Internet-of-things northern corn leaf blight early warning equipment

Abstract

The invention relates to the technical field of agricultural internet-of-things monitoring equipment, in particular to internet-of-things northern corn leaf blight early warning equipment. Comprising a meteorological data acquisition sensor used for acquiring key meteorological data such as air temperature, relative humidity, rainfall and the like; the power supply system adopts a solar power supply system and is provided with a solar cell panel and a storage battery; the data acquisition host is responsible for receiving sensor data and uploading the sensor data to the intelligent agricultural Internet of Things platform; data are collected through a meteorological data collection sensor, the data are uploaded to an intelligent agricultural Internet of Things platform through a data collection host, the intelligent agricultural Internet of Things platform compares and analyzes the data, the occurrence probability of northern corn leaf blight is given, and early warning is carried out in advance; according to the invention, the Internet of Things acquisition host can acquire meteorological data in real time and analyze the data on the platform; the occurrence index of northern corn leaf blight can be quickly and accurately predicted, and a more scientific prediction scheme is provided for agricultural production.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural Internet of Things monitoring devices, and particularly relates to an Internet of Things early warning device for northern corn leaf blight. Background Art

[0002] At present, the prevention and control of northern corn leaf blight mainly rely on observing the occurrence of the disease, and the probability of the disease occurrence cannot be predicted in advance.

[0003] The existing disease diagnosis and decision-making systems mainly target professional experts and scientific researchers in the field. This has led to a prominent problem, that is, for ordinary farmers, there are great difficulties in understanding and actual operation and use. Some systems even need to rely on agronomists for remote diagnosis before they can give the final conclusion, and this process often takes a lot of time. Moreover, this remote diagnosis method is greatly restricted by network conditions. In those remote areas, the network signal is poor or there is no network coverage at all, which makes it extremely inconvenient for farmers to use these systems and difficult to effectively use the systems to timely diagnose the disease situation of crops, thus causing a certain degree of obstruction and trouble to agricultural production.

[0004] Sample collection and laboratory testing: Collect samples such as suspected diseased leaves, stems, and ears, bring them back to the laboratory to describe the lesions, and confirm the pathogenic fungi through methods such as microscopic examination, tissue section observation, culture, and biochemical identification.

[0005] Spore trap: Using the principle of aerodynamics, the spore trap captures the pathogenic spores in the air onto a glass slide through a fan, and after being photographed by an internal camera and a high-power microscope, it is uploaded to the Internet of Things platform to achieve automatic statistical analysis.

[0006] At present, the prevention and control of northern corn leaf blight mainly rely on the observation and diagnosis after the disease occurs, lacking effective early warning means. Most of the existing disease diagnosis and decision-making systems are targeted at experts and scientific researchers in the professional field. Ordinary farmers have great difficulties in understanding and using these systems. Moreover, some systems rely on remote diagnosis and are significantly restricted by network conditions, making it difficult to meet the needs of farmers in remote areas. In addition, although the traditional spore trap can capture pathogenic spores, the data analysis efficiency is low and real-time early warning cannot be achieved.

[0007] The current disease prediction system is complex to operate and difficult for ordinary farmers to master;

[0008] Remote diagnosis is limited by network conditions, and it is impossible to obtain the diagnosis results in a timely manner in remote areas;

[0009] The data processing of the traditional spore trap lags behind and rapid early warning cannot be achieved;

[0010] Lack of real-time and accurate prediction means based on meteorological data. Summary of the Invention

[0011] To solve the problems of the prior art, the present invention provides an Internet of Things early warning device for northern corn leaf blight. This device collects environmental data in real time and analyzes it to predict the occurrence probability of northern corn leaf blight in advance, providing scientific and effective disease control solutions for farmers.

[0012] To achieve the above object, the present invention adopts the following technical solutions:

[0013] An Internet of Things early warning device for northern corn leaf blight, comprising:

[0014] A meteorological data collection sensor for collecting key meteorological data such as air temperature, relative humidity, rainfall, etc.;

[0015] A power supply system, using a solar power supply system, equipped with a solar panel and a storage battery;

[0016] A data collection host, responsible for receiving sensor data and uploading it to the intelligent agriculture Internet of Things platform;

[0017] Collect data through the meteorological data collection sensor, and upload the data to the intelligent agriculture Internet of Things platform through the data collection host. The intelligent agriculture Internet of Things platform will compare and analyze the data, give the occurrence probability of northern corn leaf blight, and issue an early warning in advance;

[0018] Among them, data collection: collect environmental data in real time through the meteorological data collection sensor;

[0019] Data transmission: upload the collected data to the intelligent agriculture Internet of Things platform through the data collection host;

[0020] Data analysis: The platform compares and analyzes the data, and uses the time series Prophet-LSTM model for prediction;

[0021] Early warning release: According to the analysis results, the platform gives the occurrence probability of northern corn leaf blight and displays the early warning information to users through a visual interface;

[0022] The further steps of using the time series Prophet-LSTM model for prediction include the following steps:

