Method, device and equipment for predicting occurrence of rice blast disease based on temporal hyperspectral features

Abstract

The present invention relates to plant disease monitoring technology, and discloses a method, device, equipment and storage medium for predicting the occurrence of rice blast disease based on temporal hyperspectral features. The method includes: intercepting the collected hyperspectral data set by using effective feature bands to obtain an effective hyperspectral data set; performing multi-scale feature extraction and splicing on the hyperspectral coding vector set of the effective hyperspectral data set to obtain a multi-scale feature sequence; performing temporal feature decomposition based on feature trends on the multi-scale feature sequence to obtain a first trend coding vector set and a first seasonal coding vector set; performing temporal feature decomposition based on feature trends on the hyperspectral coding vector set to obtain a second trend coding vector set and a second seasonal coding vector set, and performing an operation for predicting the occurrence of rice blast disease on each coding vector set. The present invention can realize the extraction and prediction of the occurrence characteristics of rice blast disease, and obtain an early prediction result of the occurrence of rice blast disease.

Description

Technical Field

[0001] The present invention relates to the technical field of plant disease monitoring, and in particular to a method, device, equipment and computer-readable storage medium for predicting the occurrence of rice blast disease based on temporal hyperspectral features. Background Art

[0002] Rice is the most important food crop and important industrial raw material for billions of people around the world, and is also a host for a variety of plant diseases and pests. Rice blast disease caused by the pathogenic fungus Pyricularia oryzae is one of the most devastating diseases in world rice production. The yield loss of rice caused by this disease reaches 10% - 30% every year, and in severe cases, the crop may even fail; in addition, the rice blast pathogen can also infect more than 50 other gramineous plants such as wheat and cereals. Early monitoring and warning of the occurrence of rice blast disease is crucial for restricting its epidemic outbreak.

[0003] However, traditional detection methods mainly rely on visual inspection by experts. This method is not only time-consuming and laborious, but also requires a large number of experienced professionals. In addition, under field conditions, it is difficult to detect early infections of rice blast disease when there are no visible symptoms on the leaves, resulting in a lag in the detection results of rice blast disease, which is not conducive to the prediction and scientific and accurate prevention and control of rice diseases. Summary of the Invention

[0004] The present invention provides a method, device, equipment and storage medium for predicting the occurrence of rice blast disease based on temporal hyperspectral features, and its main purpose is to achieve early prediction and warning of rice blast disease according to the early characteristics of the onset of rice blast disease.

[0005] To achieve the above object, a method for predicting the occurrence of rice blast disease based on temporal hyperspectral features provided by the present invention includes:

[0006] Real-time collect the hyperspectral data of rice in the target monitoring area to obtain a hyperspectral data set within a preset time period;

[0007] Intercept the hyperspectral data set according to a preset effective feature band to obtain an effective hyperspectral data set;

[0008] Use a pre-trained rice blast disease occurrence prediction model to perform feature encoding on the effective hyperspectral data set to obtain a hyperspectral encoding vector set;

[0009] Perform multi-scale feature extraction on the hyperspectral encoding vector set to obtain a multi-scale feature set, and splice the various multi-scale features in the multi-scale feature set to obtain a multi-scale feature sequence;

[0010] Using the pre - constructed encoder in the rice blast disease occurrence prediction model, perform time - series feature decomposition based on feature trends on the multi - scale feature sequence to obtain a first set of trend encoding vectors and a first set of seasonal encoding vectors;

[0011] Perform time - series feature decomposition based on feature trends on the hyperspectral encoding vector set to obtain a second set of trend encoding vectors and a second set of seasonal encoding vectors;

[0012] Using the pre - constructed decoder in the rice blast disease occurrence prediction model, perform rice blast disease occurrence prediction operations on the first set of trend encoding vectors, the first set of seasonal encoding vectors, the second set of trend encoding vectors, and the second set of seasonal encoding vectors to obtain the rice blast disease occurrence prediction result in the target monitoring area, where the rice blast disease occurrence prediction result includes a confirmed result and a disease development prediction result.

[0013] Optionally, before intercepting the hyperspectral data set according to the preset effective feature bands, the method further includes:

[0014] Calculate the Pearson correlation coefficient curve of the data features of each band in the hyperspectral data set for the rice blast disease according to the pre - constructed Pearson coefficient and random forest algorithm;

[0015] Judge whether early features of rice blast disease occurrence are detected in the historical hyperspectral data set before the hyperspectral data set;

[0016] When the early features are detected in the historical hyperspectral data set, select a preset first threshold as the interception threshold;

[0017] When the early features are not detected in the historical hyperspectral data set, select a preset second threshold as the interception threshold;

[0018] Screen the bands in the Pearson correlation coefficient curve whose correlation coefficients are greater than the interception threshold to obtain the effective feature bands.

[0019] Optionally, calculating the Pearson correlation coefficient curve of the data features of each band in the hyperspectral data set for the rice blast disease according to the pre - constructed Pearson coefficient and random forest algorithm includes:

[0020] Obtain the pre - constructed decision tree forest;

[0021] According to the pre - constructed OOB error analysis algorithm, calculate the out - of - bag data corresponding to the target decision tree in the decision tree forest and obtain the number of misclassified samples in the out - of - bag data;

[0022] Perform feature perturbation based on the target wavelength on the target decision tree to obtain updated out-of-bag data, and calculate the number of updated error samples in the updated out-of-bag data;

[0023] Perform iterative operations of the above-mentioned feature perturbation on each target decision tree for a preset number of times to obtain the target wavelength error quantity for the target wavelength;

[0024] Calculate the importance score of the target wavelength according to the target wavelength error quantity and the number of error samples. The importance score of the target wavelength is expressed as:

[0025]

[0026] In the formula, represents the target wavelength of the hyperspectral data, represents the target wavelength of the importance score of the hyperspectral data, represents the total number of decision trees in the decision tree forest, represents the th decision tree, represents the decision tree for the target wavelength of the target wavelength error quantity, represents the decision tree of the number of error samples;

[0027] According to the pre-constructed k-fold cross-validation algorithm and the importance score of the target wavelength, screen a preset number of optimal decision tree sets from the decision tree forest, and obtain the random forest feature wavelength set corresponding to the optimal decision tree set;

[0028] According to the Pearson coefficient, calculate the correlation of the hyperspectral data of each wavelength in the random forest feature wavelength set to obtain a Pearson correlation coefficient curve.

[0029] Optionally, before using the pre-trained rice blast disease occurrence prediction model to perform feature encoding on the effective hyperspectral data set, the method further includes:

[0030] Obtain a pre-constructed rice blast disease occurrence prediction model, and obtain a sample set recorded according to preset test conditions, and randomly group the sample set according to a preset training and testing ratio to obtain a training set and a testing set;

[0031] Extract a target sample from the training set in turn, and use the rice blast disease occurrence prediction model to predict the occurrence of rice blast disease for the target sample to obtain a primary prediction result;

[0032] Calculate the loss value between the true label corresponding to the target sample and the primary prediction result according to the pre-constructed cross-entropy loss algorithm;

[0033] Minimize the loss value to obtain the network model parameters when the loss value is minimized;

[0034] According to the pre-constructed K-fold cross-validation algorithm, calculate the mean of the network model parameters corresponding to each target sample, and perform an inverse network update operation on the mean of the network parameters to obtain an updated rice blast disease occurrence prediction model;

[0035] Using the test set, calculate the tolerance accuracy of the updated rice blast disease occurrence prediction model according to a preset tolerance value, where the tolerance accuracy is expressed as:

[0036]

[0037] In the formula, represents the tolerance accuracy, represents the primary prediction result of the th test sample, represents the true label of the th test sample, represents the tolerance value, represents the number of samples in the test set; represents: when the absolute value of the difference between the primary prediction result and the true label is less than the tolerance value, it is recorded as 1, otherwise, it is recorded as 0;

[0038] Judge whether the tolerance accuracy is greater than a preset qualified threshold;

[0039] When the tolerance accuracy is less than or equal to the qualified threshold, return to the step of randomly grouping the sample set according to the preset training and testing ratio, and reassign the training set and the test set to train the rice blast prediction model;

[0040] When the tolerance accuracy is greater than the qualified threshold, stop the training process to obtain a trained rice blast disease occurrence prediction model.

