. For example, sort out the monitoring marks that the air temperature is maintained in the range of 20℃ to 30℃ and the water supply is in a relatively sufficient state. Match these environmental data with the intervention priority score and screen them line by line. If the score is within the previously established range of 0.5 to 1.0, it is necessary to confirm whether the lesion type is a fast-spreading type or a slow-spreading type. In this process, the lesion type index table can be retrieved to compare with its common corresponding medication instructions and a list of targeted treatment measures. If the lesion type is marked as a easily spread disease and the score reaches 0.7 or above, it will be recorded in the key focus list and the available drug specifications, application methods and times. If the lesion type is marked as a slowly spreading disease and the score is between 0.5 and 0.7, it is classified as a general concern and does not require immediate special chemical intervention. Instead, conventional management measures can be given priority. Combined with subsequent changes in the growth environment, such as whether the temperature breaks through the 0°C to 40°C range or whether the light intensity is between 10,000lx and 150,000lx, if the records show that the environmental conditions are generally stable, a set of basic intervention processes that meet the characteristics of this type of disease are compiled, including spraying appropriately formulated nutrient solutions or supplementing trace elements and other measures. On the contrary, if the score is higher than 1.0 or significant abnormalities in temperature and light are found, consider upgrading to higher doses or more frequent chemical control or physical isolation. In this way, possible intervention measures are matched one by one and arranged in an executable list to ultimately generate the results of the intervention plan.

[0127] The steps to obtain the multi-scale growth analysis results are:

[0128] According to the results of the intervention plan, a sliding time window was set, and the growth index data of Artemisia argyi during its growth process were segmented at fixed time intervals. Growth characteristics including leaf area growth rate, stem height increment, and chlorophyll concentration change rate were extracted to generate a time window growth characteristic matrix.

[0129] Based on the growth feature matrix of time windows, calculate the growth fluctuation intensity, and the expression is:

[0130]

[0131] where I represents the growth fluctuation intensity, XA j is the leaf area of the j-th time window, XH j is the stem height of the j-th time window, XC j is the chlorophyll concentration of the j-th time window, and xm is the number of time windows;

[0132] Based on the growth fluctuation intensity, apply singular spectrum analysis to generate the multi-scale growth analysis results.

[0133] Specifically, according to the obtained intervention plan formulation results and set the sliding time window according to the listed time nodes. In specific operations, the Artemisia argyi growth indicators collected every 24 hours in consecutive days can be divided into a unit first. For example, the data from the morning of the 1st day to the morning of the 2nd day is regarded as a window, and the data from the morning of the 2nd day to the morning of the 3rd day is regarded as the next window. In each window, sort out and record the change value of the leaf area and the height increment of the stem from the bottom to the top. For the chlorophyll concentration, based on the multiple test results obtained by extraction and photometer measurement within the corresponding time period, and calculate its change rate according to the average value at the start and end of these concentration data. When calculating the leaf area growth rate, first divide the difference between the area value of the previous window and the current window area by the number of days in the window length. The stem height increment can be compared with the initial and final measurement values in different windows in the same way. If the measurement frequency is high, the average value can also be taken within a single window. Among them, it is necessary to check the extreme temperature and humidity conditions during measurement. If the temperature exceeds the equipment range of 0°C to 45°C or the relative humidity exceeds 0% to 100%, the reliability of this measurement should be verified first and the fault record should be excluded. In this way, the three growth characteristics of the leaf area growth rate, the stem height increment, and the chlorophyll concentration change rate can be obtained in each segmented window. When the data of all windows are completed, combine these three indicators of each window into a matrix form of three columns or multiple columns, and label the corresponding time window serial number for each row of the matrix. At this time, if it is found that there is missing information in some windows, it can be traced back according to the original measurement timestamp. Finally, arrange the three indicators of all valid windows completely and summarize them to obtain the growth feature matrix of time windows.