[0023] 1) The Prophet-LSTM model predicts by modeling from dimensions such as air temperature, air humidity, soil temperature, soil humidity, soil pH, wind speed, etc. The Internet of Things warning device for northern corn leaf blight uploads data to the server database, which is obtained through means such as SQL query, web crawler, and interfaces. The time interval for the device sensor to upload data depends on the actual situation. Here, our device sensor uploads a piece of data every about 10 minutes. Use pandas to read the data of the past year, divide it into multiple dimensions according to time points, and form a format that conforms to Prophet;

[0024] 2) Process the sample data into a format that conforms to Prophet;

[0025] 3) Finally, input the processed data into the model, and obtain the Prophet model through the time series model constructed by dataset training; create a data frame of the dates for the next month and then obtain the predicted data;

[0026] 4) Connect by calculating the residuals between the actual values and the predictions and input them into the LSTM model;

[0027] Residual calculation formula: where y t represents the actual value, represents the predicted value.

[0028] Furthermore, construct the LSTM model. Use the data processed by Prophet as samples to construct a sequence encoding Xt that conforms to the input format of the LSTM model, and train the LSTM model through the constructed sequence encoding. Specifically include:

[0029] Through the sequence encoding X t and the output h t-1 of the hidden layer at the previous time point, calculate the forget gate f t in the LSTM model, the input gate i t and the output gate o t , and calculate the neuron memory reference value at time point t through the tanh function; update the old cell state information of the LSTM model through the forget gate f t and the input gate i t ; calculate the output of the hidden layer neuron at time point t through the updated old cell state information and the output gate o t ;

[0030] Predict indicators such as air temperature and humidity, wind speed, soil temperature and humidity in the next cycle through Prophet-LSTM, input the predicted values into the BP neural network to obtain the comprehensive predicted values, calculate the Euclidean distance according to the temperature and humidity range most suitable for spore growth and the predicted value range, and finally perform normalization processing to obtain the similarity. Set that if the similarity is greater than or equal to 0.75, it is determined that it is extremely suitable for spore growth, leading to the occurrence of large spot disease.

[0031] The Prophet-LSTM prediction method decomposes Prophet into y(t) = g(t) + s(t) + h(t) + ∈(t), where g(t) is the trend term, representing the change function for the non-periodic trend in the time series, s(t) is the periodic term, used to obtain the periodic pattern in the time series, and its unit is week or year; h(t) is the holiday term, representing the impact of holidays on the time series data, and ∈(t) is the error term, used to represent the fluctuations not predicted in the model.

[0032] Construct the time series model of the trend term g(t): g(t) = (k + a(t) T δ)t + (m + a(t) T γ), where k represents the growth rate, the time point of the growth rate change is recorded as the change point c; δ represents the change amount of the growth rate; m represents the bias; a(t) is a time series generated by a binary function. If the current time point is equal to the change point, the change point position c takes 1, otherwise it takes 0, where j represents the position of the change point; γ is a time series of the same length as a(t), and each value in this time series is: γ j =-c j δ j δ j represents the change amount of the growth rate corresponding to the jth change point position.

[0033] The periodic term s(t) is constructed through the Fourier series:

[0034]

[0035] Use the parameter vector β to represent the coefficients in the formula: β = (a 1 , b 1 ,... a n , b n ), T where a n , b n are all coefficients to be fitted, N represents the order of the Fourier transform; L is the time series period;

[0036] The time series model corresponding to the holiday term h(t):

[0037]

[0038] Z(t) = [I(t ∈ D 1 )...I(t ∈ D n )]

[0039] h(t) = Z(t)κ, κ ~ Normal(0, v)

[0040] where n represents the number of holidays, K i represents the impact value of the i-th holiday, D n represents the preset time period before and after the n-th holiday; v represents the standard deviation, which is a preset value. When the standard deviation is larger, it means the impact of the holiday on the model is greater; when the value is smaller, it means the impact on the model is smaller. Here, we can ignore the impact of holidays on the model because meteorological data is collected regardless of holidays;

[0041] According to the above calculation method, the predicted value y(t) of the Prophet model is finally obtained, and then the processed y(t) value is input into the LSTM model for data training.

[0042] Furthermore, the calculation method of the LSTM network includes a forget gate, an input gate, and an output gate. Among them, the calculations of the three gates respectively control the deletion or addition of information in the unit;

[0043] Forget gate: It mainly determines which information to discard from the previous unit. According to the input x at the current moment t and the hidden state h at the previous moment t-1 it outputs a number between 0 and 1 and determines the proportion of the previous unit C t-1 passed to the current unit state C t . 1 means completely retaining C t-1 , 0 means completely deleting C t-1 . The calculation mathematical formula can be expressed as:

[0044] f t = σ(W f ·[h t-1 , x t + b f )

[0045] where x t is the sequence encoding corresponding to the current time point, h t-1 is the output of the hidden layer at the previous time point, b f is the bias term of the forget gate, W f is the weight matrix of the forget gate, and σ is the sigmoid function;

