Method for preventing and controlling insect diseases by interplanting in citrus orchard

Abstract

The invention discloses a citrus orchard interplanting pest and disease prevention and control method, which comprises the following steps: installing a drip irrigation facility in a citrus orchard, and uniformly distributing drippers of the drip irrigation facility between each citrus tree and green manure planting rows; soil humidity sensors are buried in multiple positions of the citrus orchard to monitor soil humidity data in real time through the soil humidity sensors; and determining whether to start the drip irrigation facility based on the soil humidity data acquired by the plurality of soil humidity sensors and the position relationship among the plurality of soil humidity sensors. Therefore, whether the irrigation facility is started or not can be automatically decided by monitoring the soil humidity of different positions in the citrus garden in real time, and the intelligent control mode is beneficial to timely discovering and processing the soil water shortage condition; the problems that in a traditional mode, whether irrigation is carried out or not cannot be judged according to the actual humidity condition of soil, and irrigation hysteresis exists are solved, and then a basis is provided for citrus orchard interplanting pest and disease prevention and control management.

Description

Technical Field

[0001] The present application relates to the field of emotion recognition technology, and more specifically, to a method for intercropping in citrus orchards to prevent and control pests and diseases. Background Art

[0002] In recent years, with the expansion of citrus planting areas, pests and diseases, especially citrus psyllids (the vector of Huanglongbing) and red spiders, have become major challenges for fruit farmers. Traditional chemical control methods have limited effects and are prone to drug resistance. At present, planting green manure is considered to be one of the effective measures to improve saline-alkali land. Therefore, how to effectively improve the soil conditions of saline-alkali land through green manure planting has become a key issue of concern.

[0003] Chinese patent CN117859561A proposes a method for interplanting green manure in citrus orchards to prevent and control insect diseases. It can interplant with a variety of green manures and apply coated organic fertilizers, which not only improves the soil in the citrus orchard, but also improves disease resistance. In addition, in order to ensure the emergence rate, timely irrigation is required to keep the soil moist in the event of drought or dryness. However, traditional irrigation methods usually rely on empirical judgment, threshold judgment or a fixed schedule, and cannot accurately assess the actual moisture conditions of the soil, which may lead to overwatering or insufficient watering. Excessive watering will waste water resources and may cause root diseases, while insufficient watering will affect the normal growth of plants and even cause death. In addition, when signs of drought are detected by relying on manual observation and threshold comparison and then taking action, a certain degree of damage to the crops has often been caused. Timely irrigation is particularly important for ensuring the smooth germination of green manure seeds, and any delay may result in a reduced germination rate.

[0004] Therefore, an optimized intercropping scheme for preventing and controlling insect diseases in citrus orchards is desired. Summary of the invention

[0005] The present application provides a method for intercropping to prevent and control pests and diseases in a citrus orchard, which can automatically decide whether to start irrigation facilities by real-time monitoring of soil moisture at different locations in the citrus orchard. This intelligent control method helps to promptly detect and deal with soil water shortages, avoiding the problem that traditional methods cannot judge whether to implement irrigation based on the actual moisture conditions of the soil and that there is a lag in irrigation, thereby providing a basis for the management of intercropping to prevent and control pests and diseases in citrus orchards.

[0006] In a first aspect, a method for intercropping pest control in a citrus orchard is provided, comprising: evenly spreading organic fertilizer in the citrus orchard; in spring, selecting fast-growing and early-maturing leguminous green manure varieties, soaking green manure seeds of the leguminous green manure varieties in lime water before sowing, and planting the mixed green manure seeds in the citrus orchard after trenching in the field; in autumn, selecting sagittaria var. vulgaris green manure for sowing, and mixing the sagittaria var. vulgaris green manure seeds with calcium magnesium phosphate before sowing; characterized in that the method for intercropping pest control in a citrus orchard further comprises:

[0007] Installing a drip irrigation facility in the citrus orchard, wherein the drippers of the drip irrigation facility are evenly distributed between each citrus tree and green manure planting row;

[0008] burying soil moisture sensors at multiple locations of the citrus orchard to monitor soil moisture data in real time through the soil moisture sensors;

[0009] Based on the soil moisture data collected by the multiple soil moisture sensors and the positional relationship between the multiple soil moisture sensors, it is determined whether to start the drip irrigation facility.

[0010] In a possible implementation, determining whether to start the drip irrigation facility based on the soil moisture data collected by the multiple soil moisture sensors and the positional relationship between the multiple soil moisture sensors includes:

[0011] The soil moisture data collected by each soil moisture sensor are sorted according to the time dimension to obtain a time series input vector of moisture values ​​of multiple soil position points;

[0012] Performing sequence encoding on the multiple soil position point moisture value time series input vectors to obtain multiple soil position point moisture value time series implicit correlation feature vectors;

[0013] Performing global feature dynamic aggregation analysis on the implicit correlation feature vectors of the time series of the moisture values ​​of the plurality of soil position points to obtain a significant aggregation feature vector of the global time series of soil moisture;

[0014] Constructing a position distribution matrix between the plurality of soil moisture sensors, and performing spatial position feature compensation on the soil moisture global time series significant aggregation feature vector based on the position distribution matrix to obtain a spatially compensated soil moisture global time series feature vector;

[0015] Based on the global time series feature vector of the spatially compensated soil moisture, a control instruction is determined, where the control instruction is used to indicate whether to start the drip irrigation facility.

[0016] In one possible implementation, the multiple soil location point moisture value time series input vectors are sequence encoded to obtain multiple soil location point moisture value time series implicit correlation feature vectors, including: inputting the multiple soil location point moisture value time series input vectors into a sequence encoder based on an RNN-LSTM hybrid model to obtain the multiple soil location point moisture value time series implicit correlation feature vectors.

[0017] In a possible implementation, performing global feature dynamic aggregation analysis on the implicitly associated feature vectors of the time series of the moisture values ​​of the multiple soil position points to obtain a significant aggregate feature vector of the global time series of soil moisture includes:

[0018] Inputting each soil position point moisture value time series implicit correlation feature vector of the plurality of soil position point moisture value time series implicit correlation feature vectors into a characteristic static potential energy measurement network to obtain a set of soil position point moisture value time series characteristic static potential energy measurement coefficients;

[0019] Based on the set of static potential energy measurement coefficients of the soil position point moisture value time series characteristics, the implicit correlation feature vectors of the soil position point moisture value time series are screened and aggregated to obtain a pseudo-anchored aggregated center representation vector of the soil position point moisture value time series;

[0020] Calculate the aggregation movement direction of each soil position point moisture value time series implicit association feature vector in the plurality of soil position point moisture value time series implicit association feature vectors relative to the soil position point moisture value time series pseudo-anchor aggregation center representation vector to obtain a set of soil position point moisture value time series aggregation running directions;

[0021] Based on the set of aggregation running directions of the soil location point moisture value time series, the implicitly associated feature vectors of the multiple soil location point moisture value time series are dynamically aggregated toward the pseudo-anchored aggregation center representation vector of the soil location point moisture value time series to obtain the significant aggregation feature vector of the soil moisture global time series.

