A method for estimating the control period of tea green leafhopper based on automatic capture

Abstract

The present invention provides a method for estimating the tea green leafhopper control period based on automatic capture, which relates to the field of tea pest control, comprising: step S1, active capture; step S2, identification and counting: using a recognition model to identify and count tea green leafhoppers in the image obtained in step S1, to obtain the number and average length of tea green leafhoppers; step S3, environmental data collection; step S4, pest control period prediction: using a prediction model completed through offline pre-training to predict the current tea green leafhopper control period, and correcting the prevention period result using a prediction time correction method. The method uses an automated system to accurately predict the pest outbreak period based on factors such as the number of tea green leafhoppers, insect length, temperature, and humidity, with high prediction accuracy and low error, effectively reducing dependence on professional technicians and avoiding the errors and lags caused by manual prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of tea garden pest control, and in particular to a method for estimating the tea green leafhopper control period based on automatic capture. Background Art

[0002] The tea green leafhopper is a small insect, generally 3 to 4 mm long and primarily pale green to yellow-green in color. With its strong reproductive and adaptability, the leafhopper is one of the most common pests in tea gardens. Currently, physical, biological, and chemical control methods are used to control the leafhopper. Physical control involves installing color plates, insect traps, and other devices in tea gardens to trap the leafhopper. Biological control involves releasing natural enemies of the leafhopper (such as predatory mites and parasitic wasps) or bio-based agents into tea gardens to control the leafhopper's population. Chemical control involves directly killing the leafhopper with chemical pesticides. Among them, physical control methods and biological control methods are most effective when the population density of tea green leafhoppers is low. When the population density of tea green leafhoppers reaches a certain level, physical control and biological control methods will not be able to effectively control the number of tea green leafhoppers and cause tea losses (that is, a large number of physical control equipment and biological control natural enemies are needed to match tea green leafhoppers to achieve the purpose of control, which not only greatly increases the cost of control, but also easily destroys the ecological balance of tea gardens); therefore, when the population density of tea green leafhoppers is high, chemical control methods need to be used to prevent and control tea green leafhoppers more efficiently and at a lower cost. However, the current pest monitoring in tea gardens is mainly through manual observation of the conditions in the tea gardens and judging the outbreak period of tea green leafhoppers (i.e., insect population density) based on past experience. This method, firstly, wastes labor productivity, increases prevention and control costs, and is time-consuming and inefficient. Secondly, the tea green leafhoppers are small in size and difficult to observe and capture manually (whether it is manual observation directly in the tea garden or observation through cameras placed in the tea garden, the tea green leafhoppers need to be captured by the naked eye). It is very easy to make mistakes in judgment or inaccurate capture, which affects subsequent prevention and control work. Thirdly, it has high requirements for professional and technical personnel and is greatly affected by human subjective factors, resulting in high judgment errors and low accuracy. Fourthly, manual judgment has a certain lag and is easy to miss the best time node for tea green leafhopper prevention and control. Summary of the Invention

[0003] In response to the problems existing in the above-mentioned prior art, the purpose of the present invention is to provide a method for estimating the control period of tea green leafhoppers based on automatic capture. This method can effectively collect the status and number of tea green leafhoppers, and accurately predict the outbreak period of tea green leafhoppers (i.e., insect population density) through an automated system, thereby reducing dependence on professional and technical personnel, reducing errors in manual judgment and prediction, improving prediction efficiency, and reducing the cost of pest prediction.

[0004] The purpose of the present invention is achieved through the following technical solutions:

[0005] A method for estimating the tea green leafhopper control period based on automatic capture, comprising:

[0006] Step S1, active capture: using an active disturbance capturer to capture the tea green leafhopper in the target tea tree planting area, and using a high-definition camera to capture the image;

[0007] Step S2, identification and counting: using the recognition model to identify and count tea green leafhoppers in the image obtained in step S1, to obtain the number and average body length of tea green leafhoppers;

[0008] Step S3, environmental collection: collecting environmental parameters through environmental collectors distributed in the target tea tree planting area;

[0009] Step S4, pest control period prediction: The pest control period of the tea green leafhopper in the current state is predicted by the prediction model completed by offline pre-training, and the result of the control period is corrected by the prediction time correction method to obtain an accurate prediction model.