[0046] Input gate: It determines which information to store in the current unit state; this includes two parts. First, the sigmoid layer of the input gate determines the proportion to be updated. The input for this step is still the input x at the current momentt and the hidden state h at the previous moment t-1 , output a number between 0 and 1; Next, the hyperbolic tangent layer creates a new candidate value vector, combining the two is the final update method of the current cell state, and its formula can be expressed as:

[0047] i t =σ(W i ·[h t-1 , x t +b i )

[0048]

[0049] Fusing the two formulas is the final update of the current cell state, that is, multiplying the old state C t-1 by f t , and then adding Its final formula can be expressed as:

[0050] Output gate, with the current cell state, we also need to know how to obtain the current hidden state from the current cell state. Similarly, use a sigmoid layer, and the input is still the input x at the current moment t and the hidden state h at the previous moment t-1 , output a number between 0 and 1; Then pass the cell state through tanh, and then multiply it by the value of the output gate to obtain the hidden state;

[0051] The calculation formula of the output gate can be expressed as: o t =σ(W o [h t-1 , x t +b 0 )

[0052] Calculation method of the hidden layer state: h(t)=o t *tanh(C t )

[0053] Through the inference calculation of the Prophet-LSTM model, calculate the similarity between the predicted result value and the actual value. The calculation of similarity first calculates the Euclidean distance, then performs normalization processing, and finally calculates the score, that is, the probability of occurrence is evaluated.

[0054] Compared with the prior art, the beneficial effects of the invention are as follows: The device can collect environmental data in real time, analyze the data using an advanced time series analysis model, predict the occurrence probability of northern leaf blight of corn in advance, and provide scientific disease control solutions for farmers; the IoT acquisition host of this device can collect meteorological data in real time and analyze the data on the platform; quickly and accurately predict the occurrence index of northern leaf blight of corn, and provide a more scientific prediction scheme for agricultural production. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments read in conjunction with the accompanying drawings.

[0056] Figure 1 It is a schematic structural diagram of the present invention.

[0057] Figure 2 It is data processing of the present invention Figure 1 .

[0058] Figure 3 It is data processing of the present invention Figure 2 .

[0059] Figure 4 It is data processing of the present invention Figure 3 .

[0060] Figure 5 It is a schematic diagram of the prediction result of the present invention Figure 1 .

[0061] Figure 6 It is a schematic diagram of the prediction result of the present invention Figure 2 . DETAILED DESCRIPTION OF THE EMBODIMENTS

[0062] The present invention will be further described in detail below through embodiments. The embodiments are only used to illustrate the present invention and do not limit the scope of the present invention.

[0063] Northern leaf blight of corn, also known as stripe disease, coal spot, leaf blight, leaf spot disease, etc., is a common fungal disease in corn production. Pathogen characteristics: The pathogen of northern leaf blight of corn is Exserohilum turcicum, belonging to the Deuteromycotina, Hyphomycetes, Exserohilum. Its conidiophores grow out from the stomata, are dark brown, solitary or in clusters of 2 - 6, unbranched, erect or with a knee-like bend at the upper part, with 2 - 6 septa, the basal cells are enlarged and darker in color, gradually lighter towards the tip, and there are obvious scar marks at the tip or the bend. It mainly damages the leaves of corn, and also damages the leaf sheaths and husks severely when it is serious. The lower leaves of the plant are diseased first, and then it spreads upward. The lesions are long fusiform, grayish brown or yellowish brown, 5 - 10 cm long and about 1 cm wide. Some lesions are even larger, or several lesions are connected into large irregular necrotic spots, and the leaves are scorched severely when it is serious.

[0064] The occurrence of northern leaf blight of maize is mainly affected by the external environment suitable for disease occurrence, which is the main factor determining whether pests and diseases can prevail and break out. Therefore, disease early warning mainly involves collecting, organizing, and analyzing information such as meteorological conditions and crop varieties at the initial stage of disease occurrence, and then predicting whether the disease will occur on a large scale within a certain period of time.

[0065] In the occurrence and development process of northern leaf blight of maize, meteorological factors play a crucial role. On the one hand, the activities of hyphae and spores, which serve as infection sources, are greatly influenced by meteorological conditions such as external temperature, relative humidity, precipitation, wind, and light. Once these meteorological factors fluctuate, the behavior of spores will change significantly. For example, suitable temperature, humidity, and precipitation conditions can prompt spores to germinate and spread rapidly, while conversely, they may inhibit their activity. On the other hand, the overall development trend and prevalence of the disease are also closely related to the external meteorology. It is clear that for the short-term outbreak and large-scale prevalence of northern leaf blight of maize, various small and medium-scale meteorological factors, especially temperature and precipitation, play a decisive role. Because of the huge influence of meteorological factors on the disease, the early warning model of northern leaf blight of maize constructed through systematic and in-depth analysis of these key meteorological factors is fully in line with the development law of the disease itself and has sufficient scientific rationality. The early warning conclusions derived from such a model naturally also have a relatively high degree of accuracy and can provide a strong reference basis for disease prevention and control. In addition, it is worth noting that both the occurrence process of the disease and the change process of the climate show a certain time continuity. This continuity means that under normal circumstances, sudden accidental external conditions will not cause a sharp change in the occurrence steps of the disease and the evolution trend of the climate. Based on this characteristic, it is feasible to use meteorological conditions to give early warning of the occurrence of northern leaf blight of maize in practical operations and is expected to become an effective means to ensure the stable development of the maize planting industry.