[0041] Optionally, the obtaining of the sample set recorded according to the preset experimental conditions includes:

[0042] Obtain pre-constructed susceptible varieties and resistant varieties, and plant the susceptible varieties and resistant varieties according to the preset cultivation conditions until the rice seedlings grow to 3.5 - 4 leaf ages;

[0043] Spray inoculation was carried out on the susceptible variety and the resistant variety until the rice seedlings of the susceptible variety and the resistant variety were completely diseased;

[0044] The reflection spectra at the 1 / 3, 1 / 2, and 2 / 3 of the base of the rice leaves of the diseased susceptible variety and resistant variety were observed for a preset number of days, and the observed data were obtained. According to the pre-constructed international rice blast seedling blast grading standard, the disease incidence of the susceptible variety and the resistant variety was graded to obtain artificial grading labels;

[0045] The observed data and artificial grading labels of the susceptible variety and the resistant variety were recorded to obtain a sample set.

[0046] Optionally, the multi-scale feature sequence is decomposed into a first trend coding vector set and a first seasonal coding vector set based on feature trends by using the pre-constructed encoder in the rice blast disease occurrence prediction model, including:

[0047] Using the pre-constructed encoder in the rice blast disease occurrence prediction model, according to the preset delay time, the autocorrelation coefficient sequence between the multi-scale feature sequence and the delay time is calculated, and the autocorrelation coefficient is expressed as:

[0048]

[0049] In the formula, represents the autocorrelation coefficient sequence between the multi-scale feature sequence and the delay time, represents the delay time, represents the multi-scale feature sequence, represents the mean value of the multi-scale feature sequence;

[0050] Performing a one-dimensional convolutional layer convolution operation on the multi-scale feature sequence and the autocorrelation coefficient sequence to obtain a trend feature, and the trend feature is expressed as:

[0051]

[0052] In the formula, represents the trend feature, represents the weighted calculation result of the multi-scale feature sequence and the autocorrelation coefficient sequence;

[0053] Subtracting the trend feature from the multi-scale feature sequence to obtain a seasonal feature, and the seasonal feature is expressed as:

[0054]

[0055] In the formula, represents the seasonal feature.

[0056] Optionally, perform rice blast disease occurrence prediction operations on the first trend coding vector set, the first seasonal coding vector set, the second trend coding vector set, and the second seasonal coding vector set by using the pre-constructed decoder in the rice blast disease occurrence prediction model to obtain the rice blast disease occurrence prediction result in the target monitoring area. The rice blast disease occurrence prediction result includes a confirmed result and a disease development prediction result, including:

[0057] Use the pre-constructed decoder in the rice blast disease occurrence prediction model to calculate the autocorrelation coefficient weights for the second seasonal coding vector set to obtain a second autocorrelated seasonal feature vector set;

[0058] Perform time series feature decomposition on the second autocorrelated seasonal feature vector set to obtain a third trend feature set and a third seasonal feature set;

[0059] Fuse the third trend feature set with the second trend coding vector set to obtain a fourth trend feature set;

[0060] Calculate the autocorrelation coefficient weights for the weighted result between the second autocorrelated seasonal feature vector set and the first seasonal coding vector set to obtain a third autocorrelated seasonal feature vector set;

[0061] Perform time series feature decomposition on the third autocorrelated seasonal feature vector set to obtain a fifth trend feature set and a fourth seasonal feature set;

[0062] Fuse the fourth trend feature set and the fifth trend feature set to obtain a sixth trend feature set;

[0063] Perform rice blast disease occurrence prediction based on the fourth seasonal feature set and the sixth trend feature set to obtain the rice blast disease prediction result in the target monitoring area.

[0064] To solve the above problems, the present invention also provides a rice blast disease prediction device based on time series hyperspectral features. The device includes:

[0065] A data acquisition and dimensionality reduction module, configured to collect hyperspectral data of rice in the target monitoring area in real time to obtain a hyperspectral data set within a preset time period, and intercept the hyperspectral data set according to preset effective feature bands to obtain an effective hyperspectral data set;

[0066] A feature extraction module, configured to use a pre-trained rice blast disease occurrence prediction model to perform feature encoding on the effective hyperspectral data set to obtain a hyperspectral encoding vector set, perform multi-scale feature extraction on the hyperspectral encoding vector set to obtain a multi-scale feature set, and splice the various multi-scale features in the multi-scale feature set to obtain a multi-scale feature sequence;

[0067] A feature decomposition module, configured to use an encoder pre-constructed in the rice blast disease occurrence prediction model to perform time-series feature decomposition based on feature trends on the multi-scale feature sequence to obtain a first trend encoding vector set and a first seasonal encoding vector set, and perform time-series feature decomposition based on feature trends on the hyperspectral encoding vector set to obtain a second trend encoding vector set and a second seasonal encoding vector set;

[0068] A rice blast disease occurrence prediction module, configured to use a decoder pre-constructed in the rice blast disease occurrence prediction model to perform rice blast disease occurrence prediction operations on the first trend encoding vector set, the first seasonal encoding vector set, the second trend encoding vector set, and the second seasonal encoding vector set to obtain a rice blast disease occurrence prediction result in the target monitoring area, where the rice blast disease occurrence prediction result includes a confirmed result and a disease development prediction result.

[0069] To solve the above problems, the present invention further provides an electronic device, where the electronic device includes:

[0070] At least one processor; and,

[0071] A memory communicatively connected to the at least one processor; wherein,

[0072] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the above-mentioned rice blast disease occurrence prediction method based on time-series hyperspectral features.

[0073] To solve the above problems, the present invention further provides a computer-readable storage medium, where at least one computer program is stored in the computer-readable storage medium, and the at least one computer program is executed by a processor in an electronic device to implement the above-mentioned rice blast disease occurrence prediction method based on time-series hyperspectral features.

[0074] In the embodiments of the present invention, first, a hyperspectral data set is obtained. Then, the hyperspectral data set is intercepted according to the effective feature bands calculated by the Pearson coefficient and the random forest algorithm, so as to achieve data dimensionality reduction and obtain an effective hyperspectral data set. Then, on the one hand, the present invention extracts features through multi-scale features, and on the other hand, directly extracts features to form two feature sequences with different scales. Then, both are decomposed into trend features and seasonal features through time-series feature decomposition based on feature extraction. The trend features corresponding to the two different scales are spliced and fused, and the seasonal features are spliced and fused. According to the finally accumulated trend features and seasonal features, the prediction of the occurrence of rice blast disease is completed. Therefore, a method, device, equipment and storage medium for predicting the occurrence of rice blast disease based on time-series hyperspectral features provided by the embodiments of the present invention can realize the early prediction and forecasting of rice blast according to the early features of the occurrence of rice blast disease. Description of the Drawings

[0075] Figure 1 It is a schematic flowchart of a method for predicting the occurrence of rice blast disease based on time-series hyperspectral features provided by an embodiment of the present invention;

[0076] Figure 2 It is a schematic diagram of the Pearson correlation coefficient curve in the method for predicting the occurrence of rice blast disease based on time-series hyperspectral features provided by an embodiment of the present invention;

[0077] Figure 3 It is the overall structure diagram of the solution in the method for predicting the occurrence of rice blast disease based on time-series hyperspectral features provided by an embodiment of the present invention;

[0078] Figure 4 It is a functional module diagram of a device for predicting the occurrence of rice blast disease based on time-series hyperspectral features provided by an embodiment of the present invention;

[0079] Figure 5 It is a schematic structural diagram of an electronic device for implementing the method for predicting the occurrence of rice blast disease based on time-series hyperspectral features provided by an embodiment of the present invention.