[0022] In a possible implementation, based on the set of static potential energy measurement coefficients of the soil position point moisture value time series characteristics, the implicit correlation feature vectors of the multiple soil position point moisture value time series are screened and aggregated to obtain a pseudo-anchored aggregated center representation vector of the soil position point moisture value time series, including:

[0023] Inputting the set of soil position point moisture value time series characteristic static potential energy measurement coefficients into the characteristic energy level screening gating unit to obtain a set of soil position point moisture value time series characteristic static potential energy weight factors;

[0024] Based on the set of static potential weight factors of the soil location point moisture value time series characteristics, the weighted sum of the implicitly associated feature vectors of the multiple soil location point moisture value time series is calculated to obtain the pseudo-anchored aggregation center representation vector of the soil location point moisture value time series.

[0025] In a possible implementation, constructing a position distribution matrix between the plurality of soil moisture sensors, and performing spatial position feature compensation on the soil moisture global time series significant aggregation feature vector based on the position distribution matrix to obtain a spatially compensated soil moisture global time series feature vector, including:

[0026] Constructing the position distribution matrix between the plurality of soil moisture sensors, wherein the value of each non-diagonal position in the position distribution matrix is ​​the distance between two corresponding soil moisture sensors;

[0027] The soil moisture global time series significant aggregation feature vector and the position distribution matrix are input into the humidity measurement spatial compensation network to obtain the spatially compensated soil moisture global time series feature vector.

[0028] In a possible implementation, the soil moisture global time series significant aggregation feature vector and the position distribution matrix are input into a humidity measurement spatial compensation network to obtain the spatially compensated soil moisture global time series feature vector, including:

[0029] Performing spatial feature activation processing on the position distribution matrix to obtain a position distribution feature matrix;

[0030] The position distribution feature matrix is ​​matrix-multiplied with the soil moisture global time series significant aggregation feature vector to obtain the spatially compensated soil moisture global time series feature vector.

[0031] In a possible implementation, performing spatial feature activation processing on the position distribution matrix to obtain a position distribution feature matrix includes: inputting the position distribution matrix into a spatial feature activation module based on a Sigmoid function to obtain the position distribution feature matrix.

[0032] In one possible implementation, a control instruction is determined based on the spatially compensated soil moisture global time series feature vector, and the control instruction is used to indicate whether to turn on the drip irrigation facility, including: inputting the spatially compensated soil moisture global time series feature vector into a classifier-based drip irrigation facility controller to obtain the control instruction.

[0033] The present application provides a method for intercropping pest control in a citrus orchard, which collects soil moisture data through real-time monitoring of these soil moisture sensors, and analyzes these soil moisture data and the spatial distribution relationship between each soil moisture sensor through a data processing algorithm based on artificial intelligence and deep learning, so as to capture the global temporal characteristics and aggregation semantics of soil moisture at different locations, and use the spatial distribution of different soil moisture sensors to perform spatial compensation for this global aggregation semantics to further optimize the expression of global temporal aggregation characteristics of soil moisture, so as to control the drip irrigation facilities to determine whether to automatically start the drip irrigation facilities. In this way, it is possible to automatically decide whether to start the irrigation facilities by real-time monitoring of the soil moisture at different locations in the citrus orchard. This intelligent control method helps to timely discover and deal with the situation of soil water shortage, avoiding the problem that the traditional method cannot judge whether to implement irrigation based on the actual moisture condition of the soil and the hysteresis of irrigation, thereby providing a basis for the management of intercropping pest control in citrus orchards. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application are briefly introduced below. Obviously, the drawings described below only relate to some embodiments of the present application, and are not intended to limit the present application.

[0035] Figure 1 This is a schematic flow chart of the method for intercropping in citrus orchards to prevent and control pests and diseases according to an embodiment of the present application.

[0036] Figure 2 In the method for intercropping in a citrus orchard to prevent and control pests and diseases in an embodiment of the present application, a schematic flowchart is provided for determining whether to turn on the drip irrigation facility based on the soil moisture data collected by the multiple soil moisture sensors and the positional relationship between the multiple soil moisture sensors.

[0037] Figure 3 In the method for intercropping in citrus orchards to prevent and control pests and diseases in an embodiment of the present application, a data flow diagram is provided for determining whether to turn on the drip irrigation facility based on the soil moisture data collected by the multiple soil moisture sensors and the positional relationship between the multiple soil moisture sensors.

[0038] Figure 4 In the method of intercropping in citrus orchards to prevent and control pests and diseases in an embodiment of the present application, a schematic flowchart is provided in which global feature dynamic aggregation analysis is performed on the implicit correlation feature vectors of the time series of moisture values ​​of the multiple soil location points to obtain a global feature vector of significant aggregation of the soil moisture time series.

[0039] Figure 5In the method of intercropping in citrus orchards to prevent and control pests and diseases in an embodiment of the present application, a position distribution matrix between the multiple soil moisture sensors is constructed, and based on the position distribution matrix, the spatial position feature compensation of the soil moisture global time series significant aggregation feature vector is performed to obtain a schematic flow chart of the spatially compensated soil moisture global time series feature vector. DETAILED DESCRIPTION

[0040] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present application.

[0041] Green manure crops, especially leguminous plants, are widely used to improve soil structure and fertility because of their ability to fix nitrogen. Fast-growing and early-maturing varieties can cover the ground in a short period of time, suppress weed growth, and their residues can increase the content of soil organic matter after decomposition. In addition, certain types of green manure crops can also serve as ecological barriers, disrupting the life cycle of pests and reducing the incidence of pests. For example, leguminous green manures sown in spring and arrow peas sown in autumn can not only improve the soil, but also resist pests and diseases through diversified planting patterns.

[0042] Based on this, in the technical solution of the present application, a method for intercropping pest control in a citrus orchard is proposed, comprising: evenly spreading organic fertilizer in the citrus orchard; in spring, selecting fast-growing and early-maturing leguminous green manure varieties, and when sowing, soaking the green manure seeds of the leguminous green manure varieties in lime water and mixing the seeds, and then planting the mixed green manure seeds in the citrus orchard after trenching in the field; in autumn, selecting sagittaria var. vulgaris green manure for sowing, and when sowing, mixing the green manure seeds of the sagittaria var. vulgaris with calcium magnesium phosphate fertilizers and then sowing. In particular, in order to be able to timely irrigate the various green manures intercropped in the citrus orchard, such as Figure 1 As shown, the method also includes: S1, installing a drip irrigation facility in the citrus orchard, wherein the drippers of the drip irrigation facility are evenly distributed between each citrus tree and green manure planting row; S2, burying soil moisture sensors at multiple locations in the citrus orchard to monitor soil moisture data in real time through the soil moisture sensors; S3, determining whether to turn on the drip irrigation facility based on the soil moisture data collected by the multiple soil moisture sensors and the positional relationship between the multiple soil moisture sensors.

[0043] Exemplarily, in step S1, a drip irrigation facility is installed in the citrus orchard, wherein the drippers of the drip irrigation facility are evenly distributed between each citrus tree and the green manure planting row. It should be understood that, through the drip irrigation facility, it can be ensured that each citrus tree and the green manure around it can get the right amount of water, avoiding the problem of overwatering or insufficient watering that may be caused by the traditional flooding method. The drip irrigation system directly transports water to the vicinity of the plant roots, reducing evaporation losses and improving water use efficiency. Evenly distributed drippers can make water penetrate into the soil more evenly, help maintain suitable soil moisture, promote soil microbial activity, and thus improve soil structure. Maintaining an appropriate soil moisture level can help reduce the risk of certain pathogens that rely on over-humid environments to reproduce, and can indirectly affect the life cycle of some pests by adjusting irrigation.

