Deep learning-based farmland disease and pest monitoring and prevention method, system and device

Abstract

The invention provides a deep learning-based farmland disease and pest monitoring and prevention method, system and device, and the method comprises the steps: carrying out the data collection of a target farmland through a plurality of sensors, obtaining the sensor data corresponding to the target farmland, carrying out the multi-level data preprocessing, obtaining the structured sensor data of the target farmland, and carrying out the multi-level data preprocessing of the structured sensor data; and performing data analysis on the crop growth image to obtain growth parameters of crops, performing pest recognition, positioning and counting on the pest situation information image through a deep learning algorithm to obtain a pest population dynamic state, constructing a pest occurrence prediction model to obtain a pest occurrence probability, and generating early warning information of a pest control opportunity. And finally obtaining a farming management decision scheme of the target farmland. According to the method, environmental factors, crop growth conditions and insect pest information are comprehensively considered, the occurrence probability of target farmland diseases and insect pests is predicted, corresponding prevention and control measures are provided, and reference is provided for farmers and managers.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent agricultural monitoring technology, and in particular to a method, system and device for monitoring and controlling farmland pests and diseases based on deep learning. Background Art

[0002] Farmland pest and disease monitoring and prevention is an important technical field in agricultural production. By accurately monitoring the occurrence and spread of pests and diseases and combining scientific prevention and control measures, the healthy growth of crops and the sustainable development of the agricultural ecological environment can be guaranteed. In recent years, with the development of remote sensing technology, drones, big data and artificial intelligence technology, farmland pest and disease monitoring has gradually moved towards intelligence, real-time and precision, but the current farmland pest and disease monitoring solutions mainly have the following problems:

[0003] (1) The monitoring method is relatively simple, and the data collection on farmland pests and diseases is not comprehensive enough;

[0004] (2) They usually focus on a single indicator, such as insect density or environmental parameters, and lack a comprehensive assessment of the overall status of the farmland system;

[0005] (3) There is a lack of scientific and quantitative pest and disease assessment models, and the overall understanding of the pest and disease situation in farmland is not accurate enough, making it difficult to output accurate and effective prevention and control decisions.

[0006] Therefore, we need to develop methods, systems and devices for monitoring and controlling farmland pests and diseases based on deep learning, which can comprehensively collect farmland data, fully consider and evaluate factors related to farmland pests and diseases, and output corresponding pest and disease control measures. Summary of the invention

[0007] The purpose of the present invention is to provide a method, system and device for monitoring and controlling farmland pests and diseases based on deep learning, so as to solve the problems mentioned in the above background technology that the data collection of existing farmland monitoring schemes is not comprehensive enough, the evaluation indicators are relatively single, and the accuracy of pest and disease monitoring is poor.

[0008] To achieve the above objectives, on the one hand, the present invention provides a method for monitoring and controlling farmland pests and diseases based on deep learning, and the method is specifically as follows:

[0009] Perform multi-level data preprocessing based on sensor data of the target farmland to obtain structured sensor data of the target farmland, wherein the structured sensor data includes environmental parameters, soil parameters, crop growth images, insect information images, pest population density, and farming operation data;

[0010] Performing data analysis on the crop growth images to obtain crop growth parameters, the growth parameters including crop canopy height, leaf area index and chlorophyll content, plotting the growth parameters at different growth periods into time series curves, and introducing a standard growth curve as a reference, comparing the difference between the actual growth process and the ideal model, and obtaining a crop growth diagnosis report for the target farmland;

[0011] The acquisition of the chlorophyll content includes:

[0012] Extract the chlorophyll content, that is, the relative content of chlorophyll in the leaves; convert the crop growth image from the RGB space to the HSV space, take the yellow-green pixels in the range of 45° to 65° in the hue channel as sampling objects, and estimate the chlorophyll content in combination with the following formula: , where H, S, V are the average values ​​of hue, saturation and brightness of the yellow-green pixels selected from the image, and C is the offset. They respectively represent the contribution or influence of hue, saturation and brightness on the prediction of chlorophyll content, reflecting the importance and sensitivity of each HSV component in estimating chlorophyll content.

[0013] Pests are identified, located and counted on the insect information image through a deep learning algorithm to obtain pest population dynamics, and a pest occurrence prediction model is constructed based on multi-source data. The probability of pest occurrence is obtained through the pest occurrence prediction model, and early warning information including the timing of pest control is generated;

[0014] Based on the sensor data, the growth parameters of the crops, the crop growth diagnosis report, the pest population dynamics, the probability of pest occurrence, and the timing of pest control, the agricultural management decision plan for the target farmland is obtained, and the selection and ratio of agricultural inputs are optimized. The agricultural inputs include seeds, fertilizers, and pesticides. The economic benefits of different agricultural management decision plans are evaluated to provide reference for farmers and managers.

[0015] Based on the above scheme, the acquisition of the leaf area index includes:

[0016] Extract the ratio of the total leaf area per unit ground area to the ground area as the leaf area index: take S blocks of pixels in the central area of ​​the crop growth image as samples, extract the proportion r of green pixels in the S blocks of pixels, and calculate the leaf area index by the following formula: LAI=k*r / (1-r), where the coefficient k is adjusted according to the crop type and growth stage;

[0017] Based on the above scheme, a dynamic correction mechanism based on temperature T and light L is used to reduce the impact of environmental conditions on the chlorophyll content. The dynamic correction mechanism is: , where the environmental correction function is defined as:

[0018]

[0019]

[0020] in is the optimum temperature; For the optimal light intensity, is the fitting coefficient.

[0021] Based on the above scheme, based on the daily and seasonal variation characteristics of the chlorophyll content, the chlorophyll content is corrected by a time series dynamic correction factor, and the formula is as follows:

[0022]

[0023] in is the sampling time in hours, is the daily variation coefficient, is the annual variation coefficient.