[0080] The realization, functional characteristics and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. Detailed Embodiments

[0081] 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.

[0082] The embodiments of the present application provide a method for predicting the occurrence of rice blast diseases based on temporal hyperspectral features. In the embodiments of the present application, the execution subject of the method for predicting the occurrence of rice blast diseases based on temporal hyperspectral features includes, but is not limited to, at least one of electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided by the embodiments of the present application. In other words, the method for predicting the occurrence of rice blast diseases based on temporal hyperspectral features can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to: a single server, a server cluster, a cloud server, or a cloud server cluster, etc. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, Content Delivery Network (CDN), and big data and artificial intelligence platforms.

[0083] Referring to Figure 1 As shown, it is a schematic flowchart of a method for predicting the occurrence of rice blast diseases based on temporal hyperspectral features provided by an embodiment of the present invention. In this embodiment, the method for predicting the occurrence of rice blast diseases based on temporal hyperspectral features includes:

[0084] S1. Real-time collect the hyperspectral data of rice in the target monitoring area to obtain a set of hyperspectral data within a preset time period.

[0085] It is reported that in the initial stage of rice blast disease, the chemical components inside the rice leaves will change, resulting in changes in the absorption rate and reflectance of the diseased leaves for different spectral bands. Therefore, the present invention detects the reflection and absorption conditions of light of different wavelengths received by crops through hyperspectral data, and obtains rich spectral information.

[0086] In the embodiments of the present invention, it is expected to identify and predict rice blast diseases based on the early characteristics of rice blast diseases to control their spread and epidemic. Therefore, the present invention needs to continuously collect the hyperspectral data of rice in the target monitoring area and save the hyperspectral data within a preset time period, such as 1-7 days, to obtain a set of hyperspectral data.

[0087] Specifically, in the embodiments of the present invention, the hyperspectral data of rice leaves is measured by a pre-built FieldSpec4 (350-2500nm) portable ground object spectrometer, and the set of hyperspectral data within 7 days is retained.

[0088] S2. Intercept the set of hyperspectral data according to a preset effective feature band to obtain a set of effective hyperspectral data.

[0089] In the embodiments of the present invention, the effective characteristic band refers to the band in which the spectral data changes most prominently due to the change in the structural performance of the leaf caused by the infection of Magnaporthe oryzae.

[0090] In the embodiments of the present invention, the data with wavelengths from 350 to 2500 are collected by a spectrometer. The amount of data is too large, so it is necessary to use the effective characteristic band for data dimensionality reduction to obtain an effective hyperspectral data set.

[0091] Specifically, in the embodiments of the present invention, before intercepting the hyperspectral data set according to the preset effective characteristic band, the method further includes:

[0092] Calculating the Pearson correlation coefficient curve of the data characteristics of each band in the hyperspectral data set for the rice blast disease according to the pre-constructed Pearson coefficient and random forest algorithm;

[0093] Judging whether early characteristics of the occurrence of rice blast disease are detected in the historical hyperspectral data set before the hyperspectral data set;

[0094] When the early characteristics are detected in the historical hyperspectral data set, a preset first threshold is selected as the interception threshold;

[0095] When the early characteristics are not detected in the historical hyperspectral data set, a preset second threshold is selected as the interception threshold;

[0096] Screening the bands in the Pearson correlation coefficient curve with a correlation coefficient greater than the interception threshold to obtain the effective characteristic band.

[0097] Among them, the Pearson coefficient, also known as the Pearson product-moment correlation coefficient, is an index commonly used in statistics to quantify the degree of linear relationship and correlation direction between two continuous variables (Pearson). The Pearson correlation coefficient is more stable and not affected by the sample size, and its value range is between -1 and 1. 1 represents a perfect positive correlation, -1 represents a perfect negative correlation, and 0 represents no linear correlation.

[0098] Furthermore, the random forest algorithm is an effective data dimensionality reduction and model optimization method, which is used to randomly permute the features and test the out-of-bag data generated after the feature permutation by the random forest. The higher the importance of the feature, the greater the change in the prediction error rate of the random forest.

[0099] The present invention can calculate the Pearson correlation coefficient curve of the data characteristics of each band in the hyperspectral data set for the rice blast disease according to the combination of the Pearson coefficient and the random forest algorithm.

[0100] Specifically, in the embodiments of the present invention, calculating the Pearson correlation coefficient curve of the data features of each band in the hyperspectral data set for the rice blast disease according to the pre-constructed Pearson coefficient and random forest algorithm includes:

[0101] Obtain a pre-constructed decision tree forest;

[0102] According to the pre-constructed OOB error analysis algorithm, calculate the out-of-bag data corresponding to the target decision tree in the decision tree forest, and obtain the number of misclassified samples in the out-of-bag data;

[0103] Perform feature perturbation on the target decision tree based on the target wavelength to obtain updated out-of-bag data, and calculate the updated number of misclassified samples in the updated out-of-bag data;

[0104] Perform iterative operations of the above-mentioned feature perturbation on each target decision tree for a preset number of times to obtain the target wavelength error quantity for the target wavelength;

[0105] According to the target wavelength error quantity and the number of misclassified samples, calculate the importance score of the target wavelength, and the importance score of the target wavelength is expressed as:

[0106]

[0107] In the formula, represents the hyperspectral data of the target wavelength ; represents the importance score of the hyperspectral data of the target wavelength ; represents the total number of decision trees in the decision tree forest, represents the th decision tree, represents the decision tree for the target wavelength of the target wavelength error quantity, represents the decision tree of the number of misclassified samples;

[0108] According to the k-fold cross-validation method and the importance score of the target wavelength, screen a preset number of optimal decision tree sets from the decision tree forest, and obtain the random forest feature wavelength set corresponding to the optimal decision tree set;

[0109] According to the Pearson coefficient, perform correlation calculation on the hyperspectral data of each wavelength in the random forest feature wavelength set to obtain the Pearson correlation coefficient curve.

[0110] Among them, the OOB error analysis is a method for evaluating the performance of the random forest (RF) algorithm, which uses the samples not used by a specific decision tree during training to evaluate the generalization ability of the model. When constructing each tree, samples are randomly selected for training, and some samples are not selected. These unselected samples are called "out-of-bag samples".

[0111] Specifically, in the embodiments of the present invention, each spectral data corresponds to a wavelength band of 350 nm to 2500 nm, and each spectrum corresponds to 2151 wavelengths, that is, 2151 features. The random forest (RF) algorithm is used to evaluate according to the error rate before and after the replacement of each feature, so as to obtain the characteristic wavelengths with higher importance.

[0112] The present invention performs data dimensionality reduction on the data based on the RF algorithm and finds the appropriate number of decision trees based on the k-fold cross-validation method. Set k to 5, specify the number of decision trees from 10 to 2150, perform cross-validation evaluation, and finally obtain that the optimal number of decision trees is 80. Finally, according to the optimal decision tree, 596 characteristic bands are extracted to obtain the random forest characteristic wavelength set. Then, the Pearson correlation coefficient is used to perform data dimensionality reduction on the data to obtain the Pearson correlation coefficient curve, as Figure 2 shown.

