Orchard pest and disease damage prediction method based on multi-source information fusion

Abstract

The invention relates to an orchard pest and disease damage prediction method based on multi-source information fusion. The orchard pest and disease damage prediction method comprises the following steps: S1, collecting multi-source data influencing the occurrence of pest and disease damage; s2, the collected multi-source data is preprocessed; s3, calculating influence factors of different data in the multi-source data, and screening characteristics highly related to pest and disease damage from the multi-source data; s4, redundant feature information is removed; s5, performing feature fusion on the multi-source data; and S6, constructing a pest and disease damage prediction model, and performing pest and disease damage prediction by taking the fused features as input. Multi-source data such as meteorological data, environmental data, soil data, fruit tree growth and pest and disease damage occurrence history are effectively fused, and the influence of different information on orchard pest and disease damage is comprehensively considered, so that the reliability and accuracy of orchard pest and disease damage prediction are improved.

Description

Technical Field

[0001] This invention relates to the field of orchard pest and disease prediction, specifically to a method for predicting orchard pests and diseases based on multi-source information fusion. Background Technology

[0002] Pests and diseases in orchards are one of the main threats to fruit tree growth, directly impacting tree development and causing significant yield reductions. Current pest and disease control methods often rely on a single data source, manual inspections, and statistical analysis for assessment, followed by the use of large quantities of chemical pesticides. This approach suffers from significant shortcomings in accuracy, timeliness, and environmental friendliness, failing to meet the demands of modern agriculture for precise, timely, and reliable pest and disease forecasting. Therefore, a multi-source information fusion-based orchard pest and disease forecasting method has emerged. This method integrates information from multiple data sources, comprehensively considers various factors influencing pest and disease occurrence, uses advanced data processing technologies and algorithms to make predictions, and provides scientific control recommendations. This achieves precise pest and disease control in orchards, protecting fruit tree health, ensuring stable yields, significantly reducing pesticide use, effectively protecting the environment, and making our orchards greener and healthier.

[0003] Patent publication number CN111602639A discloses an intelligent control method for pest and disease prevention in citrus orchards. This method predicts whether pests and diseases will occur in the orchard by considering factors such as temperature, humidity, light intensity, and carbon dioxide concentration. However, this method does not consider the impact of data such as rainfall on pest and disease occurrence and only performs a binary prediction (occurrence or non-occurrence).

[0004] Patent publication number CN103034910A discloses a regional-scale pest and disease prediction method based on multi-source information. By considering satellite remote sensing data reflecting vegetation physiological state and regional-scale meteorological data reflecting meteorological conditions, a prediction model is constructed using the Logistic regression method to predict the probability of pest and disease occurrence in different plots. Although it provides a more refined probability output for pest and disease prediction, it also suffers from the problem of incomplete consideration of factors. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for predicting orchard pests and diseases based on multi-source information fusion. This method comprehensively considers the impact of multi-source data, such as environment, weather, soil, fruit tree growth status, and the history of pest and disease occurrence, on the occurrence of pests and diseases. Multi-source data is fused according to the degree of influence of different data, thereby improving the accuracy of orchard pest and disease prediction.

[0006] To achieve the above objectives, the technical solution adopted by this invention is: a method for predicting orchard pests and diseases based on multi-source information fusion, comprising the following steps: S1. Collect multi-source data that affects the occurrence of pests and diseases; S2. Preprocess the collected multi-source data; S3. Calculate the influence factors of different data in the multi-source data, and screen the features that are highly correlated with the occurrence of pests and diseases from the multi-source data; S4. Remove redundant feature information; S5. Perform feature fusion on multi-source data; S6. Construct a pest and disease prediction model, using the fused features as input to predict pests and diseases.

[0007] As a preferred option, in step S1, the multi-source data includes temperature, soil moisture, rainfall, fruit tree growth stage, time of occurrence of pests and diseases, and severity of pests and diseases. The severity of pests and diseases is the area affected by pests and diseases, measured in mu (a Chinese unit of area, approximately 0.165 acres).

[0008] As a preferred option, in step S2, preprocessing includes data inspection, outlier removal, and data standardization.

[0009] As a preferred approach, step S3 involves calculating the influencing factor, which includes setting a first threshold of σ1 and selecting features based on their sensitivity to pest and disease occurrence. Specifically, the feature selection method involves calculating the bilinear correlation coefficient between each feature and pest and disease occurrence; this bilinear correlation coefficient is the influencing factor. Features with bilinear correlation coefficients higher than the first threshold are selected. The formula for calculating the bilinear correlation coefficient is as follows: Where ω is the point-to-point correlation coefficient, M1 is the average value of the characteristic variable when pests and diseases occur (i.e., the binary variable is 1), M0 is the average value of the characteristic variable when pests and diseases do not occur (i.e., the binary variable is 0), ρ is the proportion of pests and diseases occurring, and s y is the standard deviation of the characteristic variable.