[0024] Based on the above scheme, the multi-level data preprocessing includes first-level preprocessing and second-level preprocessing. The first-level preprocessing includes:

[0025] The sensor data is subjected to a first layer of preprocessing through the edge computing layer to obtain first preprocessed data, including: for the environmental parameters, removing random noise interference through threshold filtering and median smoothing algorithm; for the soil parameters, removing obvious outliers, detecting the integrity of the data, and performing aggregation compression in units of fixed time intervals; for the crop growth image, extracting the crop area and removing the image background by using a target detection algorithm based on deep learning; for the insect information image, highlighting the texture features of the pest target through image enhancement and feature engineering methods;

[0026] The second layer preprocessing includes: reporting the first preprocessed data to the data center layer through the message queue telemetry transmission protocol, the access module parses the data packet and verifies the signature to ensure that the data source is credible before performing the second layer preprocessing. The second layer preprocessing includes: unifying the time format and standardizing the field naming of the first preprocessed data to form the structured sensor data as the second preprocessed data.

[0027] Based on the above scheme, the construction of the pest occurrence prediction model includes:

[0028] Based on the environmental parameters, the growth parameters of the crops, the pest population dynamics and the farming operation data, through correlation analysis and factor analysis methods, key characteristic factors closely related to the occurrence of pests are screened out, including meteorological factors, crop growth indicators and farming factors, and a multi-dimensional input feature vector is constructed;

[0029] Based on the multi-dimensional input feature vector, corresponding historical observation value sequence data is collected to form a complete pest prediction training sample set as input data of the pest occurrence prediction model, and the time span of the training sample set covers the entire process of pest occurrence;

[0030] A nonlinear regression model of insect quantity and the key characteristic factors is constructed using a time series-based machine learning algorithm as the insect pest occurrence prediction model, wherein the input variables are the historical observation value sequences of various characteristic factors, and the output variables include the insect quantity or insect population density of the target farmland in the next few days;

[0031] A fixed time period is set. Before the start of each time period, the historical observation value sequence data of the previous time period is used as the training sample set, and the trained pest occurrence prediction model is input to obtain the trend of insect population density changes within the time period, monitor the number of insects before the start of each time period, and predict the probability of pest occurrence and the pest level based on the trend of insect population density changes. The fixed time period includes daily, weekly, monthly, half-monthly, half-yearly, quarterly and annually.

[0032] Based on the above scheme, the probability of pest occurrence is calculated based on the time decay factor, and the formula is as follows: ,in express The probability of pest occurrence at any time, For the The key characteristic factors are The measured values ​​at the moment, including temperature, humidity, and light environment factors, is the weight coefficient of each key characteristic factor, reflecting the contribution of different key characteristic factors to the occurrence of insect pests. is the time attenuation coefficient, which is used to describe the impact of historical data on the current forecast. is the time point of data collection, is the time decay term, which changes with the time interval between the acquisition time point and the current time point t decreases with the increase of .

[0033] On the one hand, the present invention also provides a farmland pest monitoring and control system based on deep learning, comprising:

[0034] The data collection layer is used to collect data on the target farmland through a variety of sensors to obtain sensor data corresponding to the target farmland;

[0035] An edge computing layer, used for performing multi-level data preprocessing based on the sensor data to obtain structured sensor data of the target farmland, wherein the structured sensor data includes environmental parameters, soil parameters, crop growth images, insect information images, pest population density, and farming operation data;

[0036] A data center layer, used to perform data analysis on the crop growth images to obtain growth parameters of the crops, the growth parameters including crop canopy height, leaf area index and chlorophyll content, plot the growth parameters at different growth periods into time series curves, and introduce a standard growth curve as a reference to compare the difference between the actual growth process and the ideal model, and obtain a crop growth diagnosis report for the target farmland;

[0037] An intelligent analysis layer, used to identify, locate and count pests on the insect information image through a deep learning algorithm to obtain pest population dynamics, build a pest occurrence prediction model based on multi-source data, obtain the probability of pest occurrence through the pest occurrence prediction model, and generate early warning information including the timing of pest control, wherein the pest population dynamics includes pest population density, and the multi-source data includes the environmental parameters, crop growth parameters, crop growth status assessment results in the crop growth diagnosis report, and pest population dynamics;

[0038] The decision support layer is used to obtain the agricultural management decision plan of the target farmland based on the sensor data, the growth parameters of the crops, the crop growth diagnosis report, the pest population dynamics, the probability of pest occurrence, and the timing of pest control.

[0039] On the other hand, the present invention also provides a farmland pest monitoring and control device based on deep learning, comprising:

[0040] Memory, used to store programs;

[0041] The processor is used to load the program and execute the above-mentioned intelligent farmland system processing method.

[0042] Compared with the prior art, the present invention has the following advantages and effects:

[0043] (1) Through the coordinated deployment of multi-source sensors, comprehensive perception of farmland environmental parameters, crop growth information, and pest and disease dynamics is achieved. At the same time, edge computing technology is combined to improve the real-time preprocessing and transmission efficiency of data;

[0044] (2) A crop growth parameter extraction algorithm based on the HSV color space was proposed, and combined with a chlorophyll content prediction model, a comprehensive assessment of crop health status was made to provide a reference for pest and disease prediction, thus improving the comprehensiveness of monitoring considerations;

[0045] (3) Through deep learning algorithms, automatic identification and counting of pests and diseases can be achieved. In combination with environmental factors, a pest occurrence prediction model can be established to provide accurate pest and disease prediction results and output corresponding pest and disease control measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for the specific embodiments or the description of the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual scale.