[0044] In one embodiment, a drip irrigation facility is installed in the citrus orchard, wherein the drippers of the drip irrigation facility are evenly distributed between each citrus tree and the green manure planting row, including: first determining the positions of all citrus trees in the citrus orchard, and planning the planting row spacing of green manure crops. Then, based on this information, a pipe network layout diagram of the drip irrigation system is designed to ensure that each dripper can cover the target area. Drip irrigation tapes or drip tubes suitable for the local climate and soil type are selected to ensure that their flow rate is stable and has strong anti-clogging ability. For the selection of drippers, it should be considered whether the water output meets the actual needs. Main pipelines are laid around the citrus trees, and branch pipelines are connected to each plant. Drippers are installed at a predetermined distance (usually within the edge of the citrus tree crown projection) so that they can be evenly distributed under the citrus trees and between the green manure planting rows. If possible, part of the pipeline is buried underground to protect the pipeline from damage by external factors.

[0045] Exemplarily, in step S2, soil moisture sensors are buried at multiple locations in the citrus orchard to monitor soil moisture data in real time through the soil moisture sensors. It should be understood that by setting soil moisture sensors at different locations, the actual soil moisture conditions in various areas of the orchard can be obtained, avoiding the one-sidedness caused by a single measurement point. Using the data collected by these sensors, combined with data processing algorithms based on artificial intelligence and deep learning, it is possible to capture the global temporal characteristics and aggregate semantics of soil moisture at different locations, and make intelligent decisions on whether to start drip irrigation facilities based on this. Real-time monitoring enables the system to respond immediately when soil moisture changes, ensuring that plants are always in the most suitable growth environment to prevent damage caused by drought or excessive humidity. By accurately assessing the actual water demand, over-irrigation or under-irrigation is avoided to achieve the purpose of water saving. Proper water management helps to improve the plant's own immunity and indirectly reduces the probability of pests and diseases. Long-term accumulated data can help farmers better understand the best planting patterns under local climatic conditions and guide future agricultural production activities.

[0046] In one embodiment, soil moisture sensors are buried at multiple locations in the citrus orchard to monitor soil moisture data in real time through the soil moisture sensors, including: selecting soil moisture sensors suitable for the orchard environment, such as tensiometers, capacitive sensors, etc., which should have good accuracy, stability and durability. Analyze factors such as the terrain and vegetation distribution of the citrus orchard to reasonably plan the location of the sensor. Ensure that key areas of the entire orchard are covered, especially those places where moisture differences are prone to occur (such as highlands and low-lying areas). Considering the range of crop root activity, the sensor is usually buried at a depth of about 10-30 cm from the surface. Dig holes according to the predetermined locations to ensure that the hole size is adapted to the sensor size. Insert the sensor into the pre-prepared hole and fill the surrounding space with original soil to ensure that the sensor is stable and in good contact. For wireless transmission sensors, it is also necessary to check the signal coverage to ensure that the data can be smoothly transmitted back to the monitoring center. All sensors should be connected to a central control system or cloud service platform, which is responsible for receiving data from various locations, performing necessary calculations and analysis, and ultimately outputting control instructions to the drip irrigation facilities. Regularly inspect and maintain the sensor, including cleaning the probe, replacing batteries (not necessary if it is wired powered), calibrating readings, etc. to ensure data accuracy.

[0047] Exemplarily, in step S3, based on the soil moisture data collected by the multiple soil moisture sensors and the positional relationship between the multiple soil moisture sensors, it is determined whether to turn on the drip irrigation facility. It should be understood that in order to further optimize the effectiveness of irrigation management and control, so as to automatically adjust the irrigation strategy and achieve the goal of water saving and efficiency improvement, an intelligent decision support system is also introduced. Specifically, in the process of automatic control of drip irrigation facilities, the technical concept of the present application is to monitor and collect soil moisture data in real time through these soil moisture sensors, and analyze these soil moisture data and the spatial distribution relationship between each soil moisture sensor together through a data processing algorithm based on artificial intelligence and deep learning, so as to capture the global temporal characteristics and aggregation semantics of soil moisture at different locations, and use the spatial distribution of different soil moisture sensors to perform spatial compensation for this global aggregation semantics to further optimize the expression of global temporal aggregation characteristics of soil moisture, so as to control the drip irrigation facilities to determine whether to automatically turn on the drip irrigation facilities. In this way, it is possible to automatically decide whether to start irrigation facilities by real-time monitoring of soil moisture at different locations in the citrus orchard. This intelligent control method helps to promptly detect and deal with soil water shortages, avoiding the problem that traditional methods cannot judge whether to implement irrigation based on the actual moisture conditions of the soil and that there is a lag in irrigation, thereby providing a basis for intercropping and pest control management in citrus orchards.

[0048] In one embodiment, Figure 2 and Figure 3 As shown, based on the soil moisture data collected by the multiple soil moisture sensors and the positional relationship between the multiple soil moisture sensors, determining whether to turn on the drip irrigation facility includes: S31, sorting the soil moisture data collected by each soil moisture sensor according to the time dimension to obtain multiple soil position point moisture value time series input vectors; S32, sequence encoding the multiple soil position point moisture value time series input vectors to obtain multiple soil position point moisture value time series implicit correlation feature vectors; S33, performing global feature dynamic aggregation analysis on the multiple soil position point moisture value time series implicit correlation feature vectors to obtain a soil moisture global time series significant aggregation feature vector; S34, constructing a position distribution matrix between the multiple soil moisture sensors, and based on the position distribution matrix, performing spatial position feature compensation on the soil moisture global time series significant aggregation feature vector to obtain a spatially compensated soil moisture global time series feature vector; S35, determining a control instruction based on the spatially compensated soil moisture global time series feature vector, the control instruction being used to indicate whether to turn on the drip irrigation facility.

[0049] Exemplarily, in step S31, the soil moisture data collected by each soil moisture sensor is sorted according to the time dimension to obtain a time series input vector of humidity values ​​of multiple soil position points. It should be understood that soil moisture is not static, but changes with time, weather conditions and other factors. By constructing a time series input vector, these changing trends can be recorded, which helps to understand the soil moisture state more accurately. By comparing and analyzing the humidity values ​​in different time periods, normal humidity fluctuation patterns and any abnormal conditions (such as drought or over-humidity) can be identified, so that timely measures can be taken. Based on time series data, machine learning algorithms can be used to predict the trend of soil moisture in the future, and then plan irrigation strategies in advance to ensure that crops are always in the most suitable growth environment. After understanding the specific needs of each location point at different times, water resources can be allocated more reasonably, waste can be avoided, and irrigation efficiency can be improved.

[0050] In one embodiment, the soil moisture data collected by each soil moisture sensor is sorted according to the time dimension to obtain a time series input vector of moisture values ​​of multiple soil position points, including: first, ensure that all soil moisture sensors installed in the citrus orchard work properly, collect soil moisture data regularly (for example, every hour), and upload these data to a central database or cloud platform. For each collected data, an accurate timestamp should be attached to indicate when the data was obtained. This step is very important because it ensures that data at different time points can be correctly sorted and associated in the subsequent processing process. According to the location information of the sensor, all historical data at the same location are classified together to form a data set indexed by geographic location. The purpose of this is to facilitate in-depth analysis of specific areas. For each location point of the soil moisture sensor, all humidity readings in the past period of time (such as one day, one week or one month) are collected. These readings are arranged in chronological order to form a one-dimensional array or vector, in which each element represents the soil moisture value at the corresponding moment. If there are missing values, they can be interpolated and filled according to the data of the adjacent time points, or directly marked as null values ​​for subsequent processing.

