Agricultural disease prevention cultivation system and method based on data processing

Abstract

The invention relates to the field of data processing, and particularly discloses an agricultural disease-prevention cultivation system and method based on data processing, which are characterized in that a high-precision model is constructed, and the model can predict a theoretical leaf temperature baseline of healthy crops in a specific environment in real time based on multi-dimensional environment time sequence data such as air temperature, humidity, radiation and wind speed. In order to ensure the accuracy of the prediction base line, an environment multi-factor nonlinear time sequence polymerization technology is adopted to effectively capture the comprehensive influence of the complex coupling effect among all environment factors on the leaf temperature. By comparing the actually measured leaf temperature with the dynamic base line, normal fluctuation caused by environmental factors can be effectively stripped, so that a pure physiological deviation signal is extracted. Thus, by analyzing the leaf temperature deviation flow, any continuous significant deviation can be judged to be abnormal caused by stress such as diseases and the like with high confidence, and then early warning is triggered.

Description

Technical Field

[0001] The present application relates to the field of data processing, and more specifically, to an agricultural disease prevention and cultivation system and method based on data processing. Background Art

[0002] During crop growth, the occurrence of various diseases is one of the main threats that affects their healthy growth, leading to yield decline and quality deterioration. Traditional agricultural disease control methods mainly rely on regular manual inspections and field experience judgment. This method is not only labor-intensive and inefficient, but also often cannot be discovered until disease symptoms have already become obvious, missing the optimal prevention and control opportunity. It leads to the need for large-scale use of pesticides for remediation, which not only increases production costs but also poses potential risks to the ecological environment and agricultural product safety. Therefore, the development of a cultivation technology that can provide early, automated, and accurate early warning of crop diseases is urgently needed to promote the development of modern agriculture towards intelligent and green directions.

[0003] With the development of the Internet of Things, big data, and artificial intelligence technologies, the use of sensors to monitor crop physiological status and growth environment in real time, coupled with intelligent decision-making through data analysis models, has provided a new technological path for the early diagnosis of crop diseases. Leaf temperature, a key indicator of crop physiological activity, is closely related to plant water status and stomatal conductance. When crops are exposed to stresses such as pathogen infection, their normal transpiration is inhibited, leading to stomata closure and abnormally elevated leaf surface temperature relative to healthy conditions. Leveraging this physiological response mechanism, monitoring leaf temperature changes for disease early warning has become a viable method. However, in practical applications, relying solely on the absolute value of leaf temperature or the simple difference from ambient temperature is extremely unreliable. This is because crop leaf temperature is not a constant value; it is dynamically influenced by multiple environmental factors, exhibiting complex, nonlinear fluctuations. For example, leaf temperature naturally rises under strong sunlight and decreases in windy conditions. This normal physiological fluctuation caused by the environment is intertwined with the abnormal physiological changes caused by disease stress. If a fixed threshold or simple model is used for judgment, it is very easy to produce a large number of false alarms and missed alarms, which cannot meet the needs of precision agriculture.

[0004] Therefore, an optimized agricultural disease prevention and cultivation program is desired. Summary of the Invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiments of the present application provide an agricultural disease prevention and cultivation system and method based on data processing.

[0006] According to one aspect of the present application, there is provided an agricultural disease prevention cultivation method based on data processing, comprising: Obtaining a time series of environmental factors and a time series of leaf temperature collected by a field sensor array, wherein the environmental factors include air temperature, relative humidity, incident solar radiation, and wind speed; Performing a dynamic prediction of a health status baseline based on the time series of the environmental factors to obtain a predicted leaf temperature baseline state; Extracting physiological deviation signals based on the time series of leaf temperature and the predicted leaf temperature baseline state to obtain a leaf temperature deviation flow; performing abnormal pattern recognition on the leaf temperature deviation flow based on the deviation flow to obtain abnormal diversion; Perform critical point judgment and early warning on the abnormal diversion to obtain a critical point early warning.

[0007] According to another aspect of the present application, there is provided an agricultural disease prevention and cultivation system based on data processing, comprising: a data acquisition module for acquiring a time series of environmental factors and a time series of leaf temperature collected by a field sensor array, wherein the environmental factors include air temperature, relative humidity, incident solar radiation, and wind speed; A health status baseline prediction module, configured to dynamically predict the health status baseline based on the time series of the environmental factors to obtain a predicted leaf temperature baseline state; a physiological deviation signal extraction module, configured to extract the physiological deviation signal based on the time series of the leaf temperature and the predicted leaf temperature baseline state to obtain a leaf temperature deviation flow; an abnormal pattern recognition module, configured to perform abnormal pattern recognition on the leaf temperature deviation flow based on the deviation flow to obtain abnormal diversion; The critical point warning module is used to perform critical point judgment and warning on the abnormal diversion to obtain a critical point warning.

[0008] Compared with the existing technology, the present application provides an agricultural disease prevention and cultivation system and method based on data processing. It constructs a high-precision model that can predict the theoretical leaf temperature baseline of healthy crops in a specific environment in real time based on multi-dimensional environmental time series data such as air temperature, humidity, radiation and wind speed. In order to ensure the accuracy of the predicted baseline, the environmental multi-factor nonlinear time series aggregation technology is adopted to effectively capture the comprehensive impact of the complex coupling effects between various environmental factors on leaf temperature. By comparing the actual measured leaf temperature with this dynamic baseline, the normal fluctuations caused by environmental factors can be effectively stripped away, thereby extracting pure physiological deviation signals. In this way, by analyzing the leaf temperature deviation flow, any persistent significant deviation can be determined with high confidence as an abnormality caused by stress such as disease, thereby triggering an early warning, fundamentally solving the problem of inaccurate early warning caused by the inability of traditional methods to strip away environmental interference. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 Flowchart of an agricultural disease prevention and cultivation method based on data processing according to an embodiment of the present application; Figure 2 A data flow diagram of an agricultural disease prevention and cultivation method based on data processing according to an embodiment of the present application; Figure 3 A flowchart of a method for agricultural disease prevention cultivation based on data processing according to an embodiment of the present application for dynamically predicting a health status baseline based on a time series of environmental factors to obtain a predicted leaf temperature baseline state; Figure 4 A flowchart of a method for agricultural disease prevention cultivation based on data processing according to an embodiment of the present application for dynamically predicting a health status baseline based on air temperature time series characteristics, relative humidity time series characteristics, incident solar radiation time series characteristics, and wind speed time series characteristics to obtain the predicted leaf temperature baseline state; Figure 5 4 is a block diagram of an agricultural disease prevention and cultivation system based on data processing according to an embodiment of the present application. DETAILED DESCRIPTION