[0134] The advantage of the formula lies in integrating the changes in the Artemisia argyi growth indicators within multiple consecutive time windows by taking the mean of the sum of squares. It can not only examine the overall fluctuation range among the three types of numerical values of leaf area, stem height, and chlorophyll concentration, but also perform normalization processing through the dimension of the number of time windows in the denominator, facilitating unified measurement under different collection scales, and playing a role in reflecting the multi-dimensional fluctuation intensity in the growth dynamic analysis.

[0135] XA j The steps for obtaining the parameter are as follows. After setting the observation window at each fixed time interval, the Artemisia argyi leaf areas observed within the same window (which can be converted by taking images and combining pixel counting, or converted after measurement based on direct measurement tools to cm 2 ) are aggregated and an average value is taken, and this average value is regarded as the value of XA j . If the leaf area is measured multiple times within the same time window, first sum up all the values and then divide by the number of measurements to obtain the window average area. For example, if the leaf areas recorded in the j-th window are 45 cm 2 , 48 cm 2 and 47 cm 2 , then XA j = (45 + 48 + 47) / 3 = 46.67 cm 2 .

[0136] XH j The steps for obtaining the parameter are as follows. The stem heights measured in the current time window are uniformly processed and averaged to obtain the value. The specific measurement process can use a scale to read the length from the root to the highest point in cm. If there are multiple measurements within the same window (such as twice in the morning and evening or more), these measurement results can be added and then divided by the number of measurements. Since the stem height often increases with time without obvious decline, it is necessary to check for null values or extreme values (such as instrument deviation caused by the temperature exceeding 45°C during collection). After confirmation, XH j is obtained. For example, if the stem heights measured multiple times in the j-th window are 13.5 cm, 14.0 cm, and 14.1 cm, after averaging, XH j = (13.5 + 14.0 + 14.1) / 3 = 13.87 cm.

[0137] XC j The steps for obtaining the parameter are as follows. Record the chlorophyll concentration within this time window and weight or average the results of multiple tests to obtain XC j . The chlorophyll concentration usually needs to be determined by chemical extraction and a spectrophotometer. First, dissolve the leaf sample in an organic solvent, measure the optical density after the color development is stable, and then use the conversion coefficient to obtain the concentration value in the form of mg / g. If the sample is taken 3 times in the same window and the results are 1.4 mg / g, 1.5 mg / g, and 1.6 mg / g respectively, then XCj =(1.4 + 1.5 + 1.6) / 3 = 1.5 mg / g.

[0138] The steps to obtain the xm parameter are as follows. The number of time windows is determined by the total number of windows segmented within the entire observation period. For example, if the interval is 24 hours and continuous observation is carried out for 10 days, then 10 complete time windows can be formed. If individual windows are excluded due to data loss or substandard quality, the corresponding count needs to be deducted. Finally, xm is the number of remaining windows that can participate in the calculation. When specifically executed, all windows will first be subjected to validity verification, and only the windows that meet the requirements of having complete data for all three items of leaf area, stem height, and chlorophyll concentration will be retained. After determining these windows, their number is recorded as xm.

[0139] Calculation process:

[0140] Set the number of time windows xm to 4, and record XA 0 = 40.0 cm 2 、XA 1 = 42.0 cm 2 、XA 2 = 44.5 cm 2 、XA 3 = 46.2 cm 2 、XA 4 = 47.0 cm 2 , also record the stem height XH 0 = 12.0 cm、XH 1 = 12.5 cm、XH 2 = 13.2 cm、XH 3 = 14.0 cm、XH 4 = 14.6 cm, and the chlorophyll concentration XC 0 = 1.2 mg / g、XC 1 = 1.3 mg / g、XC 2 = 1.35 mg / g、XC 3 = 1.42 mg / g、XC 4 = 1.45 mg / g, and perform the difference calculation between windows. For example, the leaf area difference between the 1st and 2nd windows is (42.0 - 40.0) = 2.0, the stem height difference is (12.5 - 12.0) = 0.5, and the chlorophyll concentration difference is (1.3 - 1.2) = 0.1. Square them and sum them up, then repeat for the 2nd to 3rd, 3rd to 4th, etc. in sequence. After accumulating these results, divide by xm = 4, and finally take the square root. The main calculation process example is as follows (only showing one segment of the calculation):