[0051] Exemplarily, in step S32, the multiple soil position point moisture value time series input vectors are sequence encoded to obtain multiple soil position point moisture value time series implicit correlation feature vectors. It should be understood that considering that the soil moisture data collected by different soil moisture sensors are all a data sequence that changes continuously over time, that is, the soil moisture values ​​of different position points have a time series correlation relationship and a dynamic change law in the time dimension. RNN (recurrent neural network) and its improved version LSTM (long short-term memory network) are particularly good at processing such time-dependent sequence data. They can remember past information and combine it with the current soil moisture value to make predictions or classifications. Therefore, in order to capture and understand the complex time series change patterns and potential relationships of soil moisture in the time dimension at each position point, in one embodiment, the multiple soil position point moisture value time series input vectors are sequence encoded to obtain multiple soil position point moisture value time series implicit correlation feature vectors, including: inputting the multiple soil position point moisture value time series input vectors into a sequence encoder based on the RNN-LSTM hybrid model to obtain the multiple soil position point moisture value time series implicit correlation feature vectors. Through the processing of the sequence encoder based on the RNN-LSTM hybrid model, the soil moisture time series characteristic information at each location point in the citrus orchard can be extracted separately, including but not limited to the trend of soil moisture change, periodic patterns, mutation points, etc., which are difficult to find by direct observation but are crucial to irrigation management. That is to say, in the actual environment, the change in soil moisture is often not linear, but is affected by many factors, such as weather conditions, crop water demand, groundwater level changes, etc. Through the processing of the sequence encoder based on the RNN-LSTM hybrid model, the soil moisture time series characteristic information at each location point in the citrus orchard and these complex nonlinear relationships can be extracted separately, including but not limited to the trend of soil moisture change, periodic patterns, mutation points, etc., which are difficult to find by direct observation but are crucial to irrigation management, so as to provide analysis results that are closer to the actual situation.

[0052] Exemplarily, in step S33, the implicitly associated characteristic vectors of the moisture value time series of the multiple soil position points are subjected to global feature dynamic aggregation analysis to obtain the soil moisture global time series significant aggregation characteristic vector. It should be understood that since the implicitly associated characteristic vectors of the moisture value time series of the multiple soil position points respectively contain the soil moisture time series characteristic information of different position points of the citrus orchard, and each soil moisture sensor is distributed at different positions of the citrus orchard, the soil moisture time series characteristic information of these different position points has a mutual correlation and interaction. However, since the soil moisture time series characteristics of not every position point have the same importance for the subsequent irrigation control task, there may be more insignificant noise and redundant information in the soil moisture time series information of some positions, and the contribution to the subsequent irrigation control task is low. Therefore, in order to better capture the overall dynamic characteristics and aggregation characteristics of soil moisture in the entire orchard, in the technical solution of the present application, the implicitly associated characteristic vectors of the moisture value time series of the multiple soil position points are further subjected to global feature dynamic aggregation analysis to obtain the soil moisture global time series significant aggregation characteristic vector.

[0053] In one embodiment, Figure 4 As shown, the global feature dynamic aggregation analysis is performed on the multiple soil position point moisture value time series implicit correlation feature vectors to obtain the soil moisture global time series significant aggregation feature vector, including: S331, each soil position point moisture value time series implicit correlation feature vector in the multiple soil position point moisture value time series implicit correlation feature vectors is input into the feature static potential energy measurement network to obtain a set of soil position point moisture value time series feature static potential energy measurement coefficients; S332, based on the set of soil position point moisture value time series feature static potential energy measurement coefficients, the multiple soil position point moisture value time series implicit correlation feature vectors are screened and aggregated to obtain the soil position point moisture value time series significant aggregation feature vector. S333, calculating the aggregation movement direction of each soil location point moisture value time series implicit association feature vector in the multiple soil location point moisture value time series implicit association feature vectors relative to the soil location point moisture value time series pseudo-anchor aggregation center representation vector to obtain a set of soil location point moisture value time series aggregation running directions; S334, based on the set of soil location point moisture value time series aggregation running directions, dynamically aggregate the multiple soil location point moisture value time series implicit association feature vectors toward the soil location point moisture value time series pseudo-anchor aggregation center representation vector to obtain the soil moisture global time series significant aggregation feature vector.

[0054] Exemplarily, in step S331, each soil position point moisture value time series implicit association feature vector in the plurality of soil position point moisture value time series implicit association feature vectors is respectively input into the characteristic static potential energy measurement network to obtain a set of soil position point moisture value time series characteristic static potential energy measurement coefficients. Specifically, the process can be expressed by the formula:

[0055] V={v1,v2,...,v i ,...,v n}

[0056]

[0057] Where V is the implicit correlation feature vector of the time series of the moisture values ​​of the multiple soil position points, v1, v2, v i ,v n are respectively the first, second, i-th and n-th soil position point moisture value time series implicit correlation feature vectors among the plurality of soil position point moisture value time series implicit correlation feature vectors, is the eigenvalue of each position in the implicitly associated eigenvector of the time series of the moisture value of the i-th soil position point, μ i and σ i They are v i The mean and standard deviation of i The number of eigenvalues ​​in , e i Yes i The static potential energy measurement coefficient of the time series characteristic of the moisture value at the corresponding soil location point.

[0058] That is, the characteristic static potential energy measurement network calculates a coefficient that reflects its "static" or "steady-state" characteristics by analyzing the soil moisture pattern of each location point over time. The higher this coefficient is, the smaller and more regular the soil moisture change at the location point is, so it can be considered that these location points are more important for maintaining the water balance in the orchard. Based on the above stability index, different weights can be assigned to each location point. Those location points that show higher stability will be given greater weights because they are more representative of the overall soil moisture status of the orchard and more reliable for irrigation control strategies. By analyzing different location points separately, some information hidden behind the data but crucial can be revealed, such as certain specific areas may be more susceptible to weather conditions and extreme humidity fluctuations. This helps the model better understand and predict these complex dynamics, so as to make more accurate irrigation decisions. In the subsequent steps, when the application aggregates the information of all location points to form a global feature representation, these measured coefficients can more accurately adjust the contribution ratio between the parts to avoid misjudgment due to certain outliers. By screening out truly representative location points, the impact of noise and irrelevant data on the final results can be effectively reduced, making the irrigation system respond more quickly and accurately.

[0059] Exemplarily, in step S332, based on the set of soil location point moisture value time series characteristic static potential energy measurement coefficients, the plurality of soil location point moisture value time series implicitly associated feature vectors are screened and aggregated to obtain the soil location point moisture value time series pseudo-anchored aggregate center representation vector, including: inputting the set of soil location point moisture value time series characteristic static potential energy measurement coefficients into the characteristic energy level screening gating unit to obtain the set of soil location point moisture value time series characteristic static potential energy weight factors; based on the set of soil location point moisture value time series characteristic static potential energy weight factors, calculating the weighted sum of the plurality of soil location point moisture value time series implicitly associated feature vectors to obtain the soil location point moisture value time series pseudo-anchored aggregate center representation vector. Specifically, the process can be expressed as follows:

[0060]

[0061]

[0062] Among them, w i Yes i The corresponding soil location point moisture value time series characteristic static potential energy measurement probability factor, w si Yes iThe corresponding static potential energy weight factor of the soil position point moisture value time series feature, θ is the predetermined threshold, mask is the mask processing, n is the number of vectors in the implicit correlation feature vector of the multiple soil position point moisture value time series, v a It represents the vector of pseudo-anchored aggregation center of soil point moisture value time series.