[0011] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0012] As used in this application and the claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not intended to refer to the singular but may include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0013] Although the present application makes various references to certain modules in the system according to embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are illustrative only, and different aspects of the system and method can use different modules.

[0014] Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the various steps may be processed in reverse order or simultaneously, as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0015] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0016] Existing technologies using leaf temperature for disease early warning cannot accurately distinguish between normal physiological fluctuations in leaf temperature caused by complex environmental factors (such as light and wind speed) and early abnormal temperature rises caused by disease stress. This interference from environmental noise leads to low accuracy of early warning models and frequent false alarms and missed alerts. To address these technical issues, the present application proposes a data-processing-based agricultural disease prevention cultivation method. Specifically, a field sensor array acquires real-time multidimensional environmental time series data, including air temperature, relative humidity, incident solar radiation, and wind speed, as well as the crop's actual leaf temperature time series. Furthermore, rather than directly analyzing raw leaf temperature, this method utilizes the collected environmental data through an innovative multi-factor nonlinear time series aggregation method to fuse these dispersed and mutually coupled environmental factors into a single feature vector that comprehensively characterizes the combined environmental impact. Combined with static crop parameters, this method then drives a deep learning model to dynamically predict the theoretical leaf temperature baseline of healthy plants under the current specific environment. The real-time leaf temperature measurements are then compared point by point with this high-precision dynamic baseline, and the difference is calculated, resulting in a pure leaf temperature deviation stream that effectively filters out environmental background interference. Finally, this deviation stream is fed into a trained anomaly pattern recognition model. If the model detects that the pattern and magnitude of the deviation consistently exceed preset health thresholds, the system immediately triggers a critical point warning, thereby accurately and reliably capturing early signs of crop disease stress.

[0017] In the technical solution of this application, a data processing-based agricultural disease prevention and cultivation method is proposed. Figure 1 Flowchart of an agricultural disease prevention and cultivation method based on data processing according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of agricultural disease prevention cultivation method based on data processing according to the embodiment of the present application. Figure 1 and Figure 2As shown, the agricultural disease prevention cultivation method based on data processing according to the embodiment of the present application includes the steps of: S100, obtaining the time series of environmental factors and the time series of leaf temperature collected by the field sensor array, the environmental factors including air temperature, relative humidity, incident solar radiation and wind speed; S200, performing dynamic prediction of the health status baseline based on the time series of the environmental factors to obtain the predicted leaf temperature baseline state; S300, performing physiological deviation signal extraction based on the time series of leaf temperature and the predicted leaf temperature baseline state to obtain a leaf temperature deviation flow; S400, performing abnormal pattern recognition based on the deviation flow on the leaf temperature deviation flow to obtain an abnormal diversion; S500, performing critical point judgment and early warning on the abnormal diversion to obtain a critical point early warning.

[0018] Specifically, in step S100, the time series of environmental factors and the time series of leaf temperature collected by the field sensor array are obtained, and the environmental factors include air temperature, relative humidity, incident solar radiation and wind speed. It should be understood that since the leaf temperature of the crop is not an isolated physiological indicator, but a direct reflection of its energy and water exchange with the surrounding microenvironment, its value is subject to the complex dynamic influence of multiple environmental factors such as air temperature, relative humidity, incident solar radiation and wind speed. If these environmental factors are not used as a reference, it is impossible to determine whether the observed leaf temperature changes are due to normal responses to the environment or physiological abnormalities caused by disease stress. Therefore, in the technical solution of the present application, the time series of environmental factors and the time series of leaf temperature collected by the field sensor array are obtained to provide comprehensive and synchronous quantitative input data for the subsequent health status baseline dynamic prediction model. In this way, it can be ensured that the basis of the model analysis is established on the complete description of the crop-environment system, thereby laying a solid data foundation for accurately removing environmental interference and effectively extracting true physiological deviation signals.

[0019] Specifically, in a specific example of the present application, for the disease warning scenario of greenhouse tomatoes, the specific implementation process of this step is to systematically deploy a sensor array in the target cultivation area, such as a greenhouse or a field. Specifically, the implementation process includes: First, selecting representative locations in the greenhouse and installing sensor nodes according to a preset grid layout to cover the entire monitoring area. Second, at each or part of the nodes, air temperature and humidity sensors, photon sensors for measuring photosynthetically active radiation, and micro anemometers are integrated at the crop canopy height to ensure that the measured data can truly reflect the actual microenvironment in which the crop leaves are located. Third, use a non-contact infrared temperature measurement probe and aim it at the back of healthy leaves in the upper, middle, and lower parts of different plants, respectively, to obtain representative leaf temperature data in a multi-point measurement manner, and avoid direct sunlight on the measurement points. Fourth, all sensors collect data synchronously through the data collector at a fixed sampling frequency, for example, once every 5 minutes, and package the air temperature, relative humidity, incident solar radiation, wind speed and leaf temperature data with timestamps, and transmit them in real time to the back-end data processing server through the wireless communication module, thus forming a one-to-one corresponding time series data set for model analysis.