[0113] According to Figure 2 the Pearson correlation coefficient curve shown, the correlation coefficients of most bands are low, and the correlation coefficients of a few bands are high. When the bands with r > 0.5 are regarded as characteristic bands, most of the characteristic bands are distributed between 573 nm and 711 nm, and there are 96 characteristic bands in total. When the bands with r > 0.6 are regarded as characteristic bands, all the characteristic bands are distributed between 692 nm and 705 nm, totaling 14.

[0114] Therefore, in the embodiments of the present invention, it is judged whether early characteristics of rice blast disease occurrence are detected in the historical hyperspectral data set before the hyperspectral data set.

[0115] When the early characteristics are not detected, more comprehensive detection is required. Therefore, the bands with r > 0.5 are regarded as characteristic bands, which is easier to discover the early characteristics of rice blast disease occurrence. When the early characteristics are detected, the bands with r > 0.6 can be regarded as characteristic bands, so as to improve the detection accuracy of rice blast disease occurrence.

[0116] In the embodiments of the present invention, the effective characteristic bands can effectively reduce the dimensionality of the spectral data and retrieve the data calculation amount.

[0117] S3. Use the pre-trained rice blast disease occurrence prediction model to perform feature encoding on the effective hyperspectral data set to obtain a hyperspectral encoding vector set.

[0118] In an embodiment of the present invention, the rice blast disease occurrence prediction model is a neural network model with multi-scale convolution based on Autoformer and having feature extraction time series decomposition, which is used to decompose hyperspectral data into seasonal features and trend features under multi-scale features, so as to predict the confirmed identification and disease development of rice blast disease occurrence.

[0119] In an embodiment of the present invention, the effective information in the set of effective hyperspectral data is calculated, and indexes such as spectral reflectance and spectral index are combined and quantized for coding to obtain a set of hyperspectral coding vectors.

[0120] Specifically, in an embodiment of the present invention, before using the pre-trained rice blast disease occurrence prediction model to perform feature coding on the set of effective hyperspectral data, the method further includes:

[0121] Obtain a pre-constructed rice blast disease occurrence prediction model, and obtain a sample set recorded according to preset test conditions, and randomly group the sample set according to a preset training and testing ratio to obtain a training set and a testing set;

[0122] Sequentially extract a target sample from the training set, and use the rice blast disease occurrence prediction model to predict the occurrence of rice blast disease for the target sample to obtain a primary prediction result;

[0123] According to a pre-constructed cross-entropy loss algorithm, calculate the loss value between the true label corresponding to the target sample and the primary prediction result;

[0124] Minimize the loss value to obtain the network model parameters when the loss value is the smallest;

[0125] According to a pre-constructed K-fold cross-validation algorithm, calculate the mean value of the network model parameters corresponding to each target sample, and perform an inverse network update operation on the mean value of the network model parameters to obtain an updated rice blast disease occurrence prediction model;

[0126] Use the testing set to calculate the tolerance accuracy of the updated rice blast disease occurrence prediction model according to a preset tolerance value, where the tolerance accuracy is expressed as:

[0127]

[0128] In the formula, represents the tolerance accuracy, represents the th primary prediction result of the test sample, represents the th true label of the test sample, represents the tolerance value, represents the number of samples in the test set; indicates that when the primary prediction result and the true label the absolute value of the difference between them is less than the tolerance value, it is recorded as 1, otherwise, it is recorded as 0;

[0129] judge whether the tolerance accuracy rate is greater than a preset qualified threshold;

[0130] When the tolerance accuracy rate is less than or equal to the qualified threshold, return to the above step of randomly grouping the sample set according to the preset training and testing ratio, and re - allocate the training set and the test set to train the rice blast prediction model;

[0131] When the tolerance accuracy rate is greater than the qualified threshold, stop the training process and obtain the trained rice blast prediction model.

[0132] Specifically, referring to Figure 3 As shown, in the embodiment of the present invention, the feature extraction stage of the rice blast disease occurrence prediction model includes two branches of multi - scale feature convolution and direct time - series decomposition, and the results of the two branches are merged in the decoder. The decoder also decomposes the data features into trend features and seasonal features, so as to finally predict and obtain the recognition result of the occurrence of rice blast disease.

[0133] Further, in the embodiment of the present invention, the obtaining of the sample set recorded according to the preset test conditions includes:

[0134] Obtain pre - constructed susceptible varieties and resistant varieties, and plant the susceptible varieties and resistant varieties according to the preset cultivation conditions until the rice seedlings grow to 3.5 - 4 leaf ages;

[0135] Spray - inoculate the susceptible varieties and resistant varieties until the rice seedlings of the susceptible varieties and resistant varieties are completely diseased;

[0136] Observe the reflection spectra at the 1 / 3, 1 / 2, and 2 / 3 of the base of the rice leaves of the diseased susceptible varieties and resistant varieties for a preset number of days, obtain the observation data, and classify the disease conditions of the susceptible varieties and resistant varieties according to the pre - constructed international rice blast seedling blast grading standard to obtain the artificial grading labels;

[0137] Record the observation data and artificial grading labels of the susceptible varieties and resistant varieties to obtain the sample set.

[0138] Specifically, in the embodiments of the present invention, one susceptible and one resistant rice variety were selected respectively. One susceptible control variety, Lijiangxintuanheigu (LTH), and one resistant variety, NIL-e1. Among them, NIL-e1 is a near-isogenic line resistant to rice blast containing a single resistant gene developed by the Institute of Plant Protection, Guangdong Academy of Agricultural Sciences, with Lijiangxintuanheigu as the background. By adding resistance in the present invention, hyperspectral data when the pathogen infects the resistant variety can be obtained, which can provide data support for predicting whether rice blast disease will prevail and break out subsequently.

[0139] Then, after the rice seeds were germinated, they were sown in pots with a diameter of 12.2 cm by hill-drop method, with three hills per pot and 5-6 plants per hill. A total of 50 pots of rice were planted, including 40 pots of susceptible rice and 10 pots of resistant rice. The rice seedlings were cultivated under preset cultivation conditions. For example, ammonium sulfate was applied when the seedlings grew to the one-leaf and one-heart stage, and the application rate per pot was 0.5 g. A total of 3 fertilizations were required before inoculation. When the rice seedlings grew to 3.5-4 leaf ages, artificial spray inoculation was carried out.

[0140] In the embodiments of the present invention, hyperspectral measurement of the rice was carried out on the day before inoculation, and then for the next 6 consecutive days, starting at 2 pm every day, the rice leaves were measured in sequence by number to obtain the observed data.

[0141] Among them, in the specific measurement process, considering that the rice samples were all in the seedling stage and the leaf size was small, the present invention carried out detailed label marking on the three hills of rice in each pot of samples. During the measurement, multiple leaves were unfolded simultaneously to ensure the accuracy and representativeness of the measurement. Three reflectance spectra were obtained at the 1 / 3, 1 / 2, and 2 / 3 of the base of the rice leaf respectively to comprehensively reflect the spectral characteristics of the leaf. During the spectral acquisition process, a leaf clip was used to fix the rice leaf to ensure the stability and accuracy of the measurement. All spectral data collected from the same rice plant were averaged to represent the average reflectance of the leaf at that location, thereby minimizing the random error during the measurement process to the greatest extent.