[0047] Figure 1 is a flow chart of a method for monitoring and controlling farmland pests and diseases based on deep learning provided by an embodiment of the present invention;

[0048] Figure 2 is a structural diagram of a farmland pest monitoring and control system based on deep learning provided by an embodiment of the present invention;

[0049] Figure 3 It is a structural diagram of a deep learning-based farmland pest monitoring and control device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to more clearly explain the purpose, technical solutions and advantages of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. The example implementation methods can be implemented in various forms and should not be understood as being limited to the examples described herein. On the contrary, these implementation methods are provided to make the present invention more comprehensive and complete, and to fully convey the concepts of the example implementation methods to those skilled in the art.

[0051] In addition, the described features, structures or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present invention. However, those skilled in the art will appreciate that the technical solution of the present invention can be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring various aspects of the present invention.

[0052] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0053] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0054] The present invention is described in detail below in conjunction with specific embodiments:

[0055] Example 1

[0056] As attached Figure 1 As shown, Embodiment 1 of the present invention provides a method for monitoring and controlling farmland pests and diseases based on deep learning, and the specific steps of the method are as follows:

[0057] Step S1, collecting data on a target farmland through a variety of sensors to obtain sensor data corresponding to the target farmland;

[0058] Step S2: performing multi-level data preprocessing based on the sensor data to obtain structured sensor data of the target farmland, wherein the structured sensor data includes environmental parameters, soil parameters, crop growth images, insect information images, pest population density, and farming operation data;

[0059] Step S3, performing data analysis on the crop growth image to obtain a crop growth diagnosis report of the target farmland;

[0060] Step S4, identifying, locating and counting pests in the insect information image by a deep learning algorithm to obtain pest population dynamics, constructing a pest occurrence prediction model based on multi-source data, obtaining pest occurrence probability through the pest occurrence prediction model, and generating early warning information including pest control timing, wherein the pest population dynamics includes pest population density, and the multi-source data includes environmental parameters, crop growth parameters, crop growth status assessment results in a crop growth diagnosis report, and pest population dynamics;

[0061] Step S5, based on the sensor data, the growth parameters of the crops, the crop growth diagnosis report, the pest population dynamics, the probability of the pest occurrence, and the pest control timing, obtain the agricultural management decision plan for the target farmland, and combine with market data to optimize the selection and ratio of agricultural inputs, the agricultural inputs include seeds, fertilizers, and pesticides, evaluate the economic benefits of different agricultural management decision plans, and provide reference for farmers and managers.

[0062] Preferably, step S1 specifically includes: based on the target farmland for which data collection and pest monitoring is required, determining the farmland information that needs to be collected, the farmland information including the diversity of pest populations, crop coverage, soil texture, and meteorological conditions of the target farmland, and selecting corresponding sensors based on the farmland information, the sensors including:

[0063] Field high-definition camera or infrared camera: used to capture crop growth images of the target farmland and update pest population data; GPS receiver: used to record the geographical location of the collection point; weather station: used to measure meteorological conditions including temperature, humidity, and wind speed; soil moisture sensor: used to measure soil parameters including soil moisture, nutrients, and pH value; spectrometer: used to measure the spectral characteristics of crops in the target farmland; laser radar: used to measure crop growth information including structure and height;

[0064] Specifically, the farming management APP is also used to enter farming operation data into the system, and the farming operation data includes fertilization, irrigation, and pesticide use;

[0065] Specifically, the sensor data includes environmental parameters, soil parameters, crop growth images, insect information images, pest population density, and farming operation data.

[0066] Preferably, the multi-level data preprocessing in step S2 is specifically as follows:

[0067] The sensor data is subjected to a first layer of preprocessing through the edge computing layer to obtain first preprocessed data, including: for the environmental parameters, removing random noise interference through threshold filtering and median smoothing algorithm; for the soil parameters, removing obvious outliers, detecting the integrity of the data, and performing aggregation compression in units of fixed time intervals; for the crop growth image, extracting the crop area and removing the image background by using a target detection algorithm based on deep learning; for the insect information image, highlighting the texture features of the pest target through image enhancement and feature engineering methods;

[0068] The first pre-processed data is reported to the data center layer through the message queue telemetry transmission protocol, and the access module parses the data packet and verifies the signature to ensure that the data source is credible before performing the second layer pre-processing. The second layer pre-processing includes: unifying the time format and standardizing the field naming of the first and processed data to form the structured sensor data as the second pre-processed data;

[0069] It should be noted that the second preprocessed data obtained by the second layer preprocessing includes structured parameters of environmental parameters, soil parameters, crop growth images, insect information images, pest population density, and farming operation data.

[0070] Preferably, the data analysis in step S3 includes extracting growth parameters of crops from the crop growth image, wherein the growth parameters include crop canopy height, leaf area index and chlorophyll content, as follows:

[0071] Step S301, extracting the vertical height from the bottom of the leaf to the top new leaf as the crop canopy height: taking the bottom of the crop growth image as the reference line, traversing row by row along the central axis of the main stem, and determining the position of the new leaf according to the continuity of the green pixel points, when no green pixels appear for more than x consecutive rows (such as 50 rows), it is considered that the canopy top is reached, and at this time, the vertical height difference between the reference line and the row where the top new leaf is located is the crop canopy height;

[0072] Step S302, extracting the ratio of the total leaf area per unit ground area to the ground area as the leaf area index (LAI): taking S blocks of pixels in the central area of ​​the crop growth image as samples, extracting the proportion r of green pixels in the S blocks of pixels, and calculating the leaf area index by the following formula: LAI=k*r / (1-r), wherein the coefficient k is adjusted according to the crop type and growth stage, and can be obtained by fitting the measured LAI value; in this embodiment, the value of S is recommended to be 1 / 16~1 / 8 of the total pixel block.