[0020] Specifically, in step S200, a dynamic prediction of the health status baseline is performed based on the time series of the environmental factors to obtain a predicted leaf temperature baseline state. It should be understood that since the actually measured leaf temperature is a comprehensive reflection of the crop's dynamic response to environmental factors and its own physiological state, it contains normal fluctuations caused by environmental changes, which seriously interferes with the identification of weak physiological abnormalities in the early stages of the disease. Therefore, in the technical solution of the present application, a dynamic prediction of the health status baseline is further performed based on the time series of the environmental factors to obtain a predicted leaf temperature baseline state, so as to construct a model that can accurately simulate the leaf temperature that healthy crops should have under specific environmental conditions. In this way, a high-fidelity dynamic reference baseline can be generated, which provides a key basis for the subsequent accurate separation and identification of physiological deviation signals caused by disease stress.

[0021] Figure 3 This is a flow chart of the agricultural disease prevention cultivation method based on data processing according to the embodiment of the present application, which performs a dynamic prediction of the health status baseline based on the time series of the environmental factors to obtain the predicted leaf temperature baseline state. Figure 3As shown, step S200 includes: S210, sorting the time series of environmental factors according to parameter samples to obtain the time series of air temperature, the time series of relative humidity, the time series of incident solar radiation and the time series of wind speed; S220, extracting the time series features of environmental factors from the time series of air temperature, the time series of relative humidity, the time series of incident solar radiation and the time series of wind speed to obtain the time series features of air temperature, the time series features of relative humidity, the time series features of incident solar radiation and the time series features of wind speed; S230, dynamically predicting the baseline of health status based on the time series features of air temperature, the time series features of relative humidity, the time series features of incident solar radiation and the time series features of wind speed to obtain the predicted leaf temperature baseline state.

[0022] In step S210, the time series of environmental factors are sorted according to the parameter samples to obtain the time series of air temperature, the time series of relative humidity, the time series of incident solar radiation, and the time series of wind speed. It should be understood that since the original environmental factor data stream collected from the sensor array often packages all environmental parameters collected at the same time into a composite data record during transmission and storage, this multivariate mixed format cannot be directly used by the subsequent feature extraction module designed to process single variable time series data. Therefore, in the technical solution of the present application, the time series of environmental factors are further sorted according to the parameter samples to obtain the time series of air temperature, the time series of relative humidity, the time series of incident solar radiation, and the time series of wind speed, so as to parse and reconstruct the original, heterogeneous data stream into multiple independent, homogeneous time series bound to specific physical meanings. In this way, it is possible to provide structured and standardized data input for subsequent independent time series feature extraction for each environmental factor, and ensure the precise alignment of different factor data in the time dimension, which is a necessary prerequisite for realizing multi-factor fusion analysis.

[0023] Specifically, in a specific example of the present application, the implementation process includes: first, the data processing server receives data frames sent from the field data collector at fixed time intervals, each data frame contains a timestamp and the corresponding measurement values ​​of air temperature, relative humidity, incident solar radiation and wind speed. Second, a data parsing program is started, which reads each data frame and accurately extracts the various values ​​belonging to air temperature, relative humidity, incident solar radiation and wind speed according to the preset data protocol or field identifier. Third, the extracted values ​​and their corresponding timestamps are used as a data point and appended to the parameter sequence storage area to which they belong. This process continues, thereby dynamically constructing and extending four parallel time-based numerical sequences. Fourth, the system performs a synchronization verification procedure to check whether the four generated time series have the same time length and completely aligned timestamps. For individual data points missing due to data transmission loss or temporary sensor failure, the system uses previous value filling or linear interpolation methods to fill them in. Finally, the system outputs four complete, clean, and strictly synchronized univariate time series, namely the time series of air temperature, relative humidity, incident solar radiation, and wind speed.

[0024] In step S220, the time series of air temperature, the time series of relative humidity, the time series of incident solar radiation and the time series of wind speed are subjected to environmental factor time series feature extraction to obtain air temperature time series features, relative humidity time series features, incident solar radiation time series features and wind speed time series features. It should be understood that since the original univariate time series is composed of a series of discrete numerical points, it has a high dimension and contains a large amount of redundant information. If it is directly used as input, the subsequent prediction model will find it difficult to efficiently capture the key dynamic information such as trends, periodicity and mutations contained therein. Therefore, in the technical solution of the present application, the time series of air temperature, the time series of relative humidity, the time series of incident solar radiation and the time series of wind speed are further subjected to environmental factor time series feature extraction to obtain air temperature time series features, relative humidity time series features, incident solar radiation time series features and wind speed time series features, so as to convert the original time series data within a period of time into an abstract feature vector of fixed dimension that contains its core dynamic characteristics. In this way, the dimension and noise of the input data can be significantly reduced, and more representational and information-dense input can be provided for subsequent multi-factor aggregation and prediction models, thereby improving the learning efficiency and accuracy of the entire baseline prediction model.

[0025] Specifically, in a specific example of this application, the implementation process includes: first, using the sliding window method to segment the four sorted environmental factor time series, setting a window length of 60 minutes (corresponding to 12 data points), sliding on each time series with a step size of 5 minutes (1 data point), thereby generating a series of overlapping time series segments. Second, for each generated time series segment, it is input into a pre-built one-dimensional convolutional neural network encoder. The encoder is independently configured for each environmental factor, and its internal convolutional layer and pooling layer are designed to automatically learn and capture the deep features of the factor in the time dimension, such as local patterns, rates of change, and short-term trends. Third, after the encoder's forward propagation calculation, each input time series segment is finally compressed and mapped into a fixed-dimensional feature vector. This process is performed independently for each time series segment of air temperature, relative humidity, incident solar radiation, and wind speed. Finally, at each time step, four independent feature vectors are generated corresponding to the air temperature time series feature, relative humidity time series feature, incident solar radiation time series feature, and wind speed time series feature for use in subsequent steps.