[0010] Based on the further optimization of the above scheme, the active disturbance catcher is evenly distributed in the target tea tree planting area, including a disturbance rod and a yellow sticky board (the size of the yellow sticky board is fixed) set in the positive direction of the disturbance rod. The corresponding tea trees in the target tea tree planting area are disturbed by the disturbance rod, so that the tea green leafhoppers attached to the tea trees fly out and are attracted and captured by the yellow sticky board set in the positive direction; the high-definition camera is set corresponding to the yellow sticky board (that is, the focal length, focus, etc. of the high-definition camera are fixed, and the distance between the high-definition camera and the yellow sticky board is fixed), which can obtain a full-area, high-definition image of the yellow sticky board.

[0011] Based on the further optimization of the above scheme, the recognition model adopts the deep neural network model completed by offline pre-training. The recognition model detects and recognizes the image obtained by the high-definition camera (that is, the high-definition image is input) and outputs the detection frame. The detection frame includes four vertex coordinates (x 1 ,y 1 ),(x 2 ,y 2 ),(x 3 ,y 3),(x 4 ,y 4 ) , step S2 is specifically as follows:

[0012] First, get the center coordinates of each detection box (x 0 ,y 0 ) :

[0013]

[0014] Then, all the center coordinates in the same image are counted to obtain the number of tea green leafhoppers corresponding to the yellow sticky board. n i , and obtain the mean of all yellow sticky boards as the measured value of the number of tea green leafhoppers in the target tea tree planting area:

[0015]

[0016] Where: N Indicates that the target tea tree planting area is set N A yellow sticky board;

[0017] Afterwards, the distance between the four sides of the detection frame is obtained by the distance formula d i (i=1,2,3,4):

[0018]

[0019] And take the maximum value of the four distances as the length of the corresponding detection frame D ;

[0020] Finally, calculate the average body length of all tea green leafhoppers on a yellow sticky board :

[0021]

[0022] Where: D i,j Indicates the i Yellow sticky board j The length of the tea green leafhopper detected.

[0023] Based on further optimization of the above solution, the environmental collector includes a temperature sensor, a humidity sensor and a timer, and the environmental parameters include the temperature, humidity and time of the target tea tree planting area at the time of collection.

[0024] Based on the further optimization of the above scheme, the main network of the prediction model adopts the MobileNetV4 lightweight model. The fully connected layer network is connected at the end of the MobileNetV4 network structure. The output dimension is 4x1, corresponding to the predicted control time, predicted pest scale, next detection time and next pest predicted scale respectively; the input of the prediction model is the state information vector v input , where the state information vector v input Composed of temperature, humidity, and pest scale:

[0025]

[0026] Where: t 0 means the average temperature of the day at the time of prediction; t his Indicates the number of days before the prediction time i The average temperature of the day is usually taken from the previous 30 days, that is, ; h 0 represents the average humidity at the time of prediction; h his Indicates the number of days before the prediction time i The average humidity of the day is generally taken from the first 30 days, that is, .

[0027] Based on the further optimization of the above scheme, the prediction time correction method includes pest scale prediction error calculation, data weight calculation and prediction model fine-tuning, specifically:

[0028] Step S41, calculation of pest scale prediction error: Based on the results of the pest control period prediction, i.e., the predicted control time, predicted pest scale, next detection time, and the next predicted pest scale, the pests are detected at the next detection time point obtained by active capture and identification counting; the pest scale prediction error is obtained by comparing the detection results with the prediction results. err , and judge whether the prediction error of tea green leafhopper pest scale is greater than the error threshold value. If it is greater than, then trigger step S42 to fine-tune the prediction model. If it is not greater than, then directly proceed to step S44; record the prediction error in real time err The number of times the error is less than the error threshold;

[0029] Step S42, data weight calculation: weight calculation is performed on all training data;

[0030] Step S43, fine-tuning the prediction model: using the training data weights obtained in step S42 to fine-tune the prediction model;

[0031] Step S44: Based on the current data status (i.e., the actual pest scale, environmental parameters, etc.), the prediction model is used to predict and output the updated predicted control time, predicted pest scale, next detection time, and next pest predicted scale, and step S41 is repeated until the latest predicted control time is reached or the continuous cumulative number of times exceeds the number threshold; wherein the continuous cumulative number is the number of times the error is calculated continuously. err The number of times the prediction error is less than the error threshold (once the prediction error is greater than the error threshold, the continuous accumulation number is reset to zero).