[0066] An Internet of Things early warning device for northern leaf blight of maize, comprising:

[0067] A meteorological data collection sensor 100, used to collect key meteorological data such as air temperature, relative humidity, and rainfall; it is installed on a vertical pole in the farmland, about 1.5 meters above the ground, to ensure that it is not affected by ground weeds and crops. The sensor includes a temperature probe, a humidity probe, and a rain gauge, which are respectively used to collect air temperature, relative humidity, and rainfall data. Each sensor communicates with the data collection host through a wireless module to transmit data in real time.

[0068] A power supply system, used to provide stable power support for the device; a solar power supply system is adopted, equipped with a solar panel and a storage battery. The solar panel provides the electricity required for the daily operation of the entire device, and the storage battery powers the device in rainy weather or at night to ensure the continuous and stable operation of the device.

[0069] The vertical pole 300 is used to fix the equipment and facilitate field installation. It is made of high-strength metal and can stably support the meteorological data collection sensor 100 and the data collection host 200. The bottom of the vertical pole is provided with installation holes, which are convenient for installation in different types of soil to ensure the stability of the equipment under harsh weather conditions.

[0070] The data collection host 200 is responsible for receiving sensor data and uploading it to the intelligent agriculture Internet of Things platform. It is installed on the top of the vertical pole and has the functions of waterproof and dustproof. The data collection host 200 internally integrates a wireless communication module (such as 4G / 5G module), a data processing module and a memory. The data collection host is responsible for receiving sensor data and uploading it to the intelligent agriculture Internet of Things platform through the wireless network.

[0071] Data is collected by the meteorological data collection sensor, and this data is uploaded to the intelligent agriculture Internet of Things platform through the data collection host. The intelligent agriculture Internet of Things platform will compare and analyze the data, give the probability of the occurrence of northern corn leaf blight, and issue an early warning in advance. Specifically, the meteorological data collection sensor is used to monitor the farmland environment data in real time, and the data collection host uploads the collected data to the intelligent agriculture Internet of Things platform. The platform uses the Prophet time series model to analyze the data and predict the probability of the occurrence of northern corn leaf blight. Once a high risk of disease occurrence is detected, warning information will be immediately released to users through multiple channels to help farmers take prevention and control measures in time.

[0072] Among them, data collection: collect environmental data in real time through the meteorological data collection sensor;

[0073] Data transmission: upload the collected data to the intelligent agriculture Internet of Things platform through the data collection host;

[0074] Data analysis: The platform compares and analyzes the data and uses the time series Prophet-LSTM model for prediction;

[0075] Warning release: According to the analysis results, the platform gives the probability of the occurrence of northern corn leaf blight and displays the warning information to users through the visualization interface.

[0076] Specific working mode: Real-time monitoring: The equipment works continuously for 24 hours to continuously collect environmental data;

[0077] Automatic analysis: The platform automatically processes and analyzes the received data;

[0078] Precise warning: Based on the high-precision prediction of the Prophet model, provide reliable warning information;

[0079] Multi-channel notification: Notify users through multiple methods such as mobile APP, web version, WeChat, etc.;

[0080] Visual display: The prediction results and warning information are presented in the form of charts and text for easy understanding by users.

[0081] Among them, the data collection process includes the following steps:

[0082] 1) Initialization: After the device starts up, it automatically initializes each sensor and performs self-check.

[0083] 2) Data collection: The meteorological data collection sensors collect environmental data at a set frequency (such as once per minute).

[0084] 3) Data processing: The data collection host preprocesses the collected data, including operations such as denoising and normalization.

[0085] 4) Data transmission: The processed data is transmitted to the intelligent agriculture Internet of Things platform through the wireless network module. 2) The data transmission process includes the following steps:

[0086] Data packaging: The preprocessed data is packaged in a predetermined format, with a timestamp and device ID attached.

[0087] Network connection: The wireless module establishes a connection with the Internet of Things platform to ensure stable signal.

[0088] Data upload: The data packet is uploaded to the platform server through the HTTPS protocol.

[0089] Confirmation mechanism: After the platform server receives the data, it returns a confirmation message to ensure the integrity and reliability of data transmission.

[0090] The data analysis process includes the following steps:

[0091] Data reception: The Internet of Things platform receives data from each terminal.

[0092] Data storage: The received data is stored in the database for subsequent query and analysis.

[0093] Data cleaning: The data is further cleaned and preprocessed to remove outliers and noise.

[0094] Model input: The cleaned data is input into the Prophet time series model.

[0095] Model training: The Prophet model is trained using historical data to optimize the model parameters.

[0096] Prediction analysis: The model makes predictions based on the input data and calculates the occurrence probability of northern leaf blight of corn.

[0097] Result storage: Store the prediction results in the database and generate warning messages.