[0026] In step S230, a dynamic prediction of the baseline of the health status is performed based on the time series characteristics of the air temperature, the time series characteristics of the relative humidity, the time series characteristics of the incident solar radiation, and the time series characteristics of the wind speed to obtain the predicted baseline state of the leaf temperature. It should be understood that since the influence of various environmental factors on the leaf temperature is not a simple linear superposition, but there is a complex nonlinear coupling relationship, for example, the warming effect of solar radiation will be jointly regulated by wind speed and humidity. If the characteristics of each factor are viewed independently, this synergistic effect cannot be captured. Therefore, in the technical solution of the present application, a dynamic prediction of the baseline of the health status is further performed based on the time series characteristics of the air temperature, the time series characteristics of the relative humidity, the time series characteristics of the incident solar radiation, and the time series characteristics of the wind speed to obtain the predicted baseline state of the leaf temperature, so as to construct a prediction model that can fully learn and express the synergistic effect of multiple environmental factors. In this way, the accuracy and robustness of the predicted baseline state of the leaf temperature can be improved, so that it can accurately reflect the true physiological response of healthy crops in complex and changing environments.

[0027] Figure 4 A flowchart of the agricultural disease prevention cultivation method based on data processing according to an embodiment of the present application is used to dynamically predict the baseline health state based on the air temperature time series characteristics, relative humidity time series characteristics, incident solar radiation time series characteristics, and wind speed time series characteristics to obtain the predicted leaf temperature baseline state. Figure 4As shown, step S230 includes: S231, performing environmental multi-factor nonlinear time series aggregation on the air temperature time series characteristics, relative humidity time series characteristics, incident solar radiation time series characteristics and wind speed time series characteristics to obtain an environmental multi-factor time series aggregation feature coding vector; S232, performing dynamic prediction of the health status baseline based on the environmental multi-factor time series aggregation feature coding vector to obtain the predicted leaf temperature baseline state.

[0028] In step S231, the air temperature time series features, relative humidity time series features, incident solar radiation time series features and wind speed time series features are subjected to environmental multi-factor nonlinear time series aggregation to obtain an environmental multi-factor time series aggregation feature encoding vector. It should be understood that since the effects of the four environmental factors of air temperature, relative humidity, incident solar radiation and wind speed on leaf temperature are not isolated or linearly additive, and there is a complex nonlinear coupling relationship between them, simply splicing or averaging their time series features will smooth out the unique and highly discriminative influence information of each factor at a specific moment, and cannot effectively capture their synergistic effect. Therefore, in the technical solution of the present application, the air temperature time series features, relative humidity time series features, incident solar radiation time series features and wind speed time series features are further subjected to environmental multi-factor nonlinear time series aggregation to obtain an environmental multi-factor time series aggregation feature encoding vector, which provides more comprehensive and powerful input for the subsequent accurate prediction of the leaf temperature baseline.

[0029] Specifically, in an embodiment of the present application, environmental multi-factor nonlinear time series aggregation is performed on the air temperature time series characteristics, relative humidity time series characteristics, incident solar radiation time series characteristics and wind speed time series characteristics to obtain an environmental multi-factor time series aggregation feature coding vector, including: calculating the positional feature mean of the air temperature time series characteristics, relative humidity time series characteristics, incident solar radiation time series characteristics and wind speed time series characteristics to obtain an environmental multi-factor time series clustering center coding vector; calculating the residual components of each time series characteristic of the air temperature time series characteristics, relative humidity time series characteristics, incident solar radiation time series characteristics and wind speed time series characteristics relative to the environmental multi-factor time series clustering center coding vector to obtain a set of environmental multi-factor time series residual component coding vectors; performing residual adaptive aggregation on the set of environmental multi-factor time series residual component coding vectors to obtain an environmental multi-factor time series residual adaptive aggregation coding vector; and fusing the environmental multi-factor time series residual adaptive aggregation coding vector and the environmental multi-factor time series clustering center coding vector to obtain an environmental multi-factor time series aggregation feature coding vector.

[0030] More specifically, the positional feature means of the air temperature time series feature, the relative humidity time series feature, the incident solar radiation time series feature, and the wind speed time series feature are calculated to obtain the environmental multi-factor time series cluster center encoding vector, which is expressed as the following formula:

[0031] in, It is a time series feature set consisting of air temperature time series features, relative humidity time series features, incident solar radiation time series features, and wind speed time series features. are the time series characteristics of air temperature, relative humidity, incident solar radiation and wind speed, is 4, It is one of the time series characteristics among the time series characteristics of air temperature, relative humidity, incident solar radiation and wind speed. It is the center encoding vector of the environmental multi-factor temporal clustering.

[0032] It should be understood that when performing the aggregation of environmental multi-factor time series features, a global benchmark reference that can represent the common trend of all environmental factor features is required as the basis for subsequent analysis and decoupling. Without this benchmark, it is impossible to quantify the degree of deviation of each independent environmental factor time series feature relative to the overall environmental state. Therefore, in the technical solution of the present application, the positional feature means of the air temperature time series features, relative humidity time series features, incident solar radiation time series features and wind speed time series features are further calculated to obtain the environmental multi-factor time series clustering center coding vector, so as to capture the first-order statistical moment of the distribution of multiple environmental factor features at the current moment, that is, its central trend, thereby forming an initial code representing the commonality of all environmental factor features. In this way, a stable and reliable reference center can be provided for the subsequent residual calculation, completing the first step of the commonality-individuality decoupling in the feature distribution, that is, extracting the common part that can characterize the macro background of the environment.

[0033] More specifically, the residual components of each of the air temperature time series features, the relative humidity time series features, the incident solar radiation time series features, and the wind speed time series features relative to the environmental multi-factor time series cluster center coding vector are calculated to obtain a set of environmental multi-factor time series residual component coding vectors, which is expressed as the following formula:

[0034] in, is the first in the set of encoding vectors of the environmental multi-factor time series residual components The encoding vector of the environmental multi-factor time series residual component.