[0032] Based on further optimization of the above scheme, the pest scale prediction error err The specific calculation method is: according to the next detection time and the next pest prediction scale output by the prediction model n next Active capture and identification counting are used to obtain the actual pest scale at the next detection time point n real ,but:

[0033] .

[0034] Based on the further optimization of the above scheme, the data weight calculation is specifically as follows: calculating the weight scores of all samples in the training data and the current sample, wherein the weight scores are obtained by weighting the body length weight score, the pest scale weight score, the temperature weight score and the humidity weight score;

[0035] Body length weight score for:

[0036]

[0037] Where: Indicates the i Average body length of pests in samples; Indicates the average body length of pests in the current sample; represents the body length scaling factor;

[0038] Pest scale weight score for:

[0039]

[0040] Where: Indicates the i The number of pest infestations per sample; Indicates the pest size of the current sample; represents the pest scale scaling factor;

[0041] Temperature weight score for:

[0042]

[0043] Where: represents the temperature scaling factor; Indicates the i The temperature vector constructed by the samples; represents the temperature vector constructed by the current sample;

[0044]

[0045] Where: Indicates the i The first temperature value of the sample; Respectively represent i The maximum, minimum and mean temperature values ​​of the samples;

[0046] Moisture weight score for:

[0047]

[0048] Where: represents the humidity scaling factor; Indicates the i The humidity vector constructed by samples; Represents the humidity vector constructed by the current sample;

[0049]

[0050] Where: Indicates the i The first humidity value of the sample; Respectively represent i The maximum, minimum and mean humidity values ​​of the samples;

[0051] Weight score Si for:

[0052]

[0053] Where: Represent the weighted values ​​of body length, pest scale, temperature and humidity respectively.

[0054] Based on the further optimization of the above scheme, the prediction model fine-tuning is specifically as follows: according to the model fine-tuning method, a smaller learning rate is used to fine-tune the original parameters, wherein the loss function of the fine-tuning is:

[0055]

[0056] Where: batch Indicates the number of samples set for a gradient update; loss j 、 S j Respectively represent j The training loss value and weight score of each sample.

[0057] The following are the technical effects of the technical solution of the present invention:

[0058] In the prediction model, the present application continuously predicts the next detection time and the next pest prediction scale, thereby decomposing the prediction of pest outbreak and prevention into multiple steps, thereby effectively reducing the difficulty of the prediction task and improving the prediction accuracy; at the same time, in the prediction process of each step, the model is corrected according to its current state, thereby achieving the effect of self-update, realizing the process detection and update adjustment of the model, further improving the prediction accuracy, and avoiding the problem of prediction deviation from the actual trajectory due to error accumulation; through the addition of weight scores, the temperature and humidity information of the preceding moment of the prediction moment is used to assist in the evaluation of pest growth, and at the same time, the maximum temperature and humidity, minimum temperature and humidity, and average temperature and humidity of the prediction moment and the preceding moment of the prediction moment are used to characterize the temperature and humidity information of the preceding moment, thereby completing the measurement of the similarity of the temperature and humidity of the two samples, thereby avoiding the excessive impact of changes in ambient temperature and humidity on the prediction results, reducing the prediction error, and improving the prediction accuracy.