[0010] As a preferred approach, step S4 involves removing redundant feature information, including setting a second threshold of σ2, calculating the Pearson correlation coefficients between each pair of features obtained in step S3, and removing features with Pearson correlation coefficients higher than the second threshold. The formula for the Pearson correlation coefficient is as follows: Where β is the Pearson correlation coefficient; X i and Y i These are the i-th observations of the two feature variables; and These are the means of the two feature variables.

[0011] As a preferred option, in step S5, a feature weighting method is used to fuse multi-source features, with the influence factor obtained in step S3 used as the weighting coefficient.

[0012] As a preferred option, in step S6, the CART regression decision tree method is used to construct the pest and disease prediction model.

[0013] The beneficial effects of this application are as follows: This application comprehensively considers the impact of multiple sources of data, such as environment, meteorology, soil, fruit tree growth status and the history of pest and disease occurrence, on the occurrence of orchard pests and diseases. It integrates multiple sources according to the degree of influence of different data, overcomes the bias and unreliability of a single data source, enhances the robustness of the model, and makes the prediction of orchard pests and diseases more accurate and reliable. Attached Figure Description

[0014] Figure 1 This is a flowchart of the method of the present invention;

[0015] Figure 2 A diagram illustrating the characteristics of pests and diseases at different growth stages of fruit trees;

[0016] Figure 3 This is a comparison chart of the predicted values ​​and the actual values ​​from the CART model. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. The following embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0018] This application provides a pest and disease prediction method based on multi-source information fusion, and the specific implementation steps are as follows:

[0019] S1. Data Acquisition: Temperature, soil moisture, and rainfall information are collected using digital temperature sensors, rain gauges, and soil moisture sensors at fixed locations in the orchard. Data on the timing and severity of pests and diseases during the fruit tree growth stages are obtained through manual observation and recording.

[0020] S2. Data Preprocessing: The collected information from different sources is divided into units of time, with Δt as the unit of time. The data in each unit of time is treated as a group. After data filtering and outlier removal, the data is normalized. The specific steps are as follows:

[0021] S21. Since high temperature and high humidity are suitable conditions for the occurrence of pests and diseases, the actan function is selected to normalize the information on temperature, soil moisture and rainfall.

[0022] S22. The growth stages of fruit trees cycle through budding, flowering, fruiting, ripening, and dormancy. During flowering and fruiting, fruit trees are susceptible to pests and diseases, while during dormancy and the initial budding stage, pests and diseases are latent and less likely to cause infection. Figure 2 As shown, since this characteristic matches the distribution of the sine function, the sin function is selected to normalize the data of the fruit tree growth stage.

[0023] S23. The timing and severity of pests and diseases during the fruit tree growth stages play a crucial role in the model construction. To preserve the original data distribution, the data must be normalized. Therefore, the Min-Max Normalization method is selected for processing. Where t1 represents temperature, soil moisture, and rainfall; t2 represents the fruit tree growth stage; and t3 represents the time and severity of pest and disease occurrence. i For the multi-source features that will be normalized, y i These are the corresponding features after normalization.

[0024] S3. Feature Selection: Calculate the binomial correlation coefficient (i.e., influence factor) between each feature and the occurrence of pests and diseases. A first threshold σ1 = 0.8 is preset. Features with binomial correlation coefficients higher than the first threshold are selected as candidate features. The specific selection rules and calculation formulas are as follows:

[0025] Where j∈[0,5], representing {0: temperature, 1: soil moisture, 2: rainfall, 3: fruit tree growth stage, 4: time of disease and pest occurrence, 5: severity of disease and pest occurrence}; ω j Let y be the point-binary correlation coefficient between the j-th feature variable and the occurrence of pests and diseases; i z is the i-th observation of the j-th feature variable; i n1 represents the observed value (i.e., 0 or 1) of the pest and disease occurrence variable corresponding to the j-th feature variable; n1 represents the occurrence of pests and diseases (i.e., z). i The number of samples when n is 1), where n0 represents the number of samples when pests and diseases have not occurred (i.e., z). i The number of samples when s is 0; y ρ represents the standard deviation of the j-th characteristic variable; ρ represents the proportion of observations where pests and diseases occur (value 1) out of all observations, and its value range is [0,1], representing the frequency of pests and diseases.

[0026] S4. Feature Filtering: Calculate the Pearson correlation coefficient between all candidate features. Set a second threshold σ² = 0.6, and remove features with Pearson correlation coefficients higher than the second threshold. The specific filtering rule calculation formula is as follows:

[0027] Where, β j,k X is the Pearson correlation coefficient between the j-th feature variable X and the k-th feature variable Y; j,k∈[0,5], X,Y∈[temperature, soil moisture, rainfall, fruit tree growth stage, time of occurrence of pests and diseases, severity of occurrence of pests and diseases]; X i Let be the i-th observation of the j-th feature variable X; Y is the mean of the j-th feature variable; i Let be the i-th observation of the k-th feature variable Y; Let be the mean of the k-th variable.