[0141] (42.0 - 40.0) 2 +(12.5 - 12.0) 2 +(1.3 - 1.2)2 = 2.0 2 + 0.5 2 + 0.1 2 = 4 + 0.25 + 0.01 = 4.26

[0142] Combining and summing up all window differences to obtain the total sum S. For example, by accumulation, S = 10.84, then

[0143]

[0144] This result indicates that when I = 1.645, it represents that during the observation periods of these adjacent windows, there are certain degrees of variation ranges in leaf area, stem height, and chlorophyll concentration. The higher the value, the more obvious the fluctuations of these indicators between time windows. A low value indicates that the changes between adjacent windows tend to be stable.

[0145] Based on the growth fluctuation intensity calculated previously, read the growth feature matrix generated in each time window and append the fluctuation intensity value to the corresponding index position of the matrix. Then, referring to an operation list containing singular spectrum analysis steps and frequency domain decomposition thresholds, first perform row vectorization processing on the fluctuation intensity data in the time series. Next, divide the sections related to the Artemisia argyi growth cycle from the frequency domain perspective, perform singular value decomposition on the fluctuation intensity of each section, and extract the principal component vectors. If the singular value of any vector decomposition exceeds the critical value range of 1.5 to 2.0 statistically obtained from the observation data of the historical stable growth period, it indicates that the fluctuation intensity of this section is higher than the general range in this frequency domain. If it does not exceed, it is considered to be at a normal level. When analyzing, it is necessary to compare the increasing and decreasing trends of the fluctuation intensity of several consecutive windows and count the contribution rates of different components after singular value decomposition. When the contribution rate of a certain component significantly increases by more than 50%, it can be considered that the fluctuation characteristics of this frequency band dominate. If multiple components share relatively balanced contribution rates, it indicates that the overall changes are dispersed on different time scales. By the above steps, the fluctuation intensities under each window are superimposed in time series and local peaks are identified. Finally, the principal components and their contribution rates of each frequency band extracted are summarized to generate the multi-scale growth analysis result.

[0146] The above is only the preferred embodiment of the present invention, and it is not intended to limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A method for monitoring the dynamic correlation between the growth status of mugwort and leafhopper infestation, characterized in that: The following steps are involved: The mugwort was photographed regularly using a multispectral camera to collect reflectance spectrum information in the near-infrared, red, and green light bands to obtain preliminary physiological parameter data; According to the preliminary physiological parameter data, assess the health status and growth trend of Artemisia argyi and generate a health assessment result; Based on the preliminary physiological parameter data, the deep learning algorithm is used for training and analysis to detect whether the plant leaves have lesions or are damaged by leafhoppers, determine the type of lesions and the degree of damage, and generate lesion damage diagnosis results; Based on the lesion invasion diagnosis result, the damaged area is marked, the damaged area ratio is calculated, and the damaged area analysis result is generated; Based on the damaged area analysis results and health assessment results, according to the damage degree and leaf health indicators, a treatment and intervention plan is formulated to generate an intervention plan formulation result; According to the results of the intervention plan formulation, a sliding time window was set to capture the short-term fluctuations and long-term trends in the growth process of Artemisia annua and generate multi-scale growth analysis results.

2. The method for monitoring the dynamic correlation between the growth status of Artemisia argyi and leafhopper infestation according to claim 1, characterized in that: The steps for obtaining the preliminary physiological parameter data are: Regularly photograph mugwort to collect reflectance spectrum information in the near-infrared, red, and green bands to obtain a spectral image dataset; Extracting spectral intensity, peaks and troughs from the spectral image data set, performing signal enhancement and background noise removal on the spectral intensity, peaks and troughs to obtain a cleaned spectral feature data set; Based on the cleaned spectral feature data set, principal component analysis is applied to identify patterns and trends in the spectral data to obtain preliminary physiological parameter data.