[0035] It should be understood that since the environmental multi-factor time series cluster center encoding vector obtained in the previous step only represents the global commonality or average trend of all environmental factor time series features, it will essentially smooth out the unique and highly discriminative impact information of each environmental factor, such as a sudden increase in solar radiation or gusts of wind, and these unique information are crucial for accurately predicting leaf temperature. Therefore, in the technical solution of the present application, the residual components of each time series feature of the air temperature time series feature, the relative humidity time series feature, the incident solar radiation time series feature and the wind speed time series feature relative to the environmental multi-factor time series cluster center encoding vector are further calculated to obtain a set of environmental multi-factor time series residual component encoding vectors, thereby drawing on the idea of ​​residual learning to transform the learning task from directly modeling the complex original feature space to modeling the residual space that is easier to learn and represents the degree of deviation. In this way, the stable global commonalities in the environmental multi-factor temporal characteristics can be effectively decoupled from the dynamic temporal characteristics of individual environmental factors, thereby clearly quantifying the unique impact or individual information of each environmental factor at the current moment, providing pure information and clear-cut input for subsequent adaptive aggregation and weighted processing.

[0036] More specifically, the set of the environmental multi-factor temporal residual component coding vectors is subjected to residual adaptive aggregation to obtain an environmental multi-factor temporal residual adaptive aggregation coding vector, which is expressed as the following formula:

[0037]

[0038]

[0039]

[0040] in, for activation function, and are the first learnable weight matrix and the second learnable weight matrix, is the environmental multi-factor temporal residual adaptive aggregation coding vector, Indicates that Projection group key vector, for function, is the point product by position, is the logarithmic function value with the natural constant e as the base, is the binorm of the vector, emphasizing significant deviations, for and The cosine value between constrains the residual direction consistency, For the Residual aggregation modulation coefficients, is the residual aggregation modulation weight.

[0041] It should be understood that the residual components decoupled from the temporal characteristics of each environmental factor are not equivalent in their final impact on leaf temperature. For example, at a certain moment, the drastic change in incident solar radiation may be the dominant factor, while at another moment, the sudden increase in wind speed may be more critical. If a simple averaging or maximization operation is used to aggregate these residual components containing individual information, it is impossible to dynamically identify and highlight the factors that currently play a decisive role, thereby losing the key information describing the synergy between factors. Therefore, in the technical solution of the present application, the set of the environmental multi-factor temporal residual component encoding vectors is further subjected to residual adaptive aggregation to obtain an environmental multi-factor temporal residual adaptive aggregation encoding vector, thereby introducing a data-driven attention module that can dynamically assign weights to the temporal residual components of each environmental factor according to the input data, so that the model can focus on those deviations that are most critical and have the richest information for describing the changes in the entire environmental state. In this way, we can essentially learn the second-order statistical characteristics of the characteristic distribution of environmental factors, intelligently screen and amplify the most significant characteristic change patterns, such as giving priority to drastic fluctuations in radiation or wind speed, while suppressing the noise and redundant information of other factors, and finally obtain an aggregated coding vector that accurately captures the dynamic and nonlinear synergy of multiple factors.

[0042] More specifically, the environment multi-factor temporal residual adaptive aggregation coding vector and the environment multi-factor temporal cluster center coding vector are fused to obtain the environment multi-factor temporal aggregation feature coding vector, which is expressed as the following formula:

[0043] in, A multi-factor temporal aggregation feature encoding vector for the environment.

[0044] It should be understood that although the single environmental multi-factor temporal residual adaptive aggregation coding vector captures the most significant dynamic changes of each environmental factor, it lacks the macro-background information of the overall environmental state; and although the single environmental multi-factor temporal cluster center coding vector can represent the global commonality, it cannot reflect the unique contribution of the key factors. Neither of them can provide a complete description of the environmental state for subsequent prediction. Therefore, in the technical solution of the present application, the environmental multi-factor temporal residual adaptive aggregation coding vector and the environmental multi-factor temporal cluster center coding vector are further fused to obtain the environmental multi-factor temporal aggregation feature coding vector, thereby constructing a hierarchical feature descriptor, which finally fuses the macro context representing the global distribution of the data with the refined micro details representing the local significant differences. In this way, a final aggregate coding vector can be generated that contains both the macro context representing the global distribution of the environment and the refined micro details representing the local significant differences, which has a high expressive power for both intra-class commonality and inter-class differences, providing the most comprehensive and powerful feature input for subsequent dynamic prediction of health status baseline.

[0045] In step S232, a dynamic prediction of the health status baseline is performed based on the environmental multi-factor time series aggregation feature coding vector to obtain the predicted leaf temperature baseline state. It should be understood that since only the environmental multi-factor time series aggregation feature coding vector can fully characterize the dynamic changes of the external environment and its synergistic effects, the physiological response of the crop, that is, the leaf temperature, is also profoundly affected by its own inherent properties. For example, there are differences in the responses of crops of different varieties and different growth periods to the same environment. Therefore, in the technical solution of the present application, a dynamic prediction of the health status baseline is further performed based on the environmental multi-factor time series aggregation feature coding vector to obtain the predicted leaf temperature baseline state, so as to effectively integrate the static physiological attributes of the crop with the dynamic environmental characteristics and construct a comprehensive model that can perform personalized and situational predictions. In this way, the predicted leaf temperature baseline can not only reflect environmental changes, but also adapt to the growth status of the crop itself, thereby greatly improving the accuracy and applicability of the baseline prediction.