[0028] S5. The effective features selected in steps S3 and S4 are fused and used as input to the pest and disease prediction model. The fusion rule calculation formula is as follows: Where y1, y2, y3, y4, and y5 represent the multi-source features after screening and preprocessing, ω0 is a constant, and ω1, ω2, ω3, ω4, and ω5 are the correlation coefficients corresponding to each feature. It represents the total feature after fusing all features within the j-th Δt.

[0029] S6. Using the fused features as input to the model and the probability of pest and disease occurrence as output, construct a CART regression decision model. Divide the multi-source data from step S1 into a training set and a test set in an 8:2 ratio. Use the training set for training the CART regression decision model and the test set for testing the CART regression decision model. The specific design and training of the CART regression decision model includes the following steps:

[0030] S61, First, z i Let τ be the total features after fusion, D be the training set, and τ be the total features. i For the true value of the sample, (z) i ,τ i )∈D.

[0031] S62. Assume the split point is v, and the corresponding feature is z. i , with feature z i Assuming the splitting condition is met, nodes that satisfy the condition are assigned to the right subtree of that node, and nodes that do not satisfy the condition are assigned to the left subtree. The resulting left and right subsets are D. left (root,z i ) and Drig h t (root,z i The loss function at the split point v is: Among them, |D left (root,z i )| and |D rig h t (root,z i )| represent the number of samples in the left and right subsets, respectively. j ,τ j ) is a sample in the dataset, τ j It is the true value of the sample. and These are the predicted values ​​for the left and right subsets of the sample, respectively.

[0032] S63. Calculate the loss function for each sample as a split point, select the split point with the smallest loss function as the best split point, and divide the dataset into two subsets, left and right, based on the selected best split point and the corresponding features.

[0033] S64. For each subset, repeat steps S62 and S63 until the stopping condition is met.

[0034] S65. Set the recursion stopping condition to the maximum depth d of the decision tree. Stop the recursion when the depth of the decision tree is greater than d.

[0035] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for predicting orchard pests and diseases based on multi-source information fusion, characterized in that, Includes the following steps: S1. Collect multi-source data that affects the occurrence of pests and diseases; S2. Preprocess the collected multi-source data; S3. Calculate the influence factors of different data in the multi-source data, and screen the features that are highly correlated with the occurrence of pests and diseases from the multi-source data; S4. Remove redundant feature information; S5. Perform feature fusion on multi-source data; S6. Construct a pest and disease prediction model, using the fused features as input to predict pests and diseases.

2. The orchard pest and disease prediction method based on multi-source information fusion as described in claim 1, characterized in that, In step S1, the multi-source data includes temperature, soil moisture, rainfall, fruit tree growth stage, time of occurrence of pests and diseases, and severity of pests and diseases. The severity of pests and diseases is measured in mu (a Chinese unit of area, approximately 0.165 acres) as the area affected by pests and diseases.

3. The orchard pest and disease prediction method based on multi-source information fusion as described in claim 1, characterized in that, In step S2, preprocessing includes data inspection, outlier removal, and data standardization.

4. The orchard pest and disease prediction method based on multi-source information fusion as described in claim 1, characterized in that, In step S3, the influencing factor calculation includes setting a first threshold of σ1 and selecting features based on their sensitivity to pest and disease occurrence. Specifically, the feature selection method is as follows: calculate the bilinear correlation coefficient between each feature and pest and disease occurrence; this bilinear correlation coefficient is the influencing factor. Features with bilinear correlation coefficients higher than the first threshold are selected. The formula for calculating the bilinear correlation coefficient is as follows: Where ω is the point-to-point correlation coefficient, M1 is the average value of the characteristic variable when pests and diseases occur (i.e., the binary variable is 1), M0 is the average value of the characteristic variable when pests and diseases do not occur (i.e., the binary variable is 0), ρ is the proportion of pests and diseases occurring, and s y is the standard deviation of the characteristic variable.

5. The orchard pest and disease prediction method based on multi-source information fusion as described in claim 1, characterized in that, In step S4, removing redundant feature information includes setting a second threshold of σ2, calculating the Pearson correlation coefficients between each pair of features obtained in step S3, and removing features with Pearson correlation coefficients higher than the second threshold. The formula for the Pearson correlation coefficient is as follows: Where β is the Pearson correlation coefficient; X i and Y i These are the i-th observations of the two feature variables; and These are the means of the two feature variables.

6. The orchard pest and disease prediction method based on multi-source information fusion as described in claim 1, characterized in that, In step S5, feature weighting is used to fuse multi-source features, with the influence factor obtained in step S3 used as the weighting coefficient.

7. The orchard pest and disease prediction method based on multi-source information fusion as described in claim 1, characterized in that, In step S6, the pest and disease prediction model is constructed using the CART regression decision tree method.

Images