[0073] Step S303, extracting the chlorophyll content (SPAD): the relative content of chlorophyll in the leaves; converting the crop growth image from the RGB space to the HSV space, taking the yellow-green pixels in the range of 45° to 65° in the hue (H) channel as sampling objects, and estimating the SPAD value in combination with the following formula: , where H, S, V are the average values ​​of hue, saturation and brightness of the yellow-green pixels selected from the image, and C is the offset. They represent the contribution or influence of hue (H), saturation (S) and brightness (V) on the prediction of SPAD value, reflecting the importance and sensitivity of each HSV component in estimating chlorophyll content. For example, if The absolute value of is large, which means that hue plays an important role in estimating the SPAD value. C can be regarded as a baseline adjustment or offset, which reflects the predicted SPAD value in the absence of HSV values ​​(i.e., when H, S, and V are all zero);

[0074] Furthermore, considering the impact of environmental conditions on chlorophyll content, a dynamic correction mechanism of temperature (T) and light (L) is introduced: , where the environmental correction function is defined as:

[0075]

[0076]

[0077] in is the optimum temperature, preferably 25°C; The optimal light intensity is 50,000 lux. is the fitting coefficient; the dynamic correction mechanism can effectively capture the influence of environmental changes on chlorophyll content;

[0078] Furthermore, considering the daily and seasonal variation characteristics of chlorophyll content, a time series dynamic correction factor is introduced in this embodiment to correct the chlorophyll content. The formula is as follows:

[0079]

[0080] in is the sampling time in hours, is the daily variation coefficient, is the annual variation coefficient. The temporal dynamic characteristics of chlorophyll content are characterized by periodic functions to make the prediction results more accurate.

[0081] Step S304, plotting the growth parameters of different growth periods into a time series curve, analyzing the change trend and rate of the growth parameters, introducing a standard growth curve as a reference, comparing the difference between the actual growth process and the ideal model, and judging the growth status of the crops in the target farmland; if it is found after comparison that the growth status of the crops deviates significantly during the critical growth period, and the environmental parameters also deviate from the preset interval, it is judged that there may be adverse stress, and a growth stress warning is issued, the critical growth period includes the jointing period and the flowering period, and while issuing the warning, the correlation between the growth parameters, the environmental parameters and the soil parameters is analyzed, the main controlling factors affecting the growth status of the crops are located, and the crop growth diagnosis report is formed.

[0082] Preferably, obtaining the pest population dynamics in step S4 specifically includes:

[0083] Step S4011: After fully collecting images of various types of pests, manually mark the boundary boxes and category labels of the pest areas to create a training data set;

[0084] Step S4012: Select a deep learning algorithm detection framework based on target detection to build an end-to-end detection model. The detection framework includes RCNN series, YOLO series, and SSD.

[0085] Step S4013, dividing the training data set into a training set and a validation set, and sending the sets to the detection model for training, and using the average precision mean as an indicator to evaluate the performance of the detection model;

[0086] Step S4014: Use the trained detection model to predict the newly collected insect information image to obtain the pest population dynamics, including pest category, pest location and pest population density.

[0087] Exemplarily, in this embodiment, the specific configuration and training process of the detection model can be: using a backbone network to extract the features of the various pest images, the backbone network includes DarkNet-53 and ResNet-101; setting detection heads on three feature maps of different scales, scaling the various pest images to a fixed size, including 416*416, using the training set as a data source, setting parameters such as batch size and learning rate, wherein the batch size is preferably 64, and the learning rate is preferably 0.01, performing forward propagation, and calculating the IOU loss and Classification loss, and back-propagation update weights, evaluate the mAP of the detection model on the validation set for each training round, and adjust hyperparameters appropriately according to the results, including reducing the learning rate and increasing the batch size, until the detection model fully converges; It should be noted that mAP is the average value of the accuracy of the model in detecting different types of pests and diseases, and the generalization performance of the trained detection model is evaluated through the test set images. If the mAP meets the practical requirements, it is determined as the final model parameter for subsequent pest occurrence prediction tasks. The practical requirements, in this embodiment, are preferably that the mAP reaches 85% or more;

[0088] Specifically, in order to improve accuracy and adaptability, algorithms such as target tracking and multi-view fusion are integrated on the basis of the detection model to establish pest movement trajectories for continuous frame images to reduce missed detection and false detection; the same crop area is photographed from different positions and angles, and the detection results of multiple perspectives are weighted and fused to improve the overall accuracy;

[0089] Specifically, in this embodiment, in order to evaluate the reliability of the detection model for pest identification, a reliable new evaluation of the pest population dynamics is performed based on a confidence calculation method of weighted samples, and the calculation formula is as follows:

[0090]

[0091] Wherein TP represents the number of true positive samples, that is, the number of pests correctly identified; TN represents the number of true negative samples, that is, the number of non-pest targets correctly excluded; FP represents the number of false positive samples, that is, the number of other targets mistakenly identified as pests; FN represents the number of false negative samples, that is, the number of actual pests that have not been identified; α and β are weight coefficients used to balance the impact of different types of recognition errors. Since missed detection (failure to identify pests) causes more harm than false detection (misjudging non-pests as pests), in this embodiment, α is set to be greater than β. By setting reasonable weight coefficients, including α=0.7 and β=0.3, the system can more accurately evaluate the reliability of the recognition results and provide a more reliable basis for subsequent prevention and control decisions; when the confidence level is lower than the preset confidence threshold, the system will automatically mark the corresponding area as an area requiring manual review, thereby improving the overall recognition accuracy.

[0092] It should be noted that the key to judging the occurrence trend and degree of damage of the target farmland pests in the future based on the pest population dynamics is to construct a mathematical model between the occurrence of pests and influencing factors, that is, the pest occurrence prediction model.