[0142] Then, according to the international rice blast grading standard: Grade 0: no lesions; Grade 1: only pinhead-sized lesions; Grade 2: slightly larger brown punctate lesions, 0.50 ≤ diameter < 1.00 mm; Grade 3: circular to oval gray lesions with a brown edge, 1.00 ≤ diameter < 2.00 mm; Grade 4: oval or narrow spindle-shaped lesions between two veins, lesion area < 2% of the leaf area... Grade 7.... Through manual scoring, an artificial scoring label was obtained. Finally, through the corresponding relationship, a sample set was recorded.

[0143] Furthermore, in the embodiments of the present invention, a rice blast disease occurrence prediction model is trained using a sample set. During the training process, the K-fold cross-validation algorithm is adopted for training, and the accuracy of the model is controlled by means of tolerance accuracy to obtain the finally trained rice blast disease occurrence prediction model. Among them, the K-fold cross-validation algorithm is a commonly used model evaluation method, aiming to more reliably evaluate the performance of machine learning models on new data. The basic idea is to divide the data set into K subsets (folds) of equal size, and then perform K times of training and validation.

[0144] S4. Perform multi-scale feature extraction on the hyperspectral encoding vector set to obtain a multi-scale feature set, and splice the multi-scale features in the multi-scale feature set to obtain a multi-scale feature sequence.

[0145] According to Figure 3 In the overall structure diagram of the shown scheme, Input Sequence is the input, Prediction is the prediction result, AdaptiveDecomp is the time-series feature decomposition based on feature extraction, Embedding represents quantization encoding, MnltiSealeConv represents multi-scale feature convolution, Concat represents feature splicing, Trend represents trend feature, Seasonal represents seasonal feature, represents the autocorrelation coefficient configuration, FeedForward represents the feed-forward network, Encoder represents the encoder, and Decoder represents the decoder.

[0146] According to Figure 3 In the part of feature extraction of the present invention embodiment, by introducing a multi-scale convolution block to simultaneously extract features of different scales, the understanding ability and prediction performance of the model for time-series data are improved.

[0147] Figure 3 In the multi-scale convolution part of, it includes multiple convolution units of different scales. These convolution units have convolution kernels of different lengths (such as lengths of 3, 5, 7, etc.). They act on the input features simultaneously, and extract short-term and long-term patterns in the time-series data through convolution operations of different scales. The output of the multi-scale convolution block is the weighted sum of the outputs of all convolution units, as shown in the formula:

[0148]

[0149] In the formula, is the final output of the multi-scale convolution block, is the output of the k-th scale convolution unit (where (k) is the index of the convolution kernel size, such as 3, 5, 7, etc.), is the weight of the 1x1 convolution layer. Here It means concatenating the outputs of convolutional units of different scales in the channel dimension.

[0150] S5. Using the pre-constructed encoder in the rice blast disease occurrence prediction model, perform temporal feature decomposition based on feature trends on the multi-scale feature sequence to obtain a first set of trend encoding vectors and a first set of seasonal encoding vectors.

[0151] In the embodiments of the present invention, the temporal feature decomposition based on feature trends can gradually extract trend features from the predicted intermediate hidden variables. Compared with the decomposition module of Autoformer that uses a moving average method to decompose time series, the present invention introduces a convolutional neural network to adaptively extract trend features from data and can capture more complex patterns.

[0152] Specifically, in the embodiments of the present invention, the step of using the pre-constructed encoder in the rice blast disease occurrence prediction model to perform temporal feature decomposition based on feature trends on the multi-scale feature sequence to obtain a first set of trend encoding vectors and a first set of seasonal encoding vectors includes:

[0153] Using the pre-constructed encoder in the rice blast disease occurrence prediction model, according to a preset delay time, calculate the autocorrelation coefficient sequence between the multi-scale feature sequence and the delay time. The autocorrelation coefficient is expressed as:

[0154]

[0155] In the formula, represents the autocorrelation coefficient sequence between the multi-scale feature sequence and the delay time, represents the delay time, represents the multi-scale feature sequence, represents the mean value of the multi-scale feature sequence;

[0156] Perform a one-dimensional convolutional layer convolution operation on the multi-scale feature sequence and the autocorrelation coefficient sequence to obtain trend features. The trend features are expressed as:

[0157]

[0158] In the formula, represents the trend features, represents the weighted calculation result of the multi-scale feature sequence and the autocorrelation coefficient sequence;

[0159] Subtract the trend features from the multi-scale feature sequence to obtain seasonal features. The seasonal features are expressed as:

[0160]

[0161] In the formula, Represents seasonal characteristics.

[0162] Specifically, the present invention uses one-dimensional convolution to smooth the periodic fluctuations and highlight the long-term trend. For an input sequence of length , the processing process is as follows: :

[0163] Trend extraction: Use a one-dimensional convolutional layer to perform a convolution operation on the transformed input data to extract the trend component of the time series:

[0164]

[0165] Seasonal component calculation: Obtain the seasonal component by subtracting the extracted trend component from the original input data:

[0166]

[0167] Summarizing the above process, the module can be expressed as:

[0168]

[0169] The module significantly enhances the effect of time series decomposition by introducing the adaptive ability of the convolutional neural network.

[0170] In addition, the encoder of the present invention also introduces an autocorrelation mechanism to capture the periodic characteristics in the data by calculating the similarity between the sequence and its delayed sequence, thereby obtaining the weighted calculation results of the multi-scale feature sequence and the autocorrelation coefficient sequence , thereby further improving the effect of time series decomposition.

[0171] S6. Perform time series feature decomposition based on feature trends on the hyperspectral encoding vector set to obtain a second trend encoding vector set and a second seasonal encoding vector set.

[0172] In the embodiment of the present invention, according to Figure 3 in the AdaptiveDecomp part, it exists in both branches of feature extraction and has the same function.

[0173] The above S5 process refers to the time series feature decomposition of multi-scale features. Here, the time series feature decomposition is directly performed on the original hyperspectral encoding vector set to obtain a second trend encoding vector set and a second seasonal encoding vector set. Among them, the second trend encoding vector set and the second seasonal encoding vector set have a feature complementary effect with the first trend encoding vector set and the first seasonal encoding vector set to improve the accuracy of subsequent model prediction.

[0174] S7. Use the pre - constructed decoder in the rice blast disease occurrence prediction model to perform rice blast disease occurrence prediction operations on the first trend coding vector set, the first season coding vector set, the second trend coding vector set, and the second season coding vector set, and obtain the rice blast disease occurrence prediction result in the target monitoring area. The rice blast disease occurrence prediction result includes a confirmed result and a disease development prediction result.

[0175] Specifically, in the embodiment of the present invention, the use of the pre - constructed decoder in the rice blast disease occurrence prediction model to perform rice blast disease occurrence prediction operations on the first trend coding vector set, the first season coding vector set, the second trend coding vector set, and the second season coding vector set, and obtain the rice blast disease occurrence prediction result in the target monitoring area includes:

[0176] Use the pre - constructed decoder in the rice blast disease occurrence prediction model to calculate the autocorrelation coefficient weights for the second season coding vector set, and obtain the second autocorrelated seasonal feature vector set;

[0177] Perform time - series feature decomposition on the second autocorrelated seasonal feature vector set to obtain a third trend feature set and a third season feature set;

[0178] Fuse the third trend feature set with the second trend coding vector set to obtain a fourth trend feature set;

[0179] Calculate the autocorrelation coefficient weights for the weighted result between the second autocorrelated seasonal feature vector set and the first season coding vector set to obtain a third autocorrelated seasonal feature vector set;

[0180] Perform time - series feature decomposition on the third autocorrelated seasonal feature vector set to obtain a fifth trend feature set and a fourth season feature set;

[0181] Fuse the fourth trend feature set and the fifth trend feature set to obtain a sixth trend feature set;

[0182] Perform rice blast disease occurrence prediction based on the fourth season feature set and the sixth trend feature set to obtain the rice blast disease occurrence prediction result in the target monitoring area.