[0046] Specifically, in this embodiment of the present application, dynamic health status baseline prediction based on an environmental multi-factor time-series aggregated feature encoding vector is performed to obtain the predicted leaf temperature baseline state. This includes extracting static crop parameter features; concatenating the static crop parameter features with the environmental multi-factor time-series aggregated feature encoding vector to obtain a static crop parameter-environmental factor synergistic feature; and performing feature decoding and regression prediction on the static crop parameter-environmental factor synergistic feature to obtain the predicted leaf temperature baseline state. Specifically, the implementation process includes: first, extracting static crop parameter features of the currently monitored object from a preset crop archive database. For example, the crop variety (e.g., Provence tomato) is converted into a feature vector using one-hot encoding, and the current growth stage (e.g., flowering and fruiting period) is mapped to a numerical value. Together, these constitute a static crop parameter feature vector. Second, at each time step, this static crop parameter feature vector is concatenated with the environmental multi-factor time-series aggregated feature encoding vector generated at the same time. This concatenation is performed end-to-end at the vector level, thereby forming a higher-dimensional, more comprehensive static crop parameter-environmental factor synergistic feature vector. Third, the collaborative feature vector of this static crop parameter-environmental factor is input into a pre-trained feature decoding regression network, which consists of several fully connected layers. Its function is to nonlinearly map the input abstract high-dimensional features to the target output space and ultimately output a single scalar value, which is the predicted leaf temperature baseline state of a specific crop in a specific environment at the current moment.

[0047] Specifically, in step S300, physiological deviation signals are extracted based on the leaf temperature time series and the predicted leaf temperature baseline state to obtain a leaf temperature deviation stream. It should be understood that since the actual measured leaf temperature time series is a mixture of the crop's normal response to the environment and physiological abnormalities caused by potential disease stress, and the predicted leaf temperature baseline state accurately simulates the influence of the former, if effective separation is not performed, environmental fluctuations will continue to mask weak early disease signals. Therefore, in the technical solution of the present application, physiological deviation signals are further extracted based on the leaf temperature time series and the predicted leaf temperature baseline state to obtain a leaf temperature deviation stream. This is done by directly comparing the actual measured values ​​with the theoretical predicted values ​​under healthy conditions, thereby quantifying and isolating temperature deviations caused solely by changes in the crop's internal physiological state. In this way, the complex, environmentally heavily interfered original leaf temperature signal can be converted into a deviation signal stream with an extremely high signal-to-noise ratio that directly reflects the physiological health status of the crop, providing a clear and reliable input for subsequent abnormal pattern recognition, greatly reducing the risk of false positives and missed positives.

[0048] More specifically, in this embodiment of the present application, physiological deviation signal extraction based on the leaf temperature time series and the predicted leaf temperature baseline state is performed to obtain a leaf temperature deviation stream, including calculating the positional difference between the leaf temperature time series and the predicted leaf temperature baseline state to obtain the leaf temperature deviation stream. Specifically, the implementation process is as follows: First, during each calculation cycle, for example, every 5 minutes, the data processing system synchronously acquires two data streams: the leaf temperature time series collected and processed by the non-contact infrared temperature measurement probe, and the predicted leaf temperature baseline state sequence output by the health baseline dynamic prediction model. Second, a positional difference calculation is performed to ensure that the two time series are strictly aligned at the timestamps. Then, for each shared timestamp, the actual measured leaf temperature value at that time is subtracted from the predicted leaf temperature baseline state value at the same time. Third, the calculated difference is treated as a new data point, attached with the corresponding timestamp, and continuously aggregated to form a continuous numerical sequence that quantifies the magnitude and direction of the physiological deviation, namely, the leaf temperature deviation stream. This data stream is then transmitted to the abnormal pattern recognition module for analysis.

[0049] Specifically, in step S400, the leaf temperature deviation flow is subjected to abnormal pattern recognition based on the deviation flow to obtain abnormal shunt. It should be understood that although the extracted leaf temperature deviation flow has filtered out environmental influences, it itself may still contain random disturbances caused by sensor noise or non-pathological instantaneous fluctuations. If a simple threshold is directly set for it to be judged, false alarms are very likely to occur. The physiological abnormalities caused by real disease stress are often manifested as the deviation flow showing a specific pattern of persistence and trend in the time dimension, rather than isolated spikes. Therefore, in the technical solution of the present application, the leaf temperature deviation flow is further subjected to abnormal pattern recognition based on the deviation flow to obtain abnormal shunt, so as to utilize an abnormality detection model trained only on health status data. The model learns and masters the intrinsic pattern of the deviation flow under normal physiological fluctuations, so as to be able to distinguish random noise from abnormal patterns with specific pathological significance. In this way, the analysis of deviation flow can be elevated from simple numerical comparison to the deep pattern recognition level. By quantifying the difference between the deviation flow and its normal pattern reconstruction result, a scoring flow that directly reflects the degree of pattern abnormality is generated, thereby achieving accurate and robust capture of early disease signals.

[0050] More specifically, in an embodiment of the present application, abnormal pattern recognition based on the leaf temperature deviation stream is performed on the leaf temperature deviation stream to obtain an abnormal branch, including: inputting the leaf temperature deviation stream into a trained anomaly detection model to obtain a reconstructed leaf temperature deviation stream; and calculating the reconstruction error between the leaf temperature deviation stream and the reconstructed leaf temperature deviation stream as the abnormal branch, where the reconstruction error is the absolute value of the difference between each leaf deviation and each reconstructed leaf temperature. Specifically, the abnormal pattern recognition process includes the following steps: first, continuously inputting the real-time leaf temperature deviation stream into a pre-trained anomaly detection model based on a long short-term memory network autoencoder in a sliding window of fixed length (for example, a time series segment containing the past two hours of data). This model is trained using only a large amount of leaf temperature deviation stream data from healthy crops under various environments and has learned how to accurately reconstruct a normal deviation stream pattern. Then, after receiving the input leaf temperature deviation stream segment, the anomaly detection model's encoder compresses it into a compact feature representation. The decoder then recovers the original deviation stream segment from this feature representation, thereby generating a reconstructed leaf temperature deviation stream. Recently, a reconstruction error calculation is performed. This involves calculating the absolute difference between the original deviation value in the input leaf temperature deviation stream and the corresponding reconstructed value in the reconstructed leaf temperature deviation stream output by the model at each time point. This reconstruction error, calculated over time, constitutes the abnormal diversion, and its magnitude directly reflects the degree to which the current leaf temperature deviation pattern deviates from the normal pattern.

