Tomato leaf miner identification and prediction system and method based on image identification

Abstract

The invention belongs to the technical field of agricultural pest monitoring, and particularly discloses a tomato leaf miner identification and prediction system and method based on image identification. The technical problem that the monitoring accuracy and efficiency of the tomato leaf miner are low due to the fact that the task requirement of real-time monitoring is not considered in the prior art and the capability of predicting the occurrence trend of future insect pests is not provided is solved. The parameters are adjusted according to the tomato parameter generation device, the monitoring result is generated according to the adjusted monitoring data, the prediction result is generated according to the monitoring result, and the related parameters of the acquisition device are adjusted according to the data such as the real-time growth state of the plant. The fusion weight between the multi-modal data is dynamically adjusted according to the adjusted data quality, task requirements and other parameters, so that the monitoring of the current result is more accurate and is close to the actual requirements; meanwhile, insect pests of plants in the future are predicted according to factors such as the environment, prevention can be effectively carried out in time, and the accuracy and efficiency of the tomato leaf miner identification and prediction method are improved.

Description

Technical Field

[0001] The present application belongs to the technical field of agricultural pest monitoring, and specifically is a tomato leafminer identification and prediction system and method based on image recognition. Background Art

[0002] The tomato leafminer is a highly destructive pest of tomato crops. Since spreading from its native South America, it has become a significant agricultural pest in many countries and regions around the world. This tiny moth insect particularly favors laying eggs on tomato plants. Once hatched, the larvae burrow into the leaves or fruit to feed, causing the characteristic "leafminer" damage: winding tunnels formed within the leaves, affecting the plant's photosynthesis and overall health. Furthermore, the larvae's attack on fruit directly leads to reduced yields and loss of commercial value. Due to its rapid reproduction rate and short life cycle, if not effectively controlled, large-scale outbreaks can occur within a short period of time, posing a serious threat to agricultural production.

[0003] Traditional monitoring methods rely on manual field inspections and empirical judgment, resulting in low efficiency, prone to omissions, and unable to meet the precise control needs of modern agriculture. In recent years, with the development of computer vision and artificial intelligence, automated pest and disease monitoring methods based on image recognition have gradually emerged. However, existing technologies often use fixed weights or rely on image quality for multimodal image fusion. These methods fail to consider the real-time monitoring requirements and lack the ability to predict future pest trends. This results in low accuracy and efficiency in tomato leafminer monitoring. Therefore, identification and prediction methods for tomato leafminer still need further improvement. Summary of the Invention

[0004] The present application aims to solve at least one of the technical problems existing in the prior art; to this end, the present application proposes a tomato leafminer identification and prediction system and method based on image recognition, which is used to solve the technical problems that the prior art lacks consideration of the task requirements of real-time monitoring and lacks the ability to predict future trends of insect pests, resulting in low accuracy and efficiency in monitoring tomato leafminers.

[0005] To achieve the above objectives, the first aspect of the present application provides a method for identifying and predicting tomato leafminers based on image recognition, comprising:

[0006] S1, obtain tomato parameters, environmental data and future weather forecast data;

[0007] S2. Input the tomato parameters and environmental data into the equipment parameter adjustment model to obtain a number of equipment adjustment parameters;

[0008] S3. adjusting the device based on the plurality of device adjustment parameters, and obtaining adjustment monitoring data based on the adjusted device;

[0009] S4. Generate monitoring results based on the adjusted monitoring data;

[0010] S5. Generate forecast results based on monitoring results, environmental data, and future weather forecast data;

[0011] S6. Generate an alarm signal based on the monitoring results and prediction results and perform corresponding early warning operations.

[0012] Preferably, tomato parameters include tomato root value, tomato height value, and tomato detection image; environmental data include temperature, humidity, and light intensity; equipment adjustment parameters refer to relevant parameters that need to be adjusted for relevant equipment when collecting data, including camera height and shooting angle. The equipment parameter adjustment model is constructed using an artificial intelligence model. The specific construction steps are as follows:

[0013] S21. Obtain several historical tomato parameters and historical environmental data and corresponding historical equipment adjustment parameters;

[0014] S22, dividing the historical tomato parameters and historical environment data and corresponding historical equipment adjustment parameters into training data, verification data, and test data;

[0015] S23, performing data preprocessing on the training data, the validation data, and the test data to obtain a training set, a validation set, and a test set, wherein the preprocessing includes data denoising and data enhancement;

[0016] S24. Selecting an artificial intelligence model as a basic model, wherein the selected artificial intelligence model is a neural network model;

[0017] S25. Use the training set to train the basic model, and adjust the learning rate on the validation set to obtain the pre-trained model;

[0018] S26. Use the test set to test the pre-trained model. If the test criteria are met, the device parameter adjustment model is obtained; otherwise, return to step S25 and retrain until the test criteria are met.