[0098] The warning release process includes the following steps:

[0099] Warning generation: Generate warning messages of different levels (such as low risk, medium risk, high risk) according to the prediction results;

[0100] Information release: Push the warning messages to the user terminals (such as mobile APPs, web versions) through the Internet of Things platform;

[0101] Visual display: Display the warning messages in the form of charts or text on the user terminals, which is intuitive and easy to understand;

[0102] Notification reminder: Send warning notifications through apps, WeChat, etc. to ensure that users can obtain warning messages in a timely manner.

[0103] Among them, the prediction model for the lesion area of northern corn leaf blight:

[0104] Y = 1 / (4.4115*10-4+(0.0614exp(0.1253t - 0.2159x - 0.0052y - 0.0079z)))

[0105] Note: Y is the lesion area of northern corn leaf blight; t is the number of days after sowing; x, y, and z are the average temperature, relative humidity, and rainfall

[0106] This model uses real-time meteorological data, namely temperature, relative humidity, and precipitation data, as well as the growth time of corn as input parameters. Based on such comprehensive and targeted input conditions, this model can achieve a higher level of accuracy and reliability when generating warning conclusions, and can provide a more powerful decision-making basis for the prevention and control of northern corn leaf blight.

[0107] The prediction steps using the time series Prophet-LSTM model include the following steps:

[0108] 1) The Prophet-LSTM model predicts by modeling from dimensions such as air temperature, air humidity, soil temperature, soil humidity, soil ph, and wind speed. The Internet of Things warning devices for northern corn leaf blight upload data to the server database, which is obtained through means such as SQL queries, web crawlers, and interfaces. The time interval for the device sensors to upload data depends on the actual situation. Here, our device sensors upload a piece of data every about 10 minutes. Use pandas to read the data of the past year and divide it into multiple dimensions according to time points to form a format that conforms to Prophet. Taking air temperature as an example: First, take out the maximum and minimum values of the time node data, and then calculate the average value every day, such as Figure 2As shown below, where date represents the time and date, the lt column represents the highest temperature of the day, ht represents the current lowest temperature, weather represents the weather, and tavg represents the average temperature of the day;

[0109] 2) Process the sample data into a format suitable for Prophet, that is, rename the field column names, as Figure 3 shown;

[0110] 3) Finally, input the processed data into the model, and obtain the Prophet model through the time series model constructed by training with the dataset; create a data frame of the dates for the next month and then obtain the predicted data; the prediction results are as Figure 4 shown:

[0111] 4) Calculate the residuals between the actual values and the predictions and input them into the LSTM model for connection.

[0112] Residual calculation formula: where y t represents the actual value, represents the predicted value.

[0113] Construct an LSTM model, and use the data processed by Prophet as samples to construct a sequence encoding X t that conforms to the input format of the LSTM model. Train the LSTM model through the constructed sequence encoding, specifically including:

[0114] Through the sequence encoding X t and the output h t-1 of the hidden layer at the previous time point, calculate the forget gate f t in the LSTM model, the input gate i t and the output gate o t , and calculate the neuron memory reference value at time point t through the tanh function; update the old cell state information of the LSTM model through the forget gate f t and the input gate i t ; calculate the output of the neuron in the hidden layer at time point t through the updated old cell state information and the output gate o t ;

[0115] Predict indicators such as air temperature and humidity, wind speed, soil temperature and humidity, etc. in the next cycle through Prophet-LSTM, input the predicted values into the BP neural network to obtain the comprehensive predicted values, calculate the Euclidean distance according to the temperature and humidity range most suitable for spore growth and the predicted value range, and finally perform normalization processing to obtain the similarity. Set that if the similarity is greater than or equal to 0.75, it is determined that it is extremely suitable for spore growth, leading to the occurrence of large spot disease;

[0116] The Prophet-LSTM prediction method can decompose Prophet into y(t) = g(t) + s(t) + h(t) + ∈(t), where g(t) is the trend term, representing the change function for the non-periodic trend in the time series, s(t) is the periodic term, used to obtain the periodic pattern in the time series, and its unit is week or year; h(t) is the holiday term, representing the impact of holidays on the time series data, and ε(t) is the error term, used to represent the fluctuations not predicted in the model.

[0117] Construct a time series model for the trend term g(t): g(t) = (k + a(t) T δ)t + (m + a(t) T γ), where k represents the growth rate, and the time point of the growth rate change is denoted as the change point c; δ represents the change amount of the growth rate; m represents the bias; a(t) is a time series generated by a binary function. If the current time point is equal to the change point, the change point position c takes 1, otherwise it takes 0, where j represents the position of the change point; γ is a time series of the same length as a(t), and each value in this time series is: γ j = -c j δ j δ j represents the change amount of the growth rate corresponding to the j-th change point position;

[0118] The periodic term s(t) is constructed by Fourier series:

[0119]

[0120] Use the parameter vector β to represent the coefficients in the formula: β = (a 1 , b 1 ,... a n , b n ) T , where a n , b n are all coefficients to be fitted, N represents the order of the Fourier transform; L is the time series period;

[0121] The time series model corresponding to the holiday term h(t):

[0122]

[0123] Z(t) = [I(t ∈ D 1 )... I(t ∈ D n )]

[0124] h(t) = Z(t)κ, κ ~ Normal(0, v)

[0125] where n represents the number of holidays, K iRepresents the impact value of the $i$-th festival, $D$ n Represents the preset time period before and after the $n$-th holiday; $v$ represents the standard deviation, which is a preset value. When the standard deviation is larger, it indicates that the holiday has a greater impact on the model; when the value is smaller, it indicates a smaller impact on the model. Here, we can ignore the impact of holidays on the model because meteorological data is collected regardless of holidays.