[0059] The present application uses an automated method to monitor and predict pests such as the tea green leafhopper during the tea tree planting process, effectively reducing dependence on professional and technical personnel, thereby reducing labor productivity, avoiding problems such as judgment failure, lag and inefficiency caused by human factors, and completing real-time and accurate prediction of pest outbreak periods, thereby providing accurate early warning for pest control work and preventing tea trees from being attacked by pests; in addition, the equipment used in the estimation method of the present application is simple and easy to obtain, and intelligent monitoring of pests can be achieved without large-scale transformation of the tea garden. It can be effectively promoted in tea gardens planted under various conditions and has a wide range of applications. DETAILED DESCRIPTION

[0060] The technical solutions in the embodiments of the present invention will be described clearly and completely below. In the following description, for the purpose of illustration rather than limitation, specific details such as specific system structures and technologies are proposed to facilitate a thorough understanding of the embodiments of the present invention.

[0061] Example 1:

[0062] A method for estimating the tea green leafhopper control period based on automatic capture, comprising:

[0063] Step S1, active capture: an active disturbance catcher is used to capture the tea green leafhopper in the target tea tree planting area, and an image is taken and saved by a high-definition camera; wherein the active disturbance catcher is evenly distributed in the target tea tree planting area (ensuring that the active disturbance catcher covers the entire target tea garden planting area, a dot matrix distribution can be adopted), including a disturbance rod and a yellow sticky board (the size of the yellow sticky board is fixed) set in the positive direction of the disturbance rod, and the corresponding tea tree in the target tea tree planting area is disturbed by the disturbance rod, so that the tea green leafhopper attached to the tea leaves flies out and is attracted and captured by the yellow sticky board set in the positive direction; the high-definition camera is set corresponding to the yellow sticky board (that is, the focal length and focus of the high-definition camera are fixed, and the distance between the high-definition camera and the yellow sticky board is fixed. For example: it can be set on the side opposite to the yellow sticky board and the line of sight of the high-definition camera is facing the yellow sticky board; the high-definition camera can use a common model in the agricultural field, and is not specifically limited in this embodiment), so as to obtain a full-area, high-definition image of the yellow sticky board.

[0064] Step S2, identification and counting: using the recognition model to identify and count tea green leafhoppers in the image obtained in step S1, to obtain the number and average body length of tea green leafhoppers;

[0065] The recognition model uses a deep neural network model that has been pre-trained offline (YOLOv3 is used in this embodiment, and other similar YOLO networks in the field, such as YOLOv5, YOLOv6, etc., can also be used). The recognition model detects and recognizes the image obtained by the HD camera (that is, the HD image is used as input) and outputs a detection frame. The detection frame includes the coordinates of four vertices. (x 1 ,y 1 ), (x 2 ,y 2 ),(x 3 ,y 3 ),(x 4 ,y 4 ) , step S2 is specifically as follows:

[0066] First, get the center coordinates of each detection box (x 0 ,y 0 ) :

[0067]

[0068] Then, all the center coordinates in the same image are counted to obtain the number of tea green leafhoppers corresponding to the yellow sticky board. n i , and get the mean of all yellow sticky plates As a measure of the number of tea green leafhoppers in the target tea planting area:

[0069]

[0070] Where: N Indicates setting up N yellow sticky boards in the target tea tree planting area;

[0071] Afterwards, the distance between the four sides of the detection frame is obtained by the distance formula d i (i=1,2,3,4):

[0072]

[0073] And take the maximum value of the four distances as the length of the corresponding detection frame D ;

[0074] Finally, calculate the average body length of all tea green leafhoppers on a yellow sticky board :

[0075]

[0076] Where: D i,j Indicates the i Yellow sticky board j The length of the tea green leafhopper detected.

[0077] Step S3, environmental collection: Environmental parameters are collected through environmental collectors distributed in the target tea tree planting area; the environmental collectors include a temperature sensor, a humidity sensor and a timer (wherein, the temperature sensor, humidity sensor and timer can all be common equipment models in the agricultural field. At the same time, the temperature sensor and humidity sensor are evenly distributed in the target tea tree planting area, and the temperature and humidity collected each time are the average value of multiple temperature sensors and humidity sensors). The environmental parameters include the temperature, humidity and time of the target tea tree planting area at the time of collection.