[0063] That is, through the processing of the characteristic energy level screening gated unit, the system can identify the stability and importance of each soil position point in the time series. In this process, those position points that show higher stability (i.e., their soil moisture changes are more regular) will be given greater weights, because these position points can better represent the overall soil moisture conditions of the orchard and are more reliable for irrigation control strategies. On the contrary, signals with large fluctuations or instability are suppressed or ignored, which helps to improve the quality of subsequent analysis and reduce the impact of redundant information on the final results. Next, based on the static potential energy measurement coefficient of the time series characteristics of the moisture values ​​of the screened soil position points, the system calculates the weight factor of each position point. These weight factors reflect the relative importance of each position point in the soil moisture management of the entire orchard, and can be dynamically adjusted according to changes in actual conditions. For example, in the dry season, the humidity changes in certain specific areas may become more critical. At this time, the system can reflect this change by updating the weight factor to better adapt to environmental conditions. Finally, by calculating the weighted sum of the implicitly associated feature vectors of the moisture value time series of multiple soil position points, the system can obtain a comprehensive "pseudo-anchored aggregation center representation vector". This vector is like a "center point" or "baseline" of the soil moisture in the entire orchard. It not only contains the information of each location point, but also reflects the relationship between them and the overall trend. With such a global representation vector, we can more clearly see the current soil moisture distribution and its changing trend in the orchard, and thus provide a basis for formulating a scientific and reasonable irrigation plan. For example, when a location point is found to deviate a lot from the pseudo-anchor aggregation center, it may mean that the area needs special attention and timely measures should be taken to avoid excessive water loss or water accumulation.

[0064] Exemplarily, in step S333, the aggregation movement direction of each soil position point moisture value time series implicit association feature vector in the plurality of soil position point moisture value time series implicit association feature vectors relative to the soil position point moisture value time series pseudo-anchor aggregation center representation vector is calculated to obtain a set of soil position point moisture value time series aggregation running directions. Specifically, the process can be expressed by the formula:

[0065]

[0066] Among them, ‖·‖ represents the vector norm, arccos(·) represents the inverse cosine function, θ i Indicates vi The corresponding soil location point moisture value time series aggregation running direction.

[0067] That is, by calculating the aggregate movement direction of the implicitly associated feature vector of the moisture value time series of each soil location point relative to the vector representing the pseudo-anchored aggregate center, the system is able to capture the specific trend of soil moisture changes over time at each location point. This directional information reveals whether each location point tends to be dry or wet, and how fast this change is. Such analysis not only helps identify which areas require immediate attention, but also provides an overall dynamic pattern of soil moisture throughout the orchard. For example, when most location points show a movement toward a drier direction, this may be a warning sign of the coming drought period; conversely, if most location points tend to be wetter, it may indicate the result of increased rainfall or other water replenishment.

[0068] Exemplarily, in step S334, based on the set of the soil position point moisture value time series aggregation running directions, the multiple soil position point moisture value time series implicitly associated feature vectors are dynamically aggregated toward the soil position point moisture value time series pseudo-anchor aggregation center representation vector to obtain the soil moisture global time series significant aggregation feature vector. Specifically, the process can be expressed as:

[0069]

[0070] Where sin(·) represents the sine trigonometric function, W i and b i Respectively represent v i The corresponding weight matrix and bias vector, v c is the significant aggregated feature vector of the soil moisture global time series.

[0071] That is, the implicitly associated feature vector of the time series of moisture values ​​at each soil location point is gradually brought closer to the pseudo-anchored aggregation center representation vector according to its aggregation movement direction to achieve dynamic aggregation. This is not just a simple mathematical operation, but to capture the overall dynamic pattern of all monitoring points in the orchard changing over time. In this way, the system can better understand the changing laws of soil moisture in the entire orchard, rather than being limited to data from a single or a few specific points. The global time series significant aggregation feature vector of soil moisture after dynamic aggregation is a highly generalized representation that reflects the interaction between different location points and the conduction effect of water in space. This feature vector not only contains the historical moisture information of each location point, but also integrates the correlation and evolution path between them, giving the model a deeper understanding of complex moisture dynamics.

[0072] In summary, the global feature dynamic aggregation analysis can evaluate and optimize the distribution of the implicit correlation feature vector of the soil moisture value time series of each location point in high-dimensional space by simulating the dynamic principles of the physical world, thereby enhancing the model's ability to understand the data intrinsic connection and dynamic evolution pattern between the soil moisture time series features of different location points. This is very important for understanding the flow of water in the orchard and predicting future trends, because this information directly affects whether the irrigation system is turned on or not. Specifically, the global feature dynamic aggregation analysis can effectively extract the soil moisture value time series implicit correlation feature representation with significant dynamic characteristics from the initial multiple soil moisture value time series implicit correlation feature vectors, thereby enhancing the model's ability to understand complex moisture dynamic correlation patterns. Specifically, the processing process of the global feature dynamic aggregation analysis automatically identifies and suppresses the insignificant or noisy soil moisture time series features of the location point through the feature energy level screening gating mechanism, and automatically selects the feature vectors that truly contribute to the irrigation decision, ensuring that the model can focus on the key location point moisture time series semantics, making the model more robust and generalizing, and improving the accuracy of subsequent irrigation control tasks. In addition, by introducing the feature static potential measurement network and the aggregation movement direction calculation module, the weight of feature aggregation is adaptively adjusted according to the intrinsic correlation between the implicit correlation feature vectors of the time series of soil moisture values ​​at different locations, thereby promoting the effective integration of the implicit correlation feature information of the time series of moisture values ​​at different soil locations, and enhancing the expression ability of the global time series aggregation semantics of soil moisture. This makes irrigation decisions not only rely on single-dimensional data, but also comprehensively considers multiple factors, providing strong support for intelligent irrigation systems.

[0073] Exemplarily, in step S34, a position distribution matrix between the plurality of soil moisture sensors is constructed, and spatial position feature compensation is performed on the soil moisture global time series significant aggregation feature vector based on the position distribution matrix to obtain a spatially compensated soil moisture global time series feature vector. It should be understood that soil moisture sensors are distributed in different areas, and the humidity data recorded by each sensor not only reflects the local conditions of the location, but may also be affected by adjacent areas, such as the natural diffusion of moisture from a wet area to a dry area. In order to fully understand these complex spatial relationships and ensure that irrigation decisions are based on the actual situation of the entire orchard, the present application constructs a position distribution matrix. This matrix describes in detail the relative positions and distances between the sensors, helping to analyze how moisture moves between different areas. By constructing such a position distribution matrix, the present application can more accurately capture the moisture conduction pattern between different locations in the orchard. This helps to avoid deviations caused by relying solely on data from a single or a few sensors. In actual environments, soil moisture often exhibits the characteristics of uneven distribution - for example, places close to water sources are usually more humid, while places far from water sources may be drier. If this spatial difference is not taken into account, directly using unadjusted data for irrigation control may result in overwatering in some areas while other areas do not get enough water. After introducing the position distribution matrix, the present application is able to spatially compensate the original soil moisture eigenvector. This means that when evaluating the soil moisture at a certain location, the situation in the surrounding area is also taken into account, thereby correcting the data deviation caused by uneven spatial distribution. This method makes the model closer to reality and provides a more accurate dynamic picture of orchard humidity. Ultimately, irrigation decisions made based on the spatially compensated soil moisture global time series eigenvector will be more scientific and reasonable, ensuring that every inch of land can obtain appropriate water supply, avoiding resource waste, and promoting healthy crop growth and effective use of water resources.