[0051] Specifically, in step S500, the abnormal shunt is subjected to critical point judgment and early warning to obtain a critical point early warning. It should be understood that since the abnormal shunt itself may contain instantaneous spikes or short-term fluctuations caused by model reconstruction errors or signal transmission noise, if an alarm is triggered based solely on a single abnormal score exceeding a certain value, it is impossible to effectively distinguish between occasional disturbances that are not pathologically significant and persistent physiological abnormalities that are truly caused by disease stress, thereby resulting in an excessively high false alarm rate for the early warning system. Therefore, in the technical solution of the present application, the abnormal shunt is further subjected to critical point judgment and early warning to obtain a critical point early warning, thereby establishing a dual judgment criterion that takes into account both the degree of abnormality and the duration. By setting a reasonable abnormal score threshold and a necessary time window, the abnormal signal is integrated and confirmed in the time dimension. In this way, the instantaneous noise and isolated abnormal points in the system can be effectively filtered out, ensuring that the critical point early warning issued is based on the judgment of a confirmed and continuously developing abnormal pattern, thereby greatly improving the reliability and accuracy of the early warning decision, and providing a high-confidence decision basis for subsequent precise intervention.

[0052] More specifically, in an embodiment of the present application, a critical point judgment and early warning are performed on the abnormal diversion to obtain a critical point early warning, including: comparing each abnormal score in the abnormal diversion with a preset early warning threshold, and if the abnormal score is greater than the preset early warning threshold and the duration exceeds the preset time, determining that a critical point early warning is triggered. In particular, here, the preset time is 3 hours, and the preset early warning threshold is 0.9. That is, more specifically, the critical point judgment and early warning process includes the following steps: First, the early warning judgment module receives the abnormal diversion in real time, and in each data update cycle, compares the latest abnormal score value with a preset early warning threshold, and the preset early warning threshold is 0.9. Then, a state timer is maintained. When the abnormal score is detected to be greater than 0.9 for the first time, the timer starts; if the subsequent consecutive abnormal scores remain above 0.9, the timer continues to accumulate time; once an abnormal score is lower than 0.9, the timer is immediately cleared and reset, and the system continuously monitors the accumulated time of the timer. Then, if and only if the system confirms that the state with an abnormality score greater than 0.9 has continued uninterrupted for 3 hours (i.e., the preset duration), the system determines that the trigger condition has been met and formally determines to trigger a critical point warning. The warning signal is then sent to the user terminal or the automated control system to prompt the need for manual inspection of the corresponding crops or the initiation of intervention measures.

[0053] In summary, the agricultural disease prevention cultivation method based on data processing according to the embodiment of the present application is explained, which constructs a high-precision model that can predict the theoretical leaf temperature baseline of healthy crops in this specific environment in real time based on multi-dimensional environmental time series data such as air temperature, humidity, radiation and wind speed. In order to ensure the accuracy of the predicted baseline, the environmental multi-factor nonlinear time series aggregation technology is adopted to effectively capture the comprehensive impact of the complex coupling effects between various environmental factors on leaf temperature. By comparing the actual measured leaf temperature with this dynamic baseline, the normal fluctuations caused by environmental factors can be effectively stripped off, thereby extracting a pure physiological deviation signal. In this way, by analyzing the leaf temperature deviation flow, any persistent significant deviation can be determined with high confidence as an anomaly caused by stress such as disease, thereby triggering an early warning, which fundamentally solves the problem of inaccurate early warning caused by the inability of traditional methods to strip off environmental interference.

[0054] Furthermore, an agricultural disease prevention and cultivation system based on data processing is also provided.

[0055] Figure 5 FIG is a block diagram of an agricultural disease prevention and cultivation system based on data processing according to an embodiment of the present application. Figure 5As shown, the agricultural disease prevention and cultivation system 100 based on data processing according to the embodiment of the present application includes: a data acquisition module 110, which is used to obtain the time series of environmental factors and the time series of leaf temperature collected by the field sensor array, and the environmental factors include air temperature, relative humidity, incident solar radiation and wind speed; a health status baseline prediction module 120, which is used to perform dynamic prediction of the health status baseline based on the time series of the environmental factors to obtain a predicted leaf temperature baseline state; a physiological deviation signal extraction module 130, which is used to extract physiological deviation signals based on the time series of leaf temperature and the predicted leaf temperature baseline state to obtain a leaf temperature deviation flow; an abnormal pattern recognition module 140, which is used to perform abnormal pattern recognition based on the deviation flow on the leaf temperature deviation flow to obtain an abnormal diversion; a critical point warning module 150, which is used to perform critical point judgment and warning on the abnormal diversion to obtain a critical point warning.

[0056] As described above, the data processing-based agricultural pest prevention and cultivation system 100 according to the embodiment of the present application can be implemented in various wireless terminals, such as a server equipped with a data processing-based agricultural pest prevention and cultivation algorithm. In one possible implementation, the data processing-based agricultural pest prevention and cultivation system 100 according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or hardware module. For example, the data processing-based agricultural pest prevention and cultivation system 100 can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the data processing-based agricultural pest prevention and cultivation system 100 can also be one of the many hardware modules of the wireless terminal.