[0078] Step S4, pest control period prediction: The pest control period of the tea green leafhopper in the current state is predicted by the prediction model completed by offline pre-training, and the result of the control period is corrected by the prediction time correction method to obtain an accurate prediction model.

[0079] The main network of the prediction model adopts the MobileNetV4 lightweight model. The fully connected layer network is connected at the end of the MobileNetV4 network structure. The output dimension is 4x1, corresponding to the predicted control time, predicted pest scale, next detection time and next pest predicted scale respectively. The input of the prediction model is the state information vector v input , where the state information vector v inputComposed of temperature, humidity, and pest scale:

[0080]

[0081] Where: t 0 means the average temperature of the day at the time of prediction; t his Indicates the number of days before the prediction time i The average temperature of the day is usually taken from the previous 30 days, that is, ; h 0 represents the average humidity at the time of prediction; h his Indicates the number of days before the prediction time i The average humidity of the day is generally taken from the first 30 days, that is, .

[0082] The training of the prediction model is as follows: the loss function uses the MSE loss function, and the optimizer uses the AdamW optimizer. The process mainly includes: the training data is sent to the network in batches for forward operation - the loss is calculated based on the forward calculation results and labels - the reverse gradient calculation and parameter update are based on the loss, and the above process is iterated until the model training is completed. The training input of the prediction model includes (v input ,l) , l Indicates the label of the corresponding sample, which is the actual control time, pest scale, next detection time and next pest prediction scale. Among them, the next detection time is set to half the time node of the actual control time, and the next pest prediction scale is the actual pest scale corresponding to the next detection time.

[0083] The prediction time correction method includes the calculation of pest scale prediction error, data weight calculation and prediction model fine-tuning, specifically:

[0084] Step S41, calculation of pest scale prediction error: Based on the results of the pest control period prediction, i.e., the predicted control time, predicted pest scale, next detection time, and the next predicted pest scale, the pests are detected at the next detection time point obtained by active capture and identification counting; the pest scale prediction error is obtained by comparing the detection results with the prediction results. err , real-time recording of prediction errors err The number of times the error threshold is less than the error threshold; among them, the next detection time and the next pest prediction scale output by the prediction model n next , use active capture and identification counting (i.e. active capture method in step S1 and identification counting method in step S2) to obtain the actual pest scale at the next detection time point n real , thereby obtaining the pest scale prediction error err:

[0085]

[0086] And determine whether the prediction error of the scale of tea green leafhopper pest is greater than the error threshold (the error threshold is obtained based on a large amount of empirical data in the laboratory). If it is, step S42 is triggered to fine-tune the prediction model. If it is not, step S44 is directly performed.

[0087] Step S42, data weight calculation: performing weight calculation on all training data; the data weight calculation specifically includes calculating the weight scores of all samples in the training data and the current sample, wherein the weight scores are obtained by weighting the body length weight score, the pest scale weight score, the temperature weight score, and the humidity weight score;

[0088] Body length weight score for:

[0089]

[0090] Where: Indicates the i Average body length of pests in samples; Indicates the average body length of pests in the current sample; represents the body length scaling factor;

[0091] Pest scale weight score for:

[0092]

[0093] Where: Indicates the i The number of pest infestations per sample; Indicates the pest size of the current sample; represents the pest scale scaling factor;

[0094] Temperature weight score for:

[0095]

[0096] Where: represents the temperature scaling factor; Indicates the i The temperature vector constructed by the samples; represents the temperature vector constructed by the current sample;

[0097]

[0098] Where: Indicates the iThe first temperature value of the sample; Respectively represent i The maximum, minimum and mean temperature values ​​of the samples;

[0099] Moisture weight score for:

[0100]

[0101] Where: represents the humidity scaling factor; Indicates the i The humidity vector constructed by samples; Represents the humidity vector constructed by the current sample;

[0102]

[0103] Where: Indicates the i The first humidity value of the sample; Respectively represent i The maximum, minimum and mean humidity values ​​of the samples;

[0104] Among them, the body length scaling factor , pest scale scaling factor , temperature scaling factor , humidity scaling factor All of them are obtained through a large amount of laboratory experience data;

[0105] Weight score Si for:

[0106]

[0107] Where: They represent the weighted values ​​of body length, pest scale, temperature and humidity respectively, all of which are obtained from a large amount of empirical data in the laboratory.