[0126] According to the above calculation method, the predicted value $y(t)$ of the Prophet model is finally obtained, and then the processed $y(t)$ value is input into the LSTM model for data training. The calculation method of the LSTM network mainly includes the forget gate, input gate, and output gate, where the calculations of the three gates respectively control the deletion or addition of information in the cell.

[0127] Forget gate: mainly determines which information to discard from the previous cell. According to the input $x$ at the current time t and the hidden state $h$ at the previous moment t-1 outputs a number between 0 and 1 and determines the proportion of the previous cell state $C$ t-1 passed to the current cell state $C$ t , where 1 represents completely retaining $C$ t-1 , 0 represents completely deleting $C$ t-1 , and the calculation mathematical formula can be expressed as:

[0128] $f$ t $=\sigma(W$ f $\cdot [h$ t-1 , $x$ t $]+b$ f )

[0129] where $x$ t is the sequence encoding corresponding to the current time point, $h$ t-1 is the output of the hidden layer at the previous time point, $b$ f is the bias term of the forget gate, $W$ f is the weight matrix of the forget gate, and $\sigma$ is the sigmoid function.

[0130] Input gate: determines which information to store in the current cell state. This includes two parts. First, the sigmoid layer of the input gate determines the proportion of the update. The input for this step is still the input $x$ at the current time t and the hidden state $h$ at the previous moment t-1 , and outputs a number between 0 and 1. Next, the tanh (hyperbolic tangent) layer creates a new candidate value vector, and combining the two is the final update method for the current cell state, and its formula can be expressed as:

[0131] $i$ t $=\sigma(W$ i $\cdot [h$t-1 , x t + b i )

[0132]

[0133] Fusing the two formulas gives the final updated current cell state, that is, multiplying the old state C t-1 by f t , and then adding Its final formula can be expressed as:

[0134] Output gate. With the current cell state, we also need to know how to obtain the current hidden state from the current cell state. Similarly, using a sigmoid layer, the input is still the input x at the current time t and the hidden state h at the previous time t-1 , and output a number between 0 and 1; then pass the cell state through tanh (pushing the value between -1 and 1), and then multiply it by the value of the output gate to obtain the hidden state.

[0135] The output gate calculation formula can be expressed as: o t = σ(W o [h t-1 , x t + b 0 )

[0136] The calculation method of the hidden layer state: h(t) = o t * tanh(C t )

[0137] Through the inference calculation of the Prophet-LSTM model, calculate the similarity between the predicted result value and the actual value. The similarity calculation first calculates the Euclidean distance, then performs normalization processing, and finally calculates the score, that is, the probability of occurrence is evaluated.

[0138] Taking the weather in Chifeng City, Inner Mongolia from 2021 to 2023 as an example, obtain data samples through database query, input data in dimensions such as air temperature, humidity, wind speed, soil temperature, and humidity into the Prophet-LSTM network model for training, and then predict data in dimensions such as air temperature and humidity from June to August, as Figure 5 , Figure 6 shown;

[0139] Investigation shows that in August 2023, the highest air temperature in Chifeng City, Inner Mongolia was 34 °C, the lowest temperature was 12 °C, the average temperature was 27.64, and the average relative humidity was 75.32%. Calculate the average value according to the prediction results, and then calculate the Euclidean distance between the prediction results and the investigation results The result after normalization is approximately The final similarity is s=1-0.1283=0.8717. The similarity score is very high, indicating that the environmental conditions in August are very suitable for the growth of spores and corn is very prone to disease. The actual survey results show that the spore concentration of corn leaf blight in Chifeng City, Inner Mongolia in August 2023 has reached a peak, with an average concentration of 1002.93mL, and there are more diseased plants in the experimental range.

[0140] Euclidean distance calculation formula:

[0141] Normalization:

[0142] Similarity score calculation: S = 1-d norm .

[0143] The influencing conditions include:

[0144] Meteorological factors: temperature, relative humidity, rainfall, etc. directly affect the germination and spread of pathogen spores;

[0145] Crop varieties: Different varieties have different resistance to diseases;

[0146] Planting density: High-density planting can easily lead to the spread of diseases;

[0147] Field management measures: Management measures such as fertilization and irrigation affect the health of plants.

[0148] Climate conditions are the main factors affecting the occurrence of corn leaf blight and the key to the spread of corn leaf blight. When the average daily temperature remains at 20-25℃, when the rainfall exceeds 30mm, and the relative humidity is greater than 80%, it is very easy to cause the outbreak and spread of corn leaf blight. The optimum temperature during the infection period is 20℃. Under suitable climatic conditions, conidia will be produced on the lesions within 10 to 14 days. These spores are spread through media such as air and rain, thereby causing re-infection. In June, when the temperature and humidity conditions are suitable, conidia are spread to the leaves of the plants with air flow and rain splashes, causing the disease. In rainy years, the disease spreads faster.