[0093] Preferably, the construction of the pest occurrence prediction model in step S4 is as follows:

[0094] Step S4021: Based on the environmental parameters, the growth parameters of the crops, the pest population dynamics and the farming operation data, through correlation analysis and factor analysis methods, key characteristic factors closely related to the occurrence of pests are screened out, including meteorological factors, crop growth indicators and farming factors, and a multi-dimensional input feature vector is constructed; the key characteristics corresponding to different insect species are different and need to be selected according to local conditions; the meteorological factors include accumulated temperature and precipitation, the crop growth indicators include growth potential and phenological period, the pest population dynamics include pest category, pest location and pest population density, adult number and egg quantity, and the farming factors include sowing date and density;

[0095] Step S4022: Based on the multi-dimensional input feature vector, corresponding historical observation value sequence data is collected to form a complete pest prediction training sample set as input data of the pest occurrence prediction model. The time span of the training sample set covers the entire process of pest occurrence, and the spatial distribution is representative of the region;

[0096] Step S4023: Use a time series-based machine learning algorithm to build a nonlinear regression model of insect quantity and the key characteristic factors as the insect pest occurrence prediction model, the input variables are the historical observation value sequences of various characteristic factors, and the output variables include the insect quantity or insect population density of the target farmland in the next few days.

[0097] Step S4024, set a fixed time period. Before the start of each time period, use the historical observation value sequence data of the previous time period as the training sample set, input the trained pest occurrence prediction model, obtain the trend of insect population density changes within the time period, monitor the number of insects before the start of each time period, and predict the probability of pest occurrence and the pest level based on the trend of insect population density changes; the fixed time period includes daily, weekly, monthly, half-monthly, half-yearly, quarterly and annually.

[0098] Specifically, in this embodiment, the time attenuation factor is considered to improve the accuracy of the pest occurrence prediction model, and the pest occurrence probability is calculated, and the formula is as follows: ,in express The probability of pest occurrence at any time, For the The key characteristic factors are The measured values ​​at the moment, including environmental factors such as temperature, humidity, and light, is the weight coefficient of each key characteristic factor, reflecting the contribution of different key characteristic factors to the occurrence of insect pests. is the time attenuation coefficient, which is used to describe the impact of historical data on the current forecast. is the time point of data collection, is the time decay term, which changes with the time interval between the acquisition time point and the current time point t decreases with the increase of .

[0099] Step S4025: regularly use the newly collected training sample set to calibrate the pest occurrence prediction model, and retrain the weight parameters of the pest occurrence prediction model. If extreme meteorological conditions are encountered, the training sample set is constructed using the measured data in time to correct the pest occurrence prediction model to improve the prediction accuracy; the regularity is preferably before the end of the fixed time period, and the extreme meteorological conditions include high temperature and heavy rain.

[0100] Preferably, based on the probability of occurrence of the pests and the pest level, the corresponding pesticide control is selected and the best time to apply the pesticide is determined, and by analyzing the occurrence time, peak period, and duration of the same insect species in recent years, the occurrence law of the pest of the insect species is obtained, and based on the pest occurrence law, an early warning of the timing of control is obtained; specifically, when the insect population rises rapidly to a critical value or an economic threshold, the pest control is initiated. The critical value varies depending on the insect species. For example, the number of cotton bollworms is generally more than 10 per 100 plants. The resistance of crops to pests is weak in highly sensitive growth stages, such as the bud stage of cotton, and crop growth needs to be controlled. Real-time monitoring of the progress and early warning of the prevention and control opportunities in highly sensitive growth stages; pesticide prevention and control based on meteorological conditions. If the temperature of the target farmland exceeds the preset temperature threshold or it is rainy, the prevention and control opportunity is reselected. The preset temperature threshold is preferably 25°C. Based on the mechanism of action and duration of the selected pesticide, the number of spraying times and intervals are reasonably determined; when the temperature exceeds 25°C, the pesticide concentration is too high, which can easily cause pesticide damage; it is not suitable to spray pesticides in rainy weather, which affects the efficacy of the pesticide. The system fully considers weather forecast information, takes into account both prevention and control needs and pesticide application safety, and flexibly grasps the opportunity.

[0101] It should be noted that the data obtained in step S4 include: pest population dynamics, pest occurrence probability, pest level and prevention and control opportunity warning.

[0102] It should be noted that based on the above-mentioned functions such as crop growth diagnosis and insect pest warning, the analysis results of soil moisture conditions, nutrient requirements, etc. can be further integrated to comprehensively manage and optimize decisions on the entire agricultural production process.

[0103] Preferably, the agricultural management decision plan for the target farmland obtained in step S5 is specifically:

[0104] Step S501, sowing period selection: analyzing the historical data of environmental parameters in the sensor data, selecting a time window suitable for sowing, the environmental parameters including temperature and precipitation; based on the long-term weather forecast of the year, estimating the weather trend in the next 30 days, selecting the time period with the best soil moisture and temperature conditions for sowing, to avoid uneven emergence and poor growth due to too early or too late sowing;

[0105] Step S502, determining the planting density: based on the sowing period, combined with the seed germination rate and the emergence rate, the sowing density is obtained, and the actual emergence of seedlings is monitored by remote sensing at the seedling stage, and the planting density is adjusted in time to avoid poor ventilation and light transmission due to overcrowding, mutual influx of pests and diseases, or yield loss due to oversparse planting;

[0106] Step S503, fertilizer and water management: based on the dynamic monitoring of soil nutrient content, analyze the demand law of the target farmland for nutrients such as nitrogen, phosphorus and potassium during the growth period, formulate a scientific fertilization plan, and apply appropriate amount of topdressing during the peak period of crop nutrient growth, such as the jointing period, to promote dry matter accumulation; based on the dynamic monitoring of soil nutrient content and soil moisture monitoring, obtain the dynamic change of soil moisture, and irrigate in time according to the degree of drought and the characteristics of crop water demand, so as to avoid blind flooding and prevent drought from forcing moisture; based on the main control factors identified in the crop growth diagnosis report, determine the type and amount of fertilizer to be applied, and when the crop growth diagnosis report shows the existence of growth stress, adjust the fertilizer and water management measures in time;