[0108] Step S43, fine-tuning the prediction model: using the training data weights obtained in step S42, fine-tuning the prediction model; the prediction model fine-tuning is specifically: according to the model fine-tuning method, using a smaller learning rate, fine-tuning the original parameters, wherein the loss function of the fine-tuning is:

[0109]

[0110] Where: batch Indicates the number of samples set for a gradient update; loss j 、 S j Respectively represent jThe training loss value of each sample (obtained based on the actual training loss) and the weight score.

[0111] Step S44: Based on the current data status (i.e., the actual pest scale, environmental parameters, etc.), the prediction model is used to predict and output the updated predicted control time, predicted pest scale, next detection time, and next pest predicted scale, and step S41 is repeated until the latest predicted control time is reached or the continuous cumulative number of times exceeds the number threshold (the loop stops when either of the two conditions is met; the number threshold is obtained through a large amount of laboratory experience data); wherein the continuous cumulative number is the number of times the error is calculated repeatedly err The number of times the prediction error is less than the error threshold (once the prediction error is greater than the error threshold, the continuous accumulation number is reset to zero).

[0112] Example 2:

[0113] As another preferred embodiment, a tea garden pest prediction method, taking into account the different weights of different data, in order to obtain a fine-tuning model that is more consistent with the data scores, further adjusts the data composition used for fine-tuning on the basis of the solution in Example 1, that is, the data used for model fine-tuning is selected using the following method:

[0114] Step S431: Divide the training samples into a high-similarity sample set and a low-similarity sample set. Specifically, set the weight score threshold S th If the weight score corresponding to the training sample is greater than the weight score threshold, it is a low-similarity sample set; otherwise, it is a high-similarity sample set.

[0115] Step S202: Select all high-similarity sample sets and low-similarity sample sets that are no more than 20% of the total number of high-similarity sample sets to form a subset;

[0116] Step S203: randomly divide the subset into a training set and a test set in a ratio of 9:1, as the data set for fine-tuning the model.

Claims

1. A method for estimating the tea green leafhopper control period based on automatic capture, characterized by: include: Step S1, active capture: using an active disturbance capturer to capture the tea green leafhopper in the target tea tree planting area, and using a high-definition camera to capture the image; Step S2, identification and counting: using the recognition model to identify and count tea green leafhoppers in the image obtained in step S1, to obtain the number and average body length of tea green leafhoppers; Step S3, environmental collection: collecting environmental parameters through environmental collectors distributed in the target tea tree planting area; Step S4, pest control period prediction: predicting the pest control period of the tea green leafhopper in the current state through the offline pre-trained prediction model, and correcting the result of the control period through the prediction time correction method to obtain an accurate prediction model; The prediction time correction method includes the calculation of pest scale prediction error, data weight calculation and prediction model fine-tuning, specifically: Step S41, calculation of pest scale prediction error: Based on the results of the pest control period prediction, i.e., the predicted control time, predicted pest scale, next detection time, and the next predicted pest scale, the pests are detected at the next detection time point obtained by active capture and identification counting; the pest scale prediction error is obtained by comparing the detection results with the prediction results. err , and judge whether the prediction error of tea green leafhopper pest scale is greater than the error threshold value. If it is greater than, then trigger step S42 to fine-tune the prediction model. If it is not greater than, then directly proceed to step S44; record the prediction error in real time err The number of times the error is less than the error threshold; Pest scale prediction error err The specific calculation method is: according to the next detection time and the next pest prediction scale output by the prediction model n next Active capture and identification counting are used to obtain the actual pest scale at the next detection time point n real ,but: ； Step S42, data weight calculation: performing weight calculation on all training data; specifically, calculating the weight scores of all samples in the training data and the current sample, wherein the weight scores are obtained by weighting the body length weight score, the pest scale weight score, the temperature weight score, and the humidity weight score; Body length weight score for: Where: Indicates the i Average body length of pests in samples; Indicates the average body length of pests in the current sample; represents the body length scaling factor; Pest scale weight score for: Where: Indicates the i The number of pest infestations per sample; Indicates the pest size of the current sample; represents the pest scale scaling factor; Temperature weight score for: Where: represents the temperature scaling factor; Indicates the i The temperature vector constructed by the samples; represents the temperature vector constructed by the current sample; Where: Indicates the i The first temperature value of the sample; Respectively represent i The maximum, minimum and mean temperature values ​​of the samples; Moisture weight score for: Where: represents the humidity scaling factor; Indicates the i The humidity vector constructed by samples; Represents the humidity vector constructed by the current sample; Where: Indicates the i The first humidity value of the sample; Respectively represent i The maximum, minimum and mean humidity values ​​of the samples; Weight score Si for: Where: Represent the weighted values ​​of body length, pest scale, temperature and humidity respectively; Step S43, fine-tuning the prediction model: using the training data weights obtained in step S42, fine-tuning the prediction model, wherein the loss function of the fine-tuning is: Where: batch Indicates the number of samples set for a gradient update; loss j 、 S j Respectively represent j The training loss value and weight score of each sample; Step S44: Based on the current data status, the prediction model is used to predict and output the updated predicted control time, predicted pest scale, next detection time and next pest predicted scale, and step S41 is repeated until the latest predicted control time is reached or the continuous cumulative number of times exceeds the threshold value; wherein the continuous cumulative number is the number of times the error is calculated continuously. err The number of times the error is less than the error threshold.