[0074] In one embodiment, Figure 5 As shown, a position distribution matrix between the multiple soil moisture sensors is constructed, and based on the position distribution matrix, spatial position feature compensation is performed on the soil moisture global time series significant aggregation feature vector to obtain the spatially compensated soil moisture global time series feature vector, including: S341, constructing the position distribution matrix between the multiple soil moisture sensors, the value of each position on the non-diagonal position in the position distribution matrix is ​​the distance between the corresponding two soil moisture sensors; S342, inputting the soil moisture global time series significant aggregation feature vector and the position distribution matrix into the humidity measurement spatial compensation network to obtain the spatially compensated soil moisture global time series feature vector.

[0075] Exemplarily, in step S341, the position distribution matrix between the multiple soil moisture sensors is constructed, and the value of each position on the non-diagonal position in the position distribution matrix is ​​the distance between the corresponding two soil moisture sensors. It should be understood that since the soil moisture sensors are distributed in different positions in the citrus orchard, the soil moisture time series characteristics of each position are not only affected by local conditions, but may also interact with other positions, for example, the moisture diffusion effect between adjacent areas. Therefore, in order to better analyze the spatial structure between different position points, so as to optimize and compensate for the temporal significance aggregation feature expression of the entire soil moisture domain, in the technical solution of the present application, a position distribution matrix between the multiple soil moisture sensors is further constructed, and the value of each position on the non-diagonal position in the position distribution matrix is ​​the distance between the corresponding two soil moisture sensors, which not only helps to understand the geometric structure of the sensor layout, but also captures the influence of moisture conduction and diffusion between different positions.

[0076] Exemplarily, in step S342, the soil moisture global time series significant aggregation feature vector and the position distribution matrix are input into the humidity measurement spatial compensation network to obtain the spatially compensated soil moisture global time series feature vector. It should be understood that since the soil moisture in the actual environment may be unevenly distributed in different locations (for example, places close to water sources are usually more humid than places far from water sources), and there is also a spatial transfer effect of soil moisture, directly using unadjusted feature vectors for irrigation control may lead to spatial deviations in decision-making. Therefore, the soil moisture global time series significant aggregation feature vector and the position distribution matrix are further input into the humidity measurement spatial compensation network to obtain the spatially compensated soil moisture global time series feature vector. The spatial compensation network can correct this spatial deviation based on the information provided by the position distribution matrix, so that the model can better understand the humidity dynamics in the entire orchard, rather than just being limited to the data of a single or a few specific points. This is crucial to achieving a globally optimal irrigation strategy.

[0077] In one embodiment, the soil moisture global time series significant aggregation feature vector and the position distribution matrix are input into the humidity measurement spatial compensation network to obtain the spatially compensated soil moisture global time series feature vector, including: performing spatial feature activation processing on the position distribution matrix to obtain the position distribution feature matrix; matrix multiplying the position distribution feature matrix with the soil moisture global time series significant aggregation feature vector to obtain the spatially compensated soil moisture global time series feature vector.

[0078] In one embodiment, performing spatial feature activation processing on the position distribution matrix to obtain a position distribution feature matrix includes: inputting the position distribution matrix into a spatial feature activation module based on a Sigmoid function to obtain the position distribution feature matrix.

[0079] Exemplarily, in step S35, a control instruction is determined based on the global time series feature vector of the spatially compensated soil moisture, and the control instruction is used to indicate whether to turn on the drip irrigation facility. In one embodiment, a control instruction is determined based on the global time series feature vector of the spatially compensated soil moisture, and the control instruction is used to indicate whether to turn on the drip irrigation facility, including: inputting the global time series feature vector of the spatially compensated soil moisture into a drip irrigation facility controller based on a classifier to obtain the control instruction. In other words, the global time series feature aggregation representation information of the spatially compensated soil moisture is used to perform classification processing, so as to control the drip irrigation facility to determine whether to automatically turn on the drip irrigation facility. In this way, it is possible to automatically decide whether to start the irrigation facility by real-time monitoring of the soil moisture at different locations in the citrus orchard. This intelligent control method helps to timely discover and deal with soil water shortages, avoiding the problem that the traditional method cannot determine whether to implement irrigation based on the actual moisture conditions of the soil and the lag in irrigation, thereby providing a basis for the management of intercropping and pest control in citrus orchards.

[0080] In one embodiment, the classifier uses a support vector machine (SVM) algorithm. The SVM-based classifier classifies and judges the input feature vector. According to the classification results, the controller generates corresponding control instructions: if the soil moisture in a certain area is lower than the set standard, an instruction to turn on the drip irrigation facility is issued; otherwise, if the humidity is suitable, the drip irrigation facility is kept as it is or is turned off. In order to facilitate operator monitoring and intervention, the present application is also equipped with a user-friendly monitoring interface (such as computer software or mobile phone application). Through this interface, the operator can view the soil moisture status of each location point and the automatic decision results of the system in real time. Although the system can automatically handle most decisions, in some special cases (such as extreme weather events or equipment failures), the operator can manually adjust the status of the drip irrigation facility through the monitoring interface. For example, the irrigation amount of a specific area can be temporarily increased or decreased as needed. Once the control instructions are generated, whether they are automatically generated or manually adjusted, they will be sent to the actual drip irrigation facility. After receiving the instructions, the drip irrigation facility performs the corresponding actions according to the instructions, such as opening or closing valves, adjusting the water flow size, etc. In order to ensure the correct execution of the instructions, the system is provided with a feedback mechanism. After the drip irrigation facility completes the operation, it will report the status to the controller, such as whether it has been successfully turned on or off. If there is any abnormal situation, the system will also promptly alarm the operator to check and deal with it.

[0081] In another embodiment, the classifier adopts a random forest algorithm, which is an integrated learning method that makes predictions by constructing multiple decision trees and outputs the mode or average of the prediction results of each decision tree as the final result. In the present application, it is first necessary to use the global time series feature vector of spatially compensated soil moisture as an input feature, and combine historical irrigation decisions as labels for data preparation. Then, these features and labels are used to train the random forest model. During the training process, each decision tree will select split nodes based on a randomly selected feature subset to increase model diversity. Then, the model performance is evaluated through cross-validation and other techniques, and the parameters are adjusted to optimize the accuracy. Finally, the trained model can receive the global time series feature vector of spatially compensated soil moisture monitored in real time, and output a control instruction on whether to turn on the drip irrigation facility based on the learned pattern.

[0082] In another embodiment, the classifier uses a deep neural network, which is an artificial intelligence algorithm that simulates the structure and function of the human brain and is particularly suitable for processing complex data patterns. In the present application, a multilayer perceptron (MLP) is first designed, comprising an input layer, a plurality of hidden layers, and an output layer, each layer consisting of a plurality of neurons, and the neurons between two adjacent layers are fully connected. Next, the spatially compensated soil moisture global time series feature vector is converted into a form suitable for input to the DNN, such as a normalized numerical array, and this step is called feature engineering. After that, the back propagation algorithm is used to minimize the loss function, such as the cross entropy loss, to update the network weights, and the batch gradient descent method or other optimization algorithms can be used to accelerate the convergence speed. In addition, the hyperparameters including the learning rate, batch size, number of hidden layers, and number of neurons in each layer must be experimentally determined to determine the optimal configuration. When the model is trained, it is deployed in the drip irrigation control system. Whenever new soil moisture data is received, the DNN will calculate and give corresponding irrigation suggestions, and a feedback mechanism is set to ensure that the system can respond to any abnormal situation in a timely manner.