[0149] Different corn varieties have relatively large differences in disease resistance. Planting susceptible varieties or large-scale cultivation of the same variety is the main reason for the outbreak. Generally speaking, late-maturing varieties have higher disease resistance than early-maturing varieties. Due to the large number of physiological subspecies of the disease, most of the currently promoted varieties are specialized varieties, and their resistance decreases after a few years of promotion. Therefore, the most economical and effective way to prevent and control corn leaf blight is to cultivate disease-resistant varieties.

[0150] During the process of corn planting, tillage patterns often lead to a significant increase in the incidence of northern corn leaf blight, and the disease condition is also more severe. In contrast, the adoption of intercropping or crop rotation planting patterns can alleviate this phenomenon. However, with the continuous expansion of the corn planting scale, this planting method is often difficult to implement, which results in the continuous accumulation of the initial pathogen source in the fields, and the incidence of northern corn leaf blight also shows an upward trend. In addition, corn planted in low-lying areas, or in fields with a relatively high planting density, leads to mutual shading between plants, affecting air circulation and photosynthesis, and affecting the evaporation of soil moisture, resulting in an increase in humidity. These conditions provide a favorable environment for the outbreak of northern corn leaf blight.

[0151] The preventive spraying strategy has shown more significant effects in the control of northern corn leaf blight compared to the therapeutic spraying. Approximately 10 days before the leaves are infected is the optimal control period. If a better control effect is desired, the pesticide should be immediately sprayed when water-soaked spots first appear on the leaves. Using 18.7% fluxapyroxad·propiconazole, diluted 1300 times, and spraying once a week, the application effect is the best. In addition, the experimental results in Guanghe County, Gansu Province show that 32% propiconazole·azoxystrobin suspension concentrate, 80% mancozeb wettable powder, and 30% metconazole·propiconazole emulsifiable concentrate all perform well in the control of northern corn leaf blight. The specific dosage is 35 mL, 50 g, and 15 mL per 0.067 hectares respectively, which are mixed with 30 kg of water for spray control. It is recommended to start spraying at the initial stage of the disease, spray once every 7 days, and the best effect can be achieved by continuous use three times. Advance the prevention and control period to the large flare-up stage of corn, and adopt the mixed combination of 18.7% propiconazole·azoxystrobin suspension concentrate and 5% chlorantraniliprole suspension concentrate, which can not only effectively control northern corn leaf blight, but also achieve the integrated prevention and control of multiple pests and diseases, thus achieving the effect of one spray for multiple protections.

[0152] Although the specific implementation manners of the present invention are described above, those skilled in the art should understand that this is only an example. The protection scope of the present invention is defined by the appended claims. Without departing from the principles and essence of the present invention, those skilled in the art can make various changes or modifications to these implementation manners, but these changes and modifications all fall within the protection scope of the present invention.

Claims

1. An Internet of Things corn leaf spot early warning device, characterized in that: include: Meteorological data collection sensors are used to collect key meteorological data such as air temperature, relative humidity, and rainfall; The power supply system adopts a solar power supply system equipped with solar panels and batteries; The data collection host is responsible for receiving sensor data and uploading it to the smart agriculture IoT platform; The data is collected through meteorological data collection sensors and uploaded to the smart agriculture IoT platform through the data collection host. The smart agriculture IoT platform will compare and analyze the data, give the probability of corn leaf blight, and issue early warning. Data collection: Real-time collection of environmental data through meteorological data collection sensors; Data transmission: upload the collected data to the smart agriculture IoT platform through the data collection host; Data analysis: The platform compares and analyzes the data and uses the time series Prophet-LSTM model for prediction; Warning release: Based on the analysis results, the platform gives the probability of occurrence of corn leaf blight and displays the warning information to users through a visual interface.

2. The Internet of Things corn leaf blight early warning device according to claim 1, characterized in that: Using time series The prediction steps of the Prophet-LSTM model include the following steps: 1) The Prophet-LSTM model predicts and models the data from the dimensions of air temperature, air humidity, soil temperature, soil humidity, soil pH, wind speed, etc. The IoT corn leaf blight early warning device uploads data to the server database and obtains it through SQL query, web crawler, interface and other means. The time interval for the device sensor to upload data depends on the actual situation. Here, the device sensor uploads a piece of data every 10 minutes or so. Pandas is used to read the data of the past year, and multiple dimensions are divided according to the time point to form a format that conforms to Prophet; 2) Process the sample data into a format that conforms to Prophet; 3) Finally, the processed data is passed into the model, and the Prophet model is obtained by training the time series model constructed by the data set; a data frame of the next month's date is created to obtain the predicted data; 4) Calculate the residual between the actual value and the prediction and input it into the LSTM model for connection; Residual calculation formula: where y t Indicates the actual value, Represents the predicted value.