[0183] In the embodiment of the present invention, refer to Figure 3For the decoder part, first calculate the autocorrelation coefficient weights for the second seasonal coding vector set, then perform the time series feature decomposition operation, and transfer them to the trend feature and seasonal feature respectively. The obtained seasonal feature can be fused with the first trend coding vector set and the first seasonal coding vector set obtained in the encoder, thus starting a new round of the above operations. Finally, after the seasonal feature and the trend feature are accumulated, the prediction of rice blast disease occurrence is performed to obtain the prediction result of rice blast disease occurrence. Since the data of the pathogen infecting the resistant variety is learned in the rice blast disease occurrence prediction model, not only the diagnosis result can be obtained, but also the result of the disease development can be predicted.

[0184] In the embodiment of the present invention, first, a hyperspectral data set is obtained. Then, the hyperspectral data set is intercepted according to the effective feature bands calculated by the Pearson coefficient and the random forest algorithm, so as to realize data dimensionality reduction and obtain an effective hyperspectral data set. Then, on the one hand, the present invention extracts features through multi-scale features, and on the other hand, directly extracts features to form two feature sequences of different scales. Then, both are decomposed into trend features and seasonal features through time series feature decomposition based on feature extraction, and the trend features corresponding to the two different scales are spliced and fused, and the seasonal features are spliced and fused. According to the finally accumulated trend features and seasonal features, the prediction of rice blast disease occurrence is completed. Therefore, a method for predicting the occurrence of rice blast disease based on time series hyperspectral features provided by the embodiment of the present invention can realize the early prediction and forecast of rice blast according to the early features of the occurrence of rice blast disease in rice.

[0185] As Figure 4 shown, it is a functional module diagram of a rice blast prediction device based on time series hyperspectral features provided by an embodiment of the present invention.

[0186] The rice blast disease occurrence prediction device 100 based on time series hyperspectral features of the present invention can be installed in an electronic device. According to the functions achieved, the rice blast disease occurrence prediction device 100 based on time series hyperspectral features can include a data acquisition and dimensionality reduction module 101, a feature extraction module 102, a feature decomposition module 103, and a rice blast disease occurrence prediction module 104. The modules of the present invention can also be referred to as units, which refer to a series of computer program segments that can be executed by the processor of an electronic device and can complete fixed functions, and are stored in the memory of the electronic device.

[0187] In this embodiment, the functions of each module / unit are as follows:

[0188] The data acquisition and dimensionality reduction module 101 is used to collect the hyperspectral data of rice in the target monitoring area in real time to obtain a hyperspectral data set within a preset time period, and intercept the hyperspectral data set according to the preset effective feature bands to obtain an effective hyperspectral data set;

[0189] The feature extraction module 102 is configured to perform feature encoding on the effective hyperspectral data set by using a pre-trained rice blast disease occurrence prediction model to obtain a hyperspectral encoding vector set, perform multi-scale feature extraction on the hyperspectral encoding vector set to obtain a multi-scale feature set, and splice the respective multi-scale features in the multi-scale feature set to obtain a multi-scale feature sequence;

[0190] The feature decomposition module 103 is configured to perform time-series feature decomposition based on feature trends on the multi-scale feature sequence by using a pre-constructed encoder in the rice blast disease occurrence prediction model to obtain a first trend encoding vector set and a first seasonal encoding vector set, and perform time-series feature decomposition based on feature trends on the hyperspectral encoding vector set to obtain a second trend encoding vector set and a second seasonal encoding vector set;

[0191] The rice blast disease occurrence prediction module 104 is configured to perform a rice blast disease occurrence prediction operation on the first trend encoding vector set, the first seasonal encoding vector set, the second trend encoding vector set, and the second seasonal encoding vector set by using a pre-constructed decoder in the rice blast disease occurrence prediction model to obtain a rice blast disease occurrence prediction result in the target monitoring area, where the rice blast disease occurrence prediction result includes a confirmed result and a disease development prediction result.

[0192] Specifically, each module in the rice blast disease prediction device 100 based on time-series hyperspectral features in the embodiment of the present application adopts the same technical means as those in the above Figures 1 to 3 described rice blast disease prediction method based on time-series hyperspectral features and can produce the same technical effects, which will not be elaborated here.

[0193] As Figure 5 shown, it is a schematic structural diagram of an electronic device 1 for implementing a rice blast disease occurrence prediction method based on time-series hyperspectral features provided by an embodiment of the present invention.

[0194] The electronic device 1 may include a processor 10, a memory 11, a communication bus 12, and a communication interface 13, and may further include a computer program stored in the memory 11 and executable on the processor 10, such as a rice blast disease occurrence prediction program based on time-series hyperspectral features.

[0195] Among them, in some embodiments, the processor 10 may be composed of an integrated circuit. For example, it may be composed of a single packaged integrated circuit, or may be composed of multiple packaged integrated circuits with the same or different functions, including a combination of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 10 is the control core (Control Unit) of the electronic device 1, connecting various components of the entire electronic device through various interfaces and circuits. By running or executing programs or modules stored in the memory 11 (such as executing a program for predicting the occurrence of rice blast disease based on temporal hyperspectral features, etc.), and calling data stored in the memory 11, it performs various functions of the electronic device and processes data.

[0196] The memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disks, multimedia cards, card-type memories (such as SD or DX memories, etc.), magnetic memories, magnetic disks, optical discs, etc. In some embodiments, the memory 11 may be an internal storage unit of the electronic device, such as the mobile hard disk of the electronic device. In some other embodiments, the memory 11 may also be an external storage device of the electronic device, such as a plug-in mobile hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. equipped on the electronic device. Further, the memory 11 may also include both an internal storage unit and an external storage device of the electronic device. The memory 11 can be used not only to store application software installed on the electronic device and various types of data, such as the code of a program for predicting the occurrence of rice blast disease based on temporal hyperspectral features, etc., but also to temporarily store data that has been output or will be output.

[0197] The communication bus 12 may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. The bus is set to enable connection communication between the memory 11 and at least one processor 10, etc.

[0198] The communication interface 13 is used for communication between the above-mentioned electronic device 1 and other devices, including a network interface and a user interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), and is generally used to establish a communication connection between this electronic device and other electronic devices. The user interface may be a display, an input unit (such as a keyboard), and optionally, the user interface may also be a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch liquid crystal display, and an OLED (Organic Light-Emitting Diode) toucher, etc. Among them, the display may also be appropriately referred to as a display screen or a display unit, and is used to display the information processed in the electronic device and to display a visual user interface.

[0199] Figure 5 Only the electronic device with components is shown. Those skilled in the art can understand that Figure 5 The shown structure does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0200] For example, although not shown, the electronic device 1 may further include a power source (such as a battery) for supplying power to each component. Preferably, the power source may be logically connected to the at least one processor 10 through a power management device, so as to implement functions such as charging management, discharging management, and power consumption management through the power management device. The power source may also include any components such as one or more DC or AC power sources, a recharge device, a power failure detection circuit, a power converter or an inverter, and a power status indicator. The electronic device 1 may also include a variety of sensors, a Bluetooth module, a Wi-Fi module, etc., which will not be elaborated here.

[0201] It should be understood that the above embodiments are only for illustration purposes and are not limited by this structure in the scope of the patent application.