3. The method for monitoring the dynamic correlation between the growth status of Artemisia argyi and leafhopper infestation according to claim 1, characterized in that: The steps for obtaining the health assessment results are: Analyzing the preliminary physiological parameter data, identifying chlorophyll content, water status and growth rate, and constructing a physiological indicator data set; Based on the physiological indicator data set, the health score of Artemisia argyi is calculated using the following formula: Among them, L i represents the chlorophyll content value of the i-th element, W i represents the i-th moisture state value, G i represents the ith growth rate, n is the number of samples, and HC is the health score; Based on the health score, combined with growth data and temperature and light conditions, trend analysis is performed to obtain health assessment results.

4. The method for monitoring the dynamic correlation between the growth status of Artemisia argyi and leafhopper infestation according to claim 1, characterized in that: The steps for obtaining the lesion invasion diagnosis result are: Based on the preliminary physiological parameter data, a deep learning model is configured and trained, and a convolutional neural network is applied to optimize the recognition accuracy by identifying the pathological features and signs of leafhopper infestation in the leaf images, thereby obtaining a trained model; The trained model is used to analyze the newly collected leaf data, and the leaf lesion type and damage degree are identified and classified by parsing the image data to generate a lesion damage diagnosis result.

5. The method for monitoring the dynamic correlation between the growth status of mugwort and leafhopper infestation according to claim 1, characterized in that: The steps for obtaining the damaged area analysis result are: Based on the lesion damage diagnosis result, pixel distribution information of the damaged area is extracted, the boundary of the lesion area on the blade surface is identified, noise is removed and the contrast of the damaged area is improved, and a binary image of the damaged area is generated; According to the binary image of the damaged area, the damaged area ratio is calculated using the following formula: Among them, X j ,Y j is the coordinate of the jth damaged pixel, X c ,Y c is the center coordinate of the damaged area, m is the total number of damaged pixels, A is the total leaf area, and R is the proportion of damaged area; Based on the damaged area ratio and in combination with the geometric features of the damaged area, the diffusion trend of the lesion is analyzed, the lesion type is classified, and the damaged area analysis result is generated.

6. The method for monitoring the dynamic correlation between the growth status of mugwort and leafhopper infestation according to claim 1, characterized in that: The steps for obtaining the results of the intervention program formulation are: According to the damaged area analysis results and health assessment results, the intervention priority score is calculated using the following formula: Among them, R represents the proportion of damaged area, H represents the health score, E represents the lesion diffusion rate, C represents the complexity of the damaged area boundary, B represents the complexity of the healthy area boundary, SG represents the leaf growth rate, F represents the leaf moisture content, and P is the intervention priority score; Based on the intervention priority score, combined with the lesion type and growth environment conditions, matching treatment and intervention measures are screened to generate intervention plan formulation results.

7. The method for monitoring the dynamic correlation between the growth status of mugwort and leafhopper infestation according to claim 1, characterized in that: The steps for obtaining the multi-scale growth analysis results are: According to the results of the intervention plan, a sliding time window is set, the growth index data of Artemisia argyi during its growth process is segmented at fixed time intervals, growth characteristics including leaf area growth rate, stem height increment and chlorophyll concentration change rate are extracted, and a time window growth characteristic matrix is ​​generated; Based on the time window growth feature matrix, the growth fluctuation intensity is calculated, and the expression is: Among them, I represents the growth fluctuation intensity, XA j is the leaf area in the jth time window, XH j is the stem height of the jth time window, XC j is the chlorophyll concentration in the jth time window, and xm is the number of time windows.

8. The method for monitoring the dynamic correlation between the growth status of mugwort and leafhopper infestation according to claim 7, characterized in that: The step of obtaining the multi-scale growth analysis result also includes: applying singular spectrum analysis based on the growth fluctuation intensity to generate the multi-scale growth analysis result.

Images