[0019] Preferably, the adjusted monitoring data in step S3 refers to tomato data obtained after assigning the device adjustment parameters to the corresponding device, including the total number of leaves, positions of several leaves, and two consecutive frames of visible light images and near-infrared images corresponding to the several leaves;

[0020] The monitoring result in step S4 refers to the current state of the plant, including the probability of insect eggs CLL, the probability of larvae YCL, the ratio of the buried channel area QDMB, and the ratio of chlorophyll loss YLSB. The specific steps of generating the monitoring result based on the adjusted monitoring data in step S4 include:

[0021] S41. Acquire and adjust monitoring data and monitoring labels, wherein the monitoring labels include current labels and predicted labels. The current labels refer to labels required to obtain the current state of the plant, and the predicted labels refer to labels required to obtain the estimated state of the plant in the future.

[0022] S42. Generate image weights corresponding to the visible light image and the near-infrared image based on the adjusted monitoring data;

[0023] S43, combining the image weight with the visible light image / near infrared image of the next frame to obtain analysis data;

[0024] S44. Input the current label and analysis data into the multi-task result generation model to obtain the monitoring result corresponding to the current label, wherein the multi-task result generation model is constructed through a machine learning model.

[0025] Preferably, the specific steps of step S42 are:

[0026] S421, continuously acquiring two frames of visible light images / near infrared images and the current label;

[0027] S422: Fusing the visible light image and the near-infrared image of the previous frame according to preset weights to obtain a fused image, and inputting the current label and the fused image into a multi-task result generation model to obtain a previous frame monitoring result corresponding to the current label;

[0028] S423, generating task sensitivity according to the previous frame monitoring result;

[0029] S424, extracting the visible image information entropy KJS and the near infrared information entropy HWS corresponding to the visible light image and the near infrared image of the next frame;

[0030] S425. Construct a visible weight function KQF (KJS, HWS, RMD) based on the task sensitivity RMD in step S423, the visible image information entropy KJS, the near-infrared information entropy HWS in step S424, and the nonlinear relationship between them and the visible light weight coefficient. The visible weight function KQF (KJS, HWS, RMD) is shown in the following formula:

[0031]

[0032] Where k is the task gain coefficient, h is the spectral difference attenuation rate, and both k and h are greater than 0;

[0033] S426. Substitute the visible image information entropy, the near-infrared information entropy, and the task sensitivity into the visible weight function to calculate the visible image weight, and perform a difference operation between the sum of the weight coefficients and the visible image weight to obtain the near-infrared image weight. The calculation formula for the near-infrared image weight HQX is as follows:

[0034] HQX = 1 - KQF

[0035] Preferably, the specific steps of generating the task sensitivity according to the previous frame monitoring result in step S423 include:

[0036] (1) Extracting the egg probability CLL and the larva probability YCL in the previous frame monitoring result;

[0037] (2) Constructing a sensitivity function MDF(CLL, YCL) according to the nonlinear relationship between the egg probability and the larva probability and the task sensitivity, wherein the sensitivity function MDF(CLL, YCL) is as follows:

[0038]

[0039] (3) Substituting the egg probability CLL and the larva probability YCL into the sensitivity function MDF(CLL, YCL) to calculate the task sensitivity RMD.

[0040] Preferably, the prediction result in step S5 refers to the estimated state of the plant in the future time, including the pest probability and the diffusion trend level; and the specific steps of generating the prediction result according to the monitoring result, the environmental data and the future weather forecast data in step S5 include:

[0041] S51, obtaining the current monitoring result, the environmental data and the future weather forecast data;

[0042] S52, obtaining a plurality of historical monitoring results, historical environmental data and prediction labels;

[0043] S53, integrating the current monitoring result, the environmental data and the future weather forecast data and the plurality of historical monitoring results and historical environmental data into a prediction sequence according to the time sequence;

[0044] S54, inputting the prediction labels and the prediction sequence into a multi-task result generation model to obtain the prediction result corresponding to the prediction labels, wherein the multi-task result generation model is constructed by a machine learning model.

[0045] Preferably, the multi-task result generation model is constructed by a machine learning model, and the specific construction steps include:

[0046] (1) Obtaining the current label and its corresponding plurality of historical analysis data and historical monitoring results, and the prediction label and its corresponding plurality of historical prediction sequences and historical prediction results;

[0047] (2) dividing the current label and its corresponding historical analysis data and historical monitoring results into training data, verification data, and test data corresponding to the current label, and performing data preprocessing on the training data, verification data, and test data corresponding to the current label to obtain a training set, a verification set, and a test set;

[0048] (3) Dividing the prediction label and its corresponding historical prediction sequences and historical prediction results into training data, verification data, and test data corresponding to the prediction label, and performing data preprocessing on the training data, verification data, and test data corresponding to the prediction label to obtain a training set, a verification set, and a test set;

[0049] (4) Select two machine learning models as the basic models for the current label and the predicted label respectively;

[0050] (5) Use the respective training sets to train their corresponding basic models, and adjust the learning rate and other hyperparameters on their respective validation sets to obtain their respective pre-trained models;

[0051] (6) Each test set is tested on its own pre-trained model. If the test criteria are met, a multi-task result generation model corresponding to each label is obtained; otherwise, return to step (5) and retrain until the test criteria are met.