[0107] Step S504, pest and disease control: Based on the crop growth diagnosis report and the pest occurrence prediction model, the occurrence and development of pests and diseases are monitored in real time, the best time for pest and disease control is predicted, and combined with the growth status assessment in the crop growth diagnosis report, special attention is paid to areas with significant deviations during the key growth period to strengthen pest and disease control; by comparing the application effects and costs of various agents, high-efficiency, low-toxic and economical pesticide products are selected, and at the same time, in conjunction with the agricultural management APP, a pesticide application plan is automatically generated, including pesticide type, number of uses, interval time, and concentration ratio, for farmers to operate with one click on the mobile phone.

[0108] Preferably, the optimization of the selection and ratio of agricultural inputs in step S5 specifically includes:

[0109] Through big data analysis of the characteristics of different crop varieties, including disease and pest resistance, stress resistance, and yield levels, the excellent varieties suitable for planting in the target farmland are evaluated; for newly introduced varieties, trial planting evaluation plans are provided to demonstrate the feasibility of large-scale promotion of the newly introduced varieties, and virtual planting experiments are carried out through crop growth models to provide theoretical references for variety layout, crop rotation, etc.;

[0110] Establish a soil testing and fertilization formula database to recommend new fertilizer products based on soil nutrient status, including slow-release fertilizers and controlled-release fertilizers, to reduce non-point source pollution while meeting crop growth needs; compare and analyze the yield-increasing effects of different fertilizers through big data, rank them by cost-effectiveness, and help farmers save costs and increase efficiency;

[0111] Collect safety assessment data of major local pesticide products in the target farmland, including toxicity, residual period, and impact on beneficial insects, establish a comprehensive evaluation index system for pesticides, conduct comprehensive score evaluation and ranking of pesticide products, and consider factors such as high efficiency, low toxicity, and low residue; when formulating the pesticide application plan, give priority to recommending pesticide products with comprehensive scores higher than the preset score threshold in the comprehensive evaluation index system of pesticides, and guide farmers to use pesticides scientifically and safely; for newly launched pesticide products, compare their similarities with existing pesticides through big data, and preliminarily evaluate their use effects;

[0112] Optimization of agricultural machinery and implements: Through big data analysis of the operating efficiency and input-output ratio of different types of agricultural machinery and different planting scales, a reference is provided for the selection and configuration of agricultural machinery and implements; for newly developed intelligent agricultural machinery and equipment such as agricultural drones, their operating paths under different plot conditions are simulated, the control strategy is optimized, and the operation accuracy is improved.

[0113] Preferably, the evaluation of the economic benefits of different agricultural management decision-making schemes in step S5 includes measuring and analyzing the input-output benefits of various new technologies, new varieties, and new pesticides under different business scales, providing a scientific decision-making basis for farmers and management departments, specifically including:

[0114] Single technical economic evaluation: Calculate the yield increase rate and cost saving rate of specific new technologies, calculate the input-output ratio, and judge its economic rationality threshold. For example, calculate the cost of saving water and fertilizer per mu after adopting water-fertilizer integration, balance it with the benefits brought by increased production, and obtain the minimum operating scale suitable for promotion; the specific new technologies include water-fertilizer integration;

[0115] Comprehensive technical program evaluation: Design multiple sets of comprehensive agricultural technology programs for the planting system in a specific area, predict the yield, cost, and profit level under different programs, conduct comparative screening, and obtain the most cost-effective recommended program. The comprehensive agricultural technology program includes elements such as excellent varieties, reasonable density planting, simplified cultivation, and mechanized management;

[0116] Productivity layout assessment: Based on land use data and the crop growth model, the planting ratio and rotation method of different crops in a certain area are optimized to maximize the total output and economic benefits; providing macro-decision-making reference for the government to formulate agricultural development plans and adjust planting structures.

[0117] It should be noted that the data obtained in step S5 include agricultural decision-making plans, agricultural inputs optimization allocation plans, economic benefit evaluation reports, etc.

[0118] It should be noted that this embodiment uses multi-source sensors to collaboratively collect farmland environment, crop growth, and pest information, and combines edge computing to achieve efficient data preprocessing and transmission. An innovative crop growth parameter extraction algorithm based on the HSV color space is proposed, and a chlorophyll content prediction model is established: SPAD = m1H + m2S + m3V + C, where m1, m2, and m3 are weight coefficients for hue, saturation, and brightness, respectively; automatic identification and counting of pests are achieved through deep learning algorithms, and a pest occurrence prediction model is established in combination with environmental factors, which can not only accurately assess the health of farmland, but also provide targeted agricultural management decision-making recommendations based on multidimensional data analysis.

[0119] In this embodiment, data is collected on the target farmland through multiple sensors, sensor data corresponding to the target farmland is obtained and multi-level data preprocessing is performed to obtain structured sensor data of the target farmland, data analysis is performed on crop growth images to obtain growth parameters of crops, pest identification, positioning and counting are performed on insect information images through deep learning algorithms to obtain pest population dynamics, and a pest occurrence prediction model is constructed to obtain the probability of pest occurrence, generate early warning information on the timing of pest control, and finally obtain a farming management decision plan for the target farmland. The present invention predicts the probability of occurrence of pests and diseases in target farmland and proposes corresponding prevention and control measures by comprehensively considering environmental factors, crop growth conditions, and pest information, providing reference for farmers and managers, and can provide solutions for application scenarios such as experimental observation, growth monitoring, and remote observation by ordinary users.