[0083] In a preferred embodiment, inputting the spatially compensated soil moisture global time series feature vector into a classifier-based drip irrigation facility controller to obtain a control instruction comprises:

[0084] The feature set of the spatially compensated soil moisture global time series feature vector is clustered to obtain a spatially compensated soil moisture global time series in-class feature set and a spatially compensated soil moisture global time series out-class feature set, namely:

[0085]

[0086] in, represents the intra-class feature set of the global time series of spatially compensated soil moisture, f 1irepresents the ith eigenvalue of the feature set within the global time series of spatially compensated soil moisture, f 2j represents the jth eigenvalue of the out-of-class feature set of the global time series of spatially compensated soil moisture;

[0087] The ratio of the number of eigenvalues ​​in the feature set of the spatially compensated soil moisture global time series class to the number of eigenvalues ​​in the feature set of the spatially compensated soil moisture global time series feature vector is calculated, that is:

[0088]

[0089] in, represents the intra-class feature set of the global time series of spatially compensated soil moisture, f 1i represents the ith eigenvalue of the feature set within the global time series of spatially compensated soil moisture, f 2j represents the jth eigenvalue of the out-of-class feature set of the global time series of spatially compensated soil moisture;

[0090] The ratio of the λ-th power of the sum of the absolute values ​​of all eigenvalues ​​in the feature set of the spatially compensated soil moisture global time series class to the λ-th power of the sum of the absolute values ​​of all eigenvalues ​​in the feature set of the spatially compensated soil moisture global time series feature vector is calculated to obtain the first spatially compensated soil moisture global time series weight, that is:

[0091]

[0092] Among them, f p represents each eigenvalue in the feature set of the spatially compensated soil moisture global time series eigenvector, w1 represents the first spatially compensated soil moisture global time series weight;

[0093] The ratio of the λ / 2 power of the sum of squares of all eigenvalues ​​in the feature set of the spatially compensated soil moisture global time series class to the λ / 2 power of the sum of squares of all eigenvalues ​​in the feature set of the spatially compensated soil moisture global time series feature vector is calculated to obtain the second spatially compensated soil moisture global time series weight, that is:

[0094]

[0095] Among them, w2 represents the global time series weight of the second spatial compensation soil moisture;

[0096] For each eigenvalue in the feature set within the spatially compensated soil moisture global time series class, the product of the eigenvalue and the second spatially compensated soil moisture global time series weight is calculated, and then the product is added to the first spatially compensated soil moisture global time series weight to obtain the optimized eigenvalue, that is:

[0097] f′ 1i =w2×f1i +w1

[0098] Among them, f′ 1i represents the i-th optimized eigenvalue of the feature set within the global time series class based on the spatially compensated soil moisture;

[0099] For each eigenvalue in the out-of-class feature set of the spatially compensated soil moisture global time series, the product of the eigenvalue and the weight of the first spatially compensated soil moisture global time series is calculated to obtain the optimized eigenvalue, that is:

[0100] f 2j ′=w1×f 2j

[0101] Among them, f 2j ′ represents the jth optimized feature value of the out-of-class feature set based on the spatially compensated soil moisture global time series;

[0102] The optimized spatially compensated soil moisture global time series feature vector based on the optimized feature values ​​of the spatially compensated soil moisture global time series in-class feature set and the spatially compensated soil moisture global time series out-class feature set is input into a classifier-based drip irrigation facility controller to obtain a control instruction.

[0103] Here, when the soil moisture global time series significant aggregation feature vector and the position distribution matrix respectively represent the soil moisture time series aggregation features and the humidity sensor position distribution features of multiple soil moisture measurement points, when performing humidity measurement spatial compensation, it is considered that some position information is implicit in the soil moisture global time series significant aggregation feature vector, which will cause position information encoding overflow in the process of humidity measurement spatial compensation. It is therefore expected to improve the semantically consistent aggregation expression effect of the spatially compensated soil moisture global time series feature vector obtained through humidity measurement spatial compensation.

[0104] Therefore, while clustering the spatially compensated soil moisture global time series feature vector, the interactive description of the key feature information of the spatially compensated soil moisture global time series feature vector in the clustering process is carried out, and the geometric equivariant topology of the feature is constructed by low-rank harmonic modulation of the clustering features of the spatially compensated soil moisture global time series feature vector and the equivariance of the feature as a whole, so as to obtain the graphical distribution translation and rotation symmetry of the clustering features of the spatially compensated soil moisture global time series feature vector relative to the feature as a whole, thereby introducing geometric message passing into the feature expression of the spatially compensated soil moisture global time series feature vector and realizing the cluster mapping symmetry of the spatially compensated soil moisture global time series feature vector through manipulation of irreducible low-rank order coefficients, thereby improving the consistency of the clustering-based feature representation of the spatially compensated soil moisture global time series feature vector, thereby improving the accuracy of the control instructions obtained by the spatially compensated soil moisture global time series feature vector input into the classifier-based drip irrigation facility controller. This intelligent control method helps to detect and deal with soil water shortage in a timely manner, avoiding the problem that the traditional method cannot judge whether to implement irrigation based on the actual moisture conditions of the soil and the lag in irrigation, thereby providing a basis for intercropping and pest control management in citrus orchards.

[0105] In summary, the method of intercropping in citrus orchards for pest control according to the embodiment of the present application is explained, which collects soil moisture data through real-time monitoring by these soil moisture sensors, and analyzes these soil moisture data and the spatial distribution relationship between each soil moisture sensor through a data processing algorithm based on artificial intelligence and deep learning, so as to capture the global temporal characteristics and aggregation semantics of soil moisture at different locations, and use the spatial distribution of different soil moisture sensors to perform spatial compensation for this global aggregation semantics to further optimize the expression of global temporal aggregation characteristics of soil moisture, so as to control the drip irrigation facilities to determine whether to automatically start the drip irrigation facilities. In this way, it is possible to automatically decide whether to start the irrigation facilities by real-time monitoring of the soil moisture at different locations in the citrus orchard. This intelligent control method helps to promptly discover and deal with soil water shortages, avoiding the problem that the traditional method cannot determine whether to implement irrigation based on the actual moisture conditions of the soil and the lag in irrigation, thereby providing a basis for the management of intercropping in citrus orchards for pest control.