3. The Internet of Things corn leaf blight early warning device according to claim 2, characterized in that: Build an LSTM model, use the Prophet processed data as samples to construct a sequence code Xt that conforms to the LSTM model input format, and train the LSTM model through the constructed sequence code, including: By encoding X t and the output h of the hidden layer at the previous time point t-1 , calculate the forget gate f in the LSTM model t , input gate i t And the output gate o t , and calculate the memory reference value of the neural point at time point t through the tanh function; through the forget gate f t AND input gate i t Update the old unit state information of the LSTM model; pass the updated old unit state information and the output gate o t Calculate the output of the hidden layer neurons at time point t; Prophet-LSTM is used to predict the air temperature and humidity, wind speed, soil temperature and humidity in the next cycle, and the predicted values ​​are input into the BP neural network to obtain a comprehensive predicted value. The Euclidean distance is calculated based on the temperature and humidity range that is most suitable for spore growth and the predicted value range. Finally, normalization is performed to obtain the similarity. If the similarity is greater than or equal to 0.75, it is determined to be very suitable for spore growth, leading to the occurrence of large spot disease. The Prophet-LSTM prediction method decomposes Prophet into y(t)=g(t)+s(t)+h(t)+∈(t), where g(t) is the trend term, which represents the change function of the non-periodic trend in the time series, s(t) is the period term, which is used to obtain the periodic law in the time series, and its unit is week or year; h(t) is the holiday term, which represents the impact of holidays on time series data, and ∈(t) is the error term, which is used to represent the fluctuations that are not predicted in the model; Construct the trend term g(t) time series model: g(t) = (k+a(t) T δ)t+(m+a(t) T γ), where k represents the growth rate, and the time point at which the growth rate changes is recorded as the change point c; δ represents the change in the growth rate; m represents the bias; a(t) is a time series generated by a binary function. If the current time point is equal to the change point, the change point position c takes 1, otherwise it takes 0, where j represents the position of the change point; γ is a time series of the same length as a(t), and each value in the time series is: γ j =-c j δ j , δ j Indicates the change in growth rate corresponding to the j-th change point position; The periodic term s(t) is constructed through the Fourier series: Use parameter vector β to represent the coefficients in the formula: β=(a1,b1,...a n ,b n ) T , where a n , b n are the coefficients that need to be fitted, N represents the order of Fourier transform; L is the time series period; The time series model corresponding to the holiday term h(t): Z(t)=[I(t∈D1)...I(t∈D n )] h(t)=Z(t)κ,κ~Normal(0,v) Where n represents the number of holidays, K i represents the impact value of the i-th festival, D n represents the preset period before and after the nth holiday; v represents the standard deviation, which is a preset value. When the standard deviation is larger, it means that the holiday has a greater impact on the model; when the value is smaller, it means that the impact on the model is smaller; we can ignore the impact of holidays on the model here, because meteorological data is collected regardless of holidays; According to the above calculation method, the predicted value y(t) of the Prophet model is finally obtained, and then the processed y(t) value is input into the LSTM model for data training.

4. The Internet of Things corn leaf blight early warning device according to claim 3, characterized in that: The LSTM network calculation method includes forget gate, input gate, and output gate, where the calculation of the three gates respectively controls the desire to delete or add information; Forget gate: mainly decides which information to discard from the previous unit, based on the current input x t and the hidden state h at the previous moment t-1 Output a number between 0 and 1 and determine the previous unit C t-1 Pass to the current unit state C t The ratio of 1 represents complete retention of C t-1 , 0 means complete deletion of C t-1 , where the mathematical formula can be expressed as: f t =σ(W f ·[h t-1 ,x t ]+b f ) where x t is the sequence code corresponding to the current time point, h t-1 is the output of the hidden layer at the previous time point, b f is the bias term W of the forget gate f is the forget gate weight matrix, σ is the sigmoid function; Input gate: determines what information to store in the current cell state; this includes two parts. First, the sigmoid layer of the input gate determines the proportion to be updated. The input at this step is still the current input x t and the hidden state h at the previous moment t-1 , output a number between 0 and 1; Next is the tangent hyperbolic layer to create a new candidate value The vector of , combining the two is the final update method of the current unit state, and its formula can be expressed as: i t =σ(W i ·[h t-1 ,x t ]+b i ) The combination of the two formulas is the final update of the current cell state, that is, the old state C t-1 Multiply by f t , then add The final formula can be expressed as: Output gate. With the current cell state, we also need to know how to output the current hidden state from the current cell state. Similarly, we use a sigmoid layer, and the input is still the current input x. t and the hidden state h at the previous moment t-1 , outputs a number between 0 and 1; then puts the cell state through tanh and then multiplies it by the value of the output gate to get the hidden state; The output gate calculation formula can be expressed as: t =σ(W o [h t-1 ,x t ]+b0) Calculation method of hidden layer state: h(t)=o t *tanh(C t ) Through the inference calculation of the Prophet-LSTM model, the predicted result value and the actual value are calculated for similarity. The similarity calculation first calculates the Euclidean distance, then performs normalization, and finally calculates the score, which is the probability of occurrence.

Images