[0120] Example 2

[0121] As attached Figure 2 As shown, Embodiment 2 of the present invention provides a farmland pest monitoring and control system S200 based on deep learning, which includes:

[0122] The data collection layer S201 is used to collect data on the target farmland through various sensors to obtain sensor data corresponding to the target farmland;

[0123] Specifically, it is deployed in the farmland monitoring area and consists of wireless sensor networks, cameras and other sensing devices. According to different monitoring indicators, temperature and humidity sensors, light sensors, soil conductivity sensors, soil pH sensors, etc. are set to collect farmland microclimate environmental parameters and soil physical and chemical properties parameters respectively. At the same time, high-definition cameras are configured to capture crop growth images and insect information images.

[0124] The edge computing layer S202 is used to perform multi-level data preprocessing based on the sensor data to obtain structured sensor data of the target farmland, wherein the structured sensor data includes environmental parameters, soil parameters, crop growth images, insect information images, pest population density, and farming operation data;

[0125] Specifically, the edge computing layer S202 has built-in data preprocessing module, feature extraction module, semantic annotation module and data compression module to perform multi-level preprocessing on the locally collected environment and image data, improve data quality, and optimize data transmission.

[0126] The data center layer S203 is used to perform data analysis on the crop growth image to obtain growth parameters of the crops, and obtain a crop growth diagnosis report of the target farmland based on the growth parameters of the crops;

[0127] Specifically, the data center layer S203 aggregates the multi-source data transmitted by the edge computing layer S202, performs data analysis based on the structured sensor data obtained by preprocessing the sensor data, and obtains a crop growth diagnosis report for the target farmland.

[0128] The intelligent analysis layer S204 is used to identify, locate and count pests on the insect information image through a deep learning algorithm to obtain pest population dynamics, build a pest occurrence prediction model based on multi-source data, obtain the probability of pest occurrence through the pest occurrence prediction model, and generate early warning information including the timing of pest control, wherein the pest population dynamics includes pest population density, and the multi-source data includes the environmental parameters, the growth parameters of the crops, the crop growth status assessment results in the crop growth diagnosis report, and the pest population dynamics;

[0129] Specifically, the intelligent analysis layer S204 takes the agricultural and insect situation correlation model as the core, integrates multi-dimensional data such as environmental factors, soil properties, crop growth, and insect pest information, and constructs functional modules such as crop yield prediction, nutrient demand diagnosis, and pest and disease identification and early warning.

[0130] The decision support layer S205 is used to obtain the agricultural management decision plan for the target farmland based on the sensor data, the growth parameters of the crops, the crop growth diagnosis report, the pest population dynamics, the probability of pest occurrence, and the timing of pest control, and optimize the selection and ratio of agricultural inputs, which include seeds, fertilizers, and pesticides, evaluate the economic benefits of different agricultural management decision plans, and provide reference for farmers and managers.

[0131] For example, based on the results of intelligent analysis and the experience of agricultural experts, decision-making plans for agricultural management such as fertilization, irrigation, and pesticide spraying are output, mainly considering factors such as current crop growth, severity of insect infestation, weather conditions, etc., matching corresponding plans in the agricultural knowledge base, and forming visual or voice-interactive auxiliary management suggestions to help farmers carry out precise agricultural operations.

[0132] Example 3

[0133] As attached Figure 3 As shown, Embodiment 3 of the present invention provides a farmland pest monitoring and control device S300 based on deep learning, and the device includes:

[0134] Memory S301, used for storing programs;

[0135] Processor S302 is used to load the program to execute the above-mentioned intelligent farmland system processing method.

[0136] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any modification, use or adaptation of the present invention, which follows the general principles of the present invention and includes common knowledge or customary techniques in the art that are not disclosed by the present invention. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present invention are indicated by the claims. It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A method for monitoring and controlling farmland pests and diseases based on deep learning, characterized in that: include: Perform multi-level data preprocessing based on sensor data of the target farmland to obtain structured sensor data of the target farmland, wherein the structured sensor data includes environmental parameters, soil parameters, crop growth images, insect information images, pest population density, and farming operation data; Performing data analysis on the crop growth images to obtain crop growth parameters, the growth parameters including crop canopy height, leaf area index and chlorophyll content, plotting the growth parameters at different growth periods into time series curves, and introducing a standard growth curve as a reference, comparing the difference between the actual growth process and the ideal model, and obtaining a crop growth diagnosis report for the target farmland; The acquisition of the chlorophyll content includes: extracting the chlorophyll content, that is, the relative content of chlorophyll in the leaves; converting the crop growth image from the RGB space to the HSV space, taking the yellow-green pixel points in the range of 45° to 65° in the hue channel as sampling objects, and estimating the chlorophyll content in combination with the following formula: , where H, S, V are the average values ​​of hue, saturation and brightness of the yellow-green pixels selected from the image, and C is the offset. They represent the contribution or influence of hue, saturation, and brightness on the prediction of chlorophyll content, respectively, reflecting the importance and sensitivity of each HSV component in estimating chlorophyll content; Pests are identified, located and counted on the insect information image through a deep learning algorithm to obtain pest population dynamics, and a pest occurrence prediction model is constructed based on multi-source data. The probability of pest occurrence is obtained through the pest occurrence prediction model, and early warning information including the timing of pest control is generated; Based on the sensor data, the growth parameters of the crops, the crop growth diagnosis report, the pest population dynamics, the probability of pest occurrence, and the timing of pest control, the agricultural management decision plan for the target farmland is obtained, and the selection and ratio of agricultural inputs are optimized. The agricultural inputs include seeds, fertilizers, and pesticides. The economic benefits of different agricultural management decision plans are evaluated to provide reference for farmers and managers.