[0106] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the methods and devices according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, a program segment, or a part of a code, and the module, program segment, or a part of the code contains at least one executable instruction for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

Claims

1. A method for preventing and controlling pests and diseases in a citrus orchard by intercropping, comprising: Spreading organic fertilizer evenly in the citrus orchard; in spring, selecting fast-growing and early-maturing leguminous green manure varieties, soaking the green manure seeds of the leguminous green manure varieties in lime water and mixing them when sowing, and planting the mixed green manure seeds in the citrus orchard after trenching in the field; in autumn, selecting arrow pea green manure for sowing, mixing the arrow pea green manure seeds with calcium magnesium phosphate fertilizer and then sowing; characterized in that the method for intercropping pest control in the citrus orchard also includes: Installing a drip irrigation facility in the citrus orchard, wherein the drippers of the drip irrigation facility are evenly distributed between each citrus tree and green manure planting row; burying soil moisture sensors at multiple locations of the citrus orchard to monitor soil moisture data in real time through the soil moisture sensors; Determining whether to start the drip irrigation facility based on the soil moisture data collected by the multiple soil moisture sensors and the positional relationship between the multiple soil moisture sensors; Wherein, determining whether to start the drip irrigation facility based on the soil moisture data collected by the multiple soil moisture sensors and the positional relationship between the multiple soil moisture sensors includes: Soil moisture data collected by each soil moisture sensor are sorted according to the time dimension to obtain a time series input vector of moisture values ​​of multiple soil position points; Performing sequence encoding on the multiple soil position point moisture value time series input vectors to obtain multiple soil position point moisture value time series implicit correlation feature vectors; Performing global feature dynamic aggregation analysis on the implicitly associated feature vectors of the time series of the moisture values ​​of the plurality of soil position points to obtain a significant aggregated feature vector of the global time series of soil moisture; Constructing a position distribution matrix between the plurality of soil moisture sensors, and performing spatial position feature compensation on the soil moisture global time series significant aggregation feature vector based on the position distribution matrix to obtain a spatially compensated soil moisture global time series feature vector; Determining a control instruction based on the global time series feature vector of the spatially compensated soil moisture, wherein the control instruction is used to indicate whether to start the drip irrigation facility; Among them, based on the global time series feature vector of spatially compensated soil moisture, a control instruction is determined, and the control instruction is used to indicate whether to turn on the drip irrigation facility, including: inputting the global time series feature vector of spatially compensated soil moisture into a classifier-based drip irrigation facility controller to obtain the control instruction.

2. The method for preventing and controlling insect diseases in citrus orchards by interplanting according to claim 1, characterized in that: The multiple soil location point moisture value time series input vectors are sequence-encoded to obtain multiple soil location point moisture value time series implicit correlation feature vectors, including: inputting the multiple soil location point moisture value time series input vectors into a sequence encoder based on an RNN-LSTM hybrid model to obtain the multiple soil location point moisture value time series implicit correlation feature vectors.

3. The method for preventing and controlling pests and diseases in citrus orchards by interplanting according to claim 2, characterized in that: Performing global feature dynamic aggregation analysis on the implicitly associated feature vectors of the time series of the moisture values ​​of the multiple soil position points to obtain a significant aggregate feature vector of the global time series of soil moisture, including: The implicitly associated characteristic vectors of each soil position point moisture value time series in the plurality of soil position point moisture value time series implicitly associated characteristic vectors are respectively input into the characteristic static potential energy measurement network to obtain a set of soil position point moisture value time series characteristic static potential energy measurement coefficients. The process is expressed by the formula: V={v1,v2,...,v i ,...,v n } Where V is the implicit correlation feature vector of the time series of the moisture values ​​of the multiple soil position points, v1, v2, v i , v n are respectively the first, second, i-th and n-th soil position point moisture value time series implicit correlation feature vectors among the plurality of soil position point moisture value time series implicit correlation feature vectors, is the eigenvalue of each position in the implicitly associated eigenvector of the time series of the moisture value of the i-th soil position point, μ i and σ i They are v i The mean and standard deviation of i The number of eigenvalues ​​in , e i Yes i The static potential energy measurement coefficient of the time series characteristic of the corresponding soil position point moisture value; Based on the set of static potential energy measurement coefficients of the soil position point moisture value time series characteristics, the implicit correlation feature vectors of the soil position point moisture value time series are screened and aggregated to obtain a pseudo-anchored aggregated center representation vector of the soil position point moisture value time series; The aggregation movement direction of each soil position point moisture value time series implicit association feature vector in the multiple soil position point moisture value time series implicit association feature vectors relative to the soil position point moisture value time series pseudo anchor aggregation center representation vector is calculated to obtain a set of soil position point moisture value time series aggregation running directions. The process is expressed by the formula: Among them, ||·|| represents the vector norm, arccos(·) represents the inverse cosine function, θ i Indicates v i The corresponding soil location point moisture value time series aggregation running direction; Based on the set of aggregation running directions of the soil position point moisture value time series, the implicitly associated feature vectors of the multiple soil position point moisture value time series are dynamically aggregated toward the pseudo-anchored aggregation center representation vector of the soil position point moisture value time series to obtain the significant aggregation feature vector of the soil moisture global time series. The process is expressed by the formula: Where sin(·) represents the sine trigonometric function, W i and b i Respectively represent v i The corresponding weight matrix and bias vector, v c is the significant aggregated feature vector of the soil moisture global time series.

4. The method for preventing and controlling pests and diseases in citrus orchards by interplanting according to claim 3, characterized in that: Based on the set of static potential energy measurement coefficients of the soil position point moisture value time series characteristics, the implicit correlation feature vectors of the multiple soil position point moisture value time series are screened and aggregated to obtain a pseudo-anchored aggregated center representation vector of the soil position point moisture value time series, including: Inputting the set of soil position point moisture value time series characteristic static potential energy measurement coefficients into the characteristic energy level screening gating unit to obtain a set of soil position point moisture value time series characteristic static potential energy weight factors; Based on the set of static potential weight factors of the soil position point moisture value time series characteristics, the weighted sum of the implicitly associated feature vectors of the multiple soil position point moisture value time series is calculated to obtain the pseudo-anchored aggregation center representation vector of the soil position point moisture value time series. The process is expressed by the formula: Among them, w i Yes i The corresponding soil location point moisture value time series characteristic static potential energy measurement probability factor, w si Yes i The corresponding soil position point moisture value time series feature static potential weight factor, θ is the predetermined threshold, mask is the mask processing, n is the number of vectors in the implicit association feature vector of the multiple soil position point moisture value time series, v a It represents the vector of pseudo-anchored aggregation center of soil point moisture value time series.

5. The method for preventing and controlling pests and diseases in citrus orchards by interplanting according to claim 4, characterized in that: Constructing a position distribution matrix between the plurality of soil moisture sensors, and performing spatial position feature compensation on the soil moisture global time series significant aggregation feature vector based on the position distribution matrix to obtain a spatially compensated soil moisture global time series feature vector, including: Constructing the position distribution matrix between the plurality of soil moisture sensors, wherein the value of each non-diagonal position in the position distribution matrix is ​​the distance between two corresponding soil moisture sensors; The soil moisture global time series significant aggregation feature vector and the position distribution matrix are input into the humidity measurement spatial compensation network to obtain the spatially compensated soil moisture global time series feature vector.

6. The method for preventing and controlling pests and diseases in citrus orchards by intercropping according to claim 5, characterized in that: Inputting the soil moisture global time series significant aggregation feature vector and the position distribution matrix into the humidity measurement spatial compensation network to obtain the spatially compensated soil moisture global time series feature vector, including: Performing spatial feature activation processing on the position distribution matrix to obtain a position distribution feature matrix; The position distribution feature matrix is ​​matrix-multiplied with the soil moisture global time series significant aggregation feature vector to obtain the spatially compensated soil moisture global time series feature vector.

7. The method for preventing and controlling pests and diseases in citrus orchards by intercropping according to claim 6, characterized in that: Performing spatial feature activation processing on the position distribution matrix to obtain a position distribution feature matrix includes: inputting the position distribution matrix into a spatial feature activation module based on a Sigmoid function to obtain the position distribution feature matrix.

Images