[0202] The rice blast disease occurrence prediction program based on temporal hyperspectral features stored in the memory 11 of the electronic device 1 is a combination of multiple instructions. When running in the processor 10, it can implement:

[0203] Real-time collect the hyperspectral data of rice in the target monitoring area to obtain a hyperspectral data set within a preset time period;

[0204] Intercept the hyperspectral data set according to a preset effective feature band to obtain an effective hyperspectral data set;

[0205] Using the pre-trained rice blast disease occurrence prediction model to perform feature encoding on the effective hyperspectral data set to obtain a hyperspectral encoding vector set;

[0206] Performing multi-scale feature extraction on the hyperspectral encoding vector set to obtain a multi-scale feature set, and splicing each multi-scale feature in the multi-scale feature set to obtain a multi-scale feature sequence;

[0207] Using the pre-constructed encoder in the rice blast disease occurrence prediction model to perform time-series feature decomposition based on feature trends on the multi-scale feature sequence to obtain a first trend encoding vector set and a first seasonal encoding vector set;

[0208] Performing time-series feature decomposition based on feature trends on the hyperspectral encoding vector set to obtain a second trend encoding vector set and a second seasonal encoding vector set;

[0209] Using the pre-constructed decoder in the rice blast disease occurrence prediction model to perform rice blast disease occurrence prediction operations on the first trend encoding vector set, the first seasonal encoding vector set, the second trend encoding vector set, and the second seasonal encoding vector set to obtain the rice blast disease occurrence prediction result in the target monitoring area, and the rice blast disease occurrence prediction result includes a confirmed result and a disease development prediction result.

[0210] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to the description of the relevant steps in the corresponding embodiment of the attached drawings, which will not be elaborated here.

[0211] Furthermore, if the modules / units integrated in the electronic device 1 are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM, Read-Only Memory).

[0212] The present invention also provides a computer-readable storage medium, and the readable storage medium stores a computer program, and when the computer program is executed by a processor of an electronic device, it can implement:

[0213] Real-time collecting hyperspectral data of rice in a target monitoring area to obtain a hyperspectral data set within a preset time period;

[0214] According to a preset effective feature band, intercepting the hyperspectral data set to obtain an effective hyperspectral data set;

[0215] Feature-encode the effective hyperspectral data set using a pre-trained rice blast disease occurrence prediction model to obtain a hyperspectral encoding vector set;

[0216] Perform multi-scale feature extraction on the hyperspectral encoding vector set to obtain a multi-scale feature set, and splice the various multi-scale features in the multi-scale feature set to obtain a multi-scale feature sequence;

[0217] Use the pre-constructed encoder in the rice blast disease occurrence prediction model to perform time-series feature decomposition based on feature trends on the multi-scale feature sequence to obtain a first trend encoding vector set and a first seasonal encoding vector set;

[0218] Perform time-series feature decomposition based on feature trends on the hyperspectral encoding vector set to obtain a second trend encoding vector set and a second seasonal encoding vector set;

[0219] Use the pre-constructed decoder in the rice blast disease occurrence prediction model to perform rice blast disease occurrence prediction operations on the first trend encoding vector set, the first seasonal encoding vector set, the second trend encoding vector set, and the second seasonal encoding vector set to obtain the rice blast disease occurrence prediction result in the target monitoring area, and the rice blast disease occurrence prediction result includes a confirmed result and a disease development prediction result.

[0220] In several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0221] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place, or they may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0222] In addition, the various functional modules in the various embodiments of the present invention can be integrated in one processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware, or in the form of a combination of hardware and software functional modules.

[0223] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above-mentioned exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention.

[0224] Therefore, from any perspective, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Thus, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be encompassed within the present invention. Any accompanying drawing reference signs in the claims should not be construed as limiting the claims involved.

[0225] The blockchain referred to in the present invention is a new application mode of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, and encryption algorithms. Blockchain, in essence, is a decentralized database, a series of data blocks generated by using cryptographic methods. Each data block contains information about a batch of network transactions, which is used to verify the validity (anti-counterfeiting) of the information and generate the next block. The blockchain can include a blockchain underlying platform, a platform product service layer, an application service layer, etc.

[0226] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Among them, Artificial Intelligence (AI) is a theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain the best results.

[0227] In addition, it is obvious that the word "including" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units or devices stated in the system claims can also be implemented by one unit or device through software or hardware. Words such as first, second, etc. are used to denote names and do not denote any specific order.

[0228] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for predicting the occurrence of rice blast disease based on time series hyperspectral features, characterized in that: The method comprises: Collecting hyperspectral data of rice in the target monitoring area in real time to obtain a set of hyperspectral data within a preset time period; Performing data interception on the hyperspectral data set according to a preset effective characteristic band to obtain an effective hyperspectral data set; Using a pre-trained rice blast disease prediction model to perform feature encoding on the effective hyperspectral data set to obtain a hyperspectral encoding vector set; Performing multi-scale feature extraction on the hyperspectral encoding vector set to obtain a multi-scale feature set, and performing feature splicing on each multi-scale feature in the multi-scale feature set to obtain a multi-scale feature sequence; Using the encoder pre-constructed in the rice blast disease occurrence prediction model to perform time series feature decomposition based on feature trends on the multi-scale feature sequence, to obtain a first trend encoding vector set and a first season encoding vector set; Performing time series feature decomposition based on feature trends on the hyperspectral coding vector set to obtain a second trend coding vector set and a second season coding vector set; The decoder pre-constructed in the rice blast disease occurrence prediction model is used to perform rice blast disease occurrence prediction operations on the first trend coding vector set, the first season coding vector set, the second trend coding vector set and the second season coding vector set to obtain the rice blast disease occurrence prediction results in the target monitoring area, and the rice blast disease occurrence prediction results include the diagnosis results and the disease progression prediction results.

2. The rice blast disease occurrence prediction method based on time series hyperspectral features according to claim 1, characterized in that: Before intercepting the hyperspectral data set according to the preset effective characteristic band, the method further includes: According to the pre-constructed Pearson coefficient and random forest algorithm, the Pearson correlation coefficient curve of the data features of each band in the hyperspectral data set for the rice blast disease is calculated; Determining whether early characteristics of rice blast disease occurrence are detected in a historical hyperspectral data set before the hyperspectral data set; When the historical hyperspectral data set detects the early feature, selecting a preset first threshold as a cutoff threshold; When the historical hyperspectral data set does not detect the early feature, selecting a preset second threshold as a cutoff threshold; The bands whose correlation coefficients in the Pearson correlation coefficient curve are greater than the interception threshold are screened to obtain effective characteristic bands.

3. The rice blast disease occurrence prediction method based on time series hyperspectral features according to claim 2, characterized in that: The method of calculating the Pearson correlation coefficient curve of the data features of each band in the hyperspectral data set for the rice blast disease according to the pre-constructed Pearson coefficient and random forest algorithm includes: Get a pre-built decision tree forest; According to a pre-built OOB error analysis algorithm, the out-of-bag data corresponding to the target decision tree in the decision tree forest is calculated, and the number of error samples in the out-of-bag data is obtained; Performing feature perturbations based on the target wavelength on the target decision tree to obtain updated out-of-bag data, and calculating the number of updated error samples of the updated out-of-bag data; Performing a preset number of iterations of the above-mentioned feature perturbation on each target decision tree to obtain a target wavelength error number for the target wavelength; According to the target wavelength error number and the number of error samples, the importance score of the target wavelength is calculated, and the importance score of the target wavelength is expressed as: In the formula, Indicates the target wavelength Hyperspectral data, Indicates the target wavelength The importance score of the hyperspectral data, represents the total number of decision trees in the decision tree forest, Indicates A decision tree, Represents a decision tree For the target wavelength The target wavelength error amount, Represents a decision tree The number of incorrect samples; According to a pre-constructed k-fold cross validation algorithm and the importance score of the target wavelength, a preset number of optimal decision tree sets are screened from the decision tree forest, and a random forest feature wavelength set corresponding to the optimal decision tree set is obtained; According to the Pearson coefficient, correlation calculation is performed on the hyperspectral data of each wavelength in the random forest characteristic wavelength set to obtain a Pearson correlation coefficient curve.