2. A method for monitoring and controlling farmland pests and diseases based on deep learning according to claim 1, characterized in that: The acquisition of the leaf area index includes: extracting the ratio of the total leaf area per unit ground area to the ground area as the leaf area index: taking S blocks of pixels in the central area of ​​the crop growth image as samples, extracting the proportion r of green pixels in the S blocks of pixels, and calculating the leaf area index by the following formula: LAI=k*r / (1-r), wherein the coefficient k is adjusted according to the crop type and growth stage.

3. The method for monitoring and controlling farmland pests and diseases based on deep learning according to claim 1, characterized in that: Based on the dynamic correction mechanism of temperature T and light L, the influence of environmental conditions on the chlorophyll content is reduced. The dynamic correction mechanism is: , where the environmental correction function is defined as: ; ,in is the optimum temperature; For the optimal light intensity, is the fitting coefficient.

4. A method for monitoring and controlling farmland pests and diseases based on deep learning according to claim 3, characterized in that: Based on the daily and seasonal variation characteristics of the chlorophyll content, the chlorophyll content is corrected by a time series dynamic correction factor, and the formula is as follows: ,in is the sampling time in hours, is the daily variation coefficient, is the annual variation coefficient.

5. The method for monitoring and controlling farmland pests and diseases based on deep learning according to claim 1, characterized in that: The multi-level data preprocessing includes a first-level preprocessing and a second-level preprocessing, wherein the first-level preprocessing includes: The sensor data is subjected to a first layer of preprocessing through the edge computing layer to obtain first preprocessed data, including: for the environmental parameters, removing random noise interference through threshold filtering and median smoothing algorithm; for the soil parameters, removing obvious outliers, detecting the integrity of the data, and performing aggregation compression in units of fixed time intervals; for the crop growth image, extracting the crop area and removing the image background by using a target detection algorithm based on deep learning; for the insect information image, highlighting the texture features of the pest target through image enhancement and feature engineering methods; The second layer preprocessing includes: reporting the first preprocessed data to the data center layer through the message queue telemetry transmission protocol, the access module parses the data packet and verifies the signature to ensure that the data source is credible before performing the second layer preprocessing. The second layer preprocessing includes: unifying the time format and standardizing the field naming of the first preprocessed data to form the structured sensor data as the second preprocessed data.

6. The method for monitoring and controlling farmland pests and diseases based on deep learning according to claim 1, characterized in that: The method of constructing the pest occurrence prediction model comprises: Based on the environmental parameters, the growth parameters of the crops, the pest population dynamics and the farming operation data, through correlation analysis and factor analysis methods, key characteristic factors closely related to the occurrence of pests are screened out, including meteorological factors, crop growth indicators and farming factors, and a multi-dimensional input feature vector is constructed; Based on the multi-dimensional input feature vector, corresponding historical observation value sequence data is collected to form a complete pest prediction training sample set as input data of the pest occurrence prediction model, and the time span of the training sample set covers the entire process of pest occurrence; A nonlinear regression model of insect quantity and the key characteristic factors is constructed using a time series-based machine learning algorithm as the insect pest occurrence prediction model, wherein the input variables are the historical observation value sequences of various characteristic factors, and the output variables include the insect quantity or insect population density of the target farmland in the next few days; A fixed time period is set. Before the start of each time period, the historical observation value sequence data of the previous time period is used as the training sample set, and the trained pest occurrence prediction model is input to obtain the trend of insect population density changes within the time period, monitor the number of insects before the start of each time period, and predict the probability of pest occurrence and the pest level based on the trend of insect population density changes. The fixed time period includes daily, weekly, monthly, half-monthly, half-yearly, quarterly and annually.

7. A method for monitoring and controlling farmland pests and diseases based on deep learning according to claim 6, characterized in that: Based on the time decay factor, the probability of pest occurrence is calculated as follows: ,in express The probability of pest occurrence at any time, For the The key characteristic factors are The measured values ​​at the moment, including temperature, humidity, and light environment factors, is the weight coefficient of each key characteristic factor, reflecting the contribution of different key characteristic factors to the occurrence of insect pests. is the time attenuation coefficient, which is used to describe the impact of historical data on the current forecast. is the time point of data collection, is the time decay term, which changes with the time interval between the acquisition time point and the current time point t decreases with the increase of .

8. A farmland pest monitoring and control system based on deep learning, characterized in that: include: The data collection layer is used to collect data on the target farmland through a variety of sensors to obtain sensor data corresponding to the target farmland; An edge computing layer, used for performing multi-level data preprocessing based on the sensor data to obtain structured sensor data of the target farmland, wherein the structured sensor data includes environmental parameters, soil parameters, crop growth images, insect information images, pest population density, and farming operation data; A data center layer, used to perform data analysis on the crop growth images to obtain growth parameters of the crops, the growth parameters including crop canopy height, leaf area index and chlorophyll content, plot the growth parameters at different growth periods into time series curves, and introduce a standard growth curve as a reference to compare the difference between the actual growth process and the ideal model, and obtain a crop growth diagnosis report for the target farmland; An intelligent analysis layer, used to identify, locate and count pests on the insect information image through a deep learning algorithm to obtain pest population dynamics, build a pest occurrence prediction model based on multi-source data, obtain the probability of pest occurrence through the pest occurrence prediction model, and generate early warning information including the timing of pest control, wherein the pest population dynamics includes pest population density, and the multi-source data includes the environmental parameters, crop growth parameters, crop growth status assessment results in the crop growth diagnosis report, and pest population dynamics; The decision support layer is used to obtain the agricultural management decision plan of the target farmland based on the sensor data, the growth parameters of the crops, the crop growth diagnosis report, the pest population dynamics, the probability of pest occurrence, and the timing of pest control.

9. A deep learning-based farmland pest monitoring and control device, characterized in that: include: Memory, used to store programs; A processor, used to load a program to execute the farmland pest and disease monitoring and assessment method based on deep learning as described in any one of claims 1-7.

Images