[0057] While various embodiments of the present disclosure have been described above, the above descriptions are illustrative, non-exhaustive, and not intended to be limiting of the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or improvements to existing technologies, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. An agricultural disease prevention and cultivation method based on data processing, characterized in that: include: Obtaining a time series of environmental factors and a time series of leaf temperature collected by a field sensor array, wherein the environmental factors include air temperature, relative humidity, incident solar radiation, and wind speed; Performing a dynamic prediction of a health status baseline based on the time series of the environmental factors to obtain a predicted leaf temperature baseline state; Extracting physiological deviation signals based on the time series of leaf temperature and the predicted leaf temperature baseline state to obtain a leaf temperature deviation flow; performing abnormal pattern recognition on the leaf temperature deviation flow based on the deviation flow to obtain abnormal diversion; Performing critical point judgment and early warning on the abnormal diversion to obtain a critical point early warning; Performing a dynamic prediction of a health status baseline based on the time series of the environmental factors to obtain a predicted leaf temperature baseline state includes: The time series of environmental factors are sorted according to parameter samples to obtain the time series of air temperature, relative humidity, incident solar radiation and wind speed; Extract environmental factor time series features from the time series of air temperature, relative humidity, incident solar radiation, and wind speed to obtain air temperature time series features, relative humidity time series features, incident solar radiation time series features, and wind speed time series features; A health status baseline dynamic prediction is performed based on the air temperature time series characteristics, the relative humidity time series characteristics, the incident solar radiation time series characteristics and the wind speed time series characteristics to obtain the predicted leaf temperature baseline state.

2. The agricultural disease prevention and cultivation method based on data processing according to claim 1, characterized in that: Performing a dynamic prediction of a health status baseline based on air temperature time series characteristics, relative humidity time series characteristics, incident solar radiation time series characteristics, and wind speed time series characteristics to obtain the predicted leaf temperature baseline state includes: Perform environmental multi-factor nonlinear time series aggregation on the air temperature time series characteristics, relative humidity time series characteristics, incident solar radiation time series characteristics and wind speed time series characteristics to obtain the environmental multi-factor time series aggregation feature encoding vector; A dynamic prediction of the health status baseline is performed based on the environmental multi-factor time series aggregated feature coding vector to obtain the predicted leaf temperature baseline state.

3. The agricultural disease prevention and cultivation method based on data processing according to claim 2, characterized in that: Perform environmental multi-factor nonlinear time series aggregation on the air temperature time series features, relative humidity time series features, incident solar radiation time series features, and wind speed time series features to obtain the environmental multi-factor time series aggregation feature encoding vector, including: Calculating the positional feature means of the air temperature time series feature, the relative humidity time series feature, the incident solar radiation time series feature, and the wind speed time series feature to obtain an environmental multi-factor time series clustering center encoding vector; Calculating the residual components of each of the air temperature time series characteristics, the relative humidity time series characteristics, the incident solar radiation time series characteristics, and the wind speed time series characteristics relative to the environmental multi-factor time series cluster center coding vector to obtain a set of environmental multi-factor time series residual component coding vectors; Performing residual adaptive aggregation on the set of the environmental multi-factor temporal residual component coding vectors to obtain an environmental multi-factor temporal residual adaptive aggregation coding vector; The environmental multi-factor temporal residual adaptive aggregation coding vector and the environmental multi-factor temporal clustering center coding vector are fused to obtain the environmental multi-factor temporal aggregation feature coding vector.

4. The agricultural disease prevention and cultivation method based on data processing according to claim 3, characterized in that: Performing a dynamic prediction of the health status baseline based on the environmental multi-factor time series aggregated feature coding vector to obtain the predicted leaf temperature baseline state includes: Extract static crop parameter features; Static crop parameter features and environmental multi-factor time series aggregation feature coding vectors are concatenated to obtain static crop parameter-environmental factor collaborative features; Feature decoding regression prediction is performed on the static crop parameter-environmental factor collaborative feature to obtain a predicted leaf temperature baseline state.

5. The agricultural disease prevention and cultivation method based on data processing according to claim 1, characterized in that: Based on the time series of leaf temperature and the predicted leaf temperature baseline state, physiological deviation signal extraction is performed to obtain a leaf temperature deviation flow, including: calculating the position difference between the time series of leaf temperature and the predicted leaf temperature baseline state to obtain the leaf temperature deviation flow.

6. The agricultural disease prevention and cultivation method based on data processing according to claim 1, characterized in that: Performing abnormal pattern recognition based on the deviation flow on the leaf temperature deviation flow to obtain abnormal diversion, including: Inputting the leaf temperature deviation flow into a trained anomaly detection model to obtain a reconstructed leaf temperature deviation flow; A reconstruction error between the leaf temperature deviation flow and the reconstructed leaf temperature deviation flow is calculated as the abnormal branch flow, where the reconstruction error is the absolute value of the difference between each blade deviation and each reconstructed leaf temperature.

7. The agricultural disease prevention and cultivation method based on data processing according to claim 1, characterized in that: Performing critical point judgment and early warning on the abnormal diversion to obtain a critical point early warning includes: Each abnormal score in the abnormal flow is compared with a preset warning threshold, and if the abnormal score is greater than the preset warning threshold and the duration exceeds a preset time, it is determined that a critical point warning is triggered.

8. The agricultural disease prevention and cultivation method based on data processing according to claim 7, characterized in that: The preset duration is 3 hours, and the preset warning threshold is 0.

9.

9. An agricultural disease prevention and cultivation system based on data processing, used to execute the method according to any one of claims 1 to 8, characterized in that: include: a data acquisition module for acquiring a time series of environmental factors and a time series of leaf temperature collected by a field sensor array, wherein the environmental factors include air temperature, relative humidity, incident solar radiation, and wind speed; A health status baseline prediction module, configured to dynamically predict the health status baseline based on the time series of the environmental factors to obtain a predicted leaf temperature baseline state; a physiological deviation signal extraction module, configured to extract the physiological deviation signal based on the time series of the leaf temperature and the predicted leaf temperature baseline state to obtain a leaf temperature deviation flow; an abnormal pattern recognition module, configured to perform abnormal pattern recognition on the leaf temperature deviation flow based on the deviation flow to obtain abnormal diversion; The critical point warning module is used to perform critical point judgment and warning on the abnormal diversion to obtain a critical point warning.

Images