4. The rice blast disease occurrence prediction method based on time series hyperspectral features according to claim 3, characterized in that: Before the feature encoding of the effective hyperspectral data set is performed using the pre-trained rice blast disease prediction model, the method further comprises: Obtaining a pre-built rice blast disease occurrence prediction model, and obtaining a sample set recorded according to preset test conditions, and randomly grouping the sample set according to a preset training-test ratio to obtain a training set and a test set; Extracting a target sample from the training set in turn, predicting the occurrence of rice blast disease on the target sample using the rice blast disease occurrence prediction model, and obtaining a primary prediction result; Calculate the loss value between the true label corresponding to the target sample and the primary prediction result according to the pre-built cross entropy loss algorithm; Minimize the loss value to obtain the network model parameters when the loss value is the smallest; According to the pre-constructed K-fold cross-validation algorithm, the network parameter mean of the network model parameter corresponding to each target sample is calculated, and the network parameter mean is subjected to a reverse network update operation to obtain an updated rice blast disease occurrence prediction model; The test set is used to calculate the tolerance accuracy of the updated rice blast disease occurrence prediction model according to the preset tolerance value, wherein the tolerance accuracy is expressed as: In the formula, represents the tolerance accuracy, Indicates The primary prediction results of the test samples are Indicates The true labels of the test samples, represents the tolerance value, represents the number of samples in the test set; Indicates: When the primary prediction result With the true label If the absolute value of the difference is less than the tolerance value, it is recorded as 1, otherwise, it is recorded as 0; Determine whether the tolerance accuracy is greater than a preset qualified threshold; When the tolerance accuracy is less than or equal to the qualified threshold, returning to the step of randomly grouping the sample set according to the preset training-test ratio, and reallocating the training set and the test set to train the rice blast disease occurrence prediction model; When the tolerance accuracy is greater than the qualified threshold, the training process is stopped to obtain a trained rice blast disease occurrence prediction model.

5. The rice blast disease occurrence prediction method based on time series hyperspectral features according to claim 4, characterized in that: The obtaining of a sample set recorded according to preset test conditions includes: Obtaining pre-constructed susceptible varieties and disease-resistant varieties, and planting the susceptible varieties and disease-resistant varieties according to preset culture conditions until the rice seedlings grow to 3.5-4 leaf age; Spray inoculate the susceptible varieties and the disease-resistant varieties until the rice seedlings of the susceptible varieties and the disease-resistant varieties are completely diseased; Observe the reflectance spectra of 1 / 3, 1 / 2 and 2 / 3 of the base of rice leaves of susceptible and resistant varieties for a preset number of days to obtain observation data, and grade the disease conditions of the susceptible and resistant varieties according to the pre-constructed international rice blast seedling grading standard to obtain manual grading labels; The observation data and manual classification labels of the susceptible and resistant varieties are recorded to obtain a sample set.

6. The rice blast disease occurrence prediction method based on time series hyperspectral features according to claim 5, characterized in that: The encoder pre-constructed in the rice blast disease occurrence prediction model is used to perform time series feature decomposition based on feature trends on the multi-scale feature sequence to obtain a first trend encoding vector set and a first season encoding vector set, including: Using the encoder pre-constructed in the rice blast disease occurrence prediction model, the autocorrelation coefficient sequence between the multi-scale feature sequence and the delay time is calculated according to the preset delay time. The autocorrelation coefficient is expressed as: In the formula, Represents the autocorrelation coefficient sequence of the multi-scale feature sequence and the delay time, represents the delay time, represents the multi-scale feature sequence, Represents the mean of the multi-scale feature sequence; A one-dimensional convolution operation is performed on the multi-scale feature sequence and the autocorrelation coefficient sequence to obtain a trend feature, which is expressed as: In the formula, Indicates trend characteristics. Represents the weighted calculation results of multi-scale feature sequences and autocorrelation coefficient sequences; Subtracting the trend feature from the multi-scale feature sequence, a seasonal feature is obtained, and the seasonal feature is expressed as: In the formula, Indicates seasonal characteristics.

7. The method for predicting the occurrence of rice blast disease based on time series hyperspectral features according to claim 6, characterized in that: The method uses the decoder pre-constructed in the rice blast disease occurrence prediction model to perform a rice blast disease occurrence prediction operation on the first trend coding vector set, the first season coding vector set, the second trend coding vector set and the second season coding vector set to obtain a rice blast disease occurrence prediction result in the target monitoring area, wherein the rice blast disease occurrence prediction result includes a diagnosis result and a disease progression prediction result, including: Using the decoder pre-built in the rice blast disease occurrence prediction model to perform autocorrelation coefficient weight calculation on the second seasonal coding vector set, to obtain a second autocorrelation seasonal feature vector set; Performing time series feature decomposition on the second autocorrelation seasonal feature vector set to obtain a third trend feature set and a third seasonal feature set; Performing feature fusion on the third trend feature set and the second trend coding vector set to obtain a fourth trend feature set; Performing autocorrelation coefficient weight calculation on the weighted results between the second autocorrelation seasonal feature vector set and the first seasonal code vector set to obtain a third autocorrelation seasonal feature vector set; Performing time series feature decomposition on the third autocorrelation seasonal feature vector set to obtain a fifth trend feature set and a fourth seasonal feature set; Performing feature fusion on the fourth trend feature set and the fifth trend feature set to obtain a sixth trend feature set; The occurrence of rice blast disease is predicted based on the fourth season feature set and the sixth trend feature set to obtain a prediction result of the occurrence of rice blast disease in the target monitoring area.

8. A device for predicting the occurrence of rice blast disease based on time series hyperspectral features, characterized in that: The device comprises: The data acquisition and dimension reduction module is used to collect the hyperspectral data of rice in the target monitoring area in real time, obtain a hyperspectral data set within a preset time period, and intercept the hyperspectral data set according to a preset effective characteristic band to obtain an effective hyperspectral data set; A feature extraction module is used to perform feature encoding on the effective hyperspectral data set using a pre-trained rice blast disease prediction model to obtain a hyperspectral encoding vector set, perform multi-scale feature extraction on the hyperspectral encoding vector set to obtain a multi-scale feature set, and perform feature splicing on each multi-scale feature in the multi-scale feature set to obtain a multi-scale feature sequence; A feature decomposition module, for performing a time series feature decomposition based on a feature trend on the multi-scale feature sequence using an encoder pre-constructed in the rice blast disease occurrence prediction model to obtain a first trend coding vector set and a first season coding vector set, and performing a time series feature decomposition based on a feature trend on the hyperspectral coding vector set to obtain a second trend coding vector set and a second season coding vector set; The rice blast disease occurrence prediction module is used to use the pre-constructed decoder in the rice blast disease occurrence prediction model to perform rice blast disease occurrence prediction operations on the first trend coding vector set, the first season coding vector set, the second trend coding vector set and the second season coding vector set to obtain the rice blast disease occurrence prediction results in the target monitoring area, and the rice blast disease occurrence prediction results include the diagnosis results and the disease progression prediction results.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the rice blast disease occurrence prediction method based on time series hyperspectral features as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for predicting the occurrence of rice blast disease based on time series hyperspectral features as described in any one of claims 1 to 7 is implemented.

Images