2. The method for estimating the tea green leafhopper control period based on automatic capture according to claim 1, characterized in that: The active disturbance catchers are evenly distributed in the target tea tree planting area, including disturbance rods and yellow sticky boards set in the positive direction of the disturbance rods, and high-definition cameras are set corresponding to the yellow sticky boards.

3. A method for estimating the tea green leafhopper control period based on automatic capture according to claim 1 or 2, characterized in that: The recognition model uses a deep neural network model that has been pre-trained offline. The recognition model detects and recognizes the image obtained by the high-definition camera and outputs a detection frame. The detection frame includes four vertex coordinates (x 1 ,y 1 ),(x 2 ,y 2 ),(x 3 , y 3 ),(x 4 ,y 4 ) , step S2 is specifically as follows: First, get the center coordinates of each detection box (x 0 ,y 0 ) : Then, all the center coordinates in the same image are counted to obtain the number of tea green leafhoppers corresponding to the yellow sticky board. n i , and get the mean of all yellow sticky plates As a measure of the number of tea green leafhoppers in the target tea planting area: Where: N Indicates setting up N yellow sticky boards in the target tea tree planting area; Afterwards, the distance between the four sides of the detection frame is obtained by the distance formula d i (i=1,2,3,4): And take the maximum value of the four distances as the length of the corresponding detection frame D ; Finally, calculate the average body length of all tea green leafhoppers on a yellow sticky board : Where: D i,j Indicates the i Yellow sticky board j The length of the tea green leafhopper detected.

4. The method for estimating the tea green leafhopper control period based on automatic capture according to claim 3, characterized in that: The environmental collector includes a temperature sensor, a humidity sensor and a timer, and the environmental parameters include the temperature, humidity and time of the target tea tree planting area at the time of collection.

5. The method for estimating the tea green leafhopper control period based on automatic capture according to claim 3, characterized in that: The main network of the prediction model adopts the MobileNetV4 lightweight model. The fully connected layer network is connected at the end of the MobileNetV4 network structure. The output dimension is 4x1, which corresponds to the predicted control time, predicted pest scale, next detection time and next pest predicted scale respectively. The input of the prediction model is the state information vector v input , Among them, the state information vector v input Composed of temperature, humidity, and pest scale: Where: t 0 means the average temperature of the day at the time of prediction; t his Indicates the number of days before the prediction time i Average temperature of the day; h 0 represents the average humidity at the time of prediction; h his Indicates the number of days before the prediction time i Average humidity for the day.

Images