[0052] Preferably, the alarm signal in step S6 includes a serious plant pest alarm signal, a minor plant pest alarm signal, a plant pest contamination risk signal, and a serious pest spread alarm signal; the specific steps of generating the alarm signal according to the monitoring results and the prediction results in step S6 include:

[0053] S61, obtaining monitoring results and prediction results corresponding to a number of blades;

[0054] S62: Extract the pest probability and spread trend level from the prediction results, and determine whether to generate a corresponding alarm signal. The specific determination process is as follows:

[0055] (1) Determine whether the probability of pests is greater than the probability threshold. If so, generate a plant pest contamination risk signal; if not, do nothing.

[0056] (2) determining whether the spreading trend level is greater than the level threshold, if so, generating a serious pest spreading alarm signal; if not, performing no operation, wherein the spreading trend level is within the range of 0-1;

[0057] S63. Generate a plant severity level based on the monitoring results, and determine whether to generate a corresponding alarm signal based on the plant severity level. The plant severity level refers to a quantitative measure of the plant pest situation. The specific determination process is as follows:

[0058] Determine whether the severity of the plant is greater than the severity threshold. If so, generate a serious plant pest alarm signal; if not, determine whether the severity of the plant is greater than D times the severity threshold. If so, generate a minor plant pest alarm signal, otherwise generate a plant health prompt signal.

[0059] Preferably, the specific steps of generating the severity of the plant according to the monitoring results in step S63 include:

[0060] S631, obtaining monitoring results corresponding to a number of blades, blade positions, and the total number of blades, wherein the blade positions include top blades, middle blades, and bottom blades;

[0061] S632. Extract the egg probability CLL, larva probability YCL, subsurface area ratio QDMB, and chlorophyll loss ratio YLSB from the monitoring results;

[0062] S633. Based on the nonlinear relationship between the egg probability CLL, the larval probability YCL, the duct area ratio QDMB, the chlorophyll loss ratio YLSB, and the leaf severity, a leaf severity function YCF(CLL, YCL, QDMB, YLSB) is constructed. The leaf severity function is shown in the following formula:

[0063] MDF(CLL,YCL,QDMB,YLSB)=max(CLL,YCL)×QDMB+μ×YLSB

[0064] Where μ represents the loss contribution coefficient, μ∈(0,1);

[0065] S634, substituting the egg probability, larva probability, duct area ratio, and chlorophyll loss ratio into the leaf severity function to calculate the leaf severity YYD of the plurality of leaves;

[0066] S635 , obtaining the number of severe leaves YS whose severity is greater than the severe threshold value, and calculating the severe ratio YB based on the ratio between the severe number and the total number of leaves;

[0067] S636. Construct a plant severity function ZCF(YYD, YW, YB) based on the nonlinear relationship between the leaf severity YYD, leaf position YW, severity ratio YB, and plant severity of the plurality of leaves. The plant severity function is shown in the following formula:

[0068]

[0069] Where i is the number of the blade, WQ i is the leaf position weight of the i-th leaf, is the severity of the plant of the i-th leaf in its corresponding YW; ω is the diffusion penalty coefficient, ω∈(0,1);

[0070] S637. Substitute the leaf severity, leaf position, and severity ratio corresponding to the plurality of leaves into the plant severity function to calculate the plant severity.

[0071] Another aspect of the present invention provides a tomato leafminer identification and prediction system based on image recognition, which is used to perform a tomato leafminer identification and prediction method based on image recognition, including: a data acquisition module, a data analysis module and an early warning module, wherein the data acquisition module is connected to the data analysis module, and the data analysis module is connected to the early warning module, wherein:

[0072] Data acquisition module: obtains tomato parameters, environmental data and future weather forecast data through data acquisition equipment;

[0073] Data analysis module: Inputs tomato parameters and environmental data into the equipment parameter adjustment model to obtain several equipment adjustment parameters; adjusts the equipment based on the several equipment adjustment parameters, and obtains adjustment monitoring data based on the adjusted equipment; generates monitoring results based on the adjusted monitoring data; generates prediction results based on the monitoring results, environmental data, and future weather forecast data; and generates alarm signals based on the monitoring results and prediction results;

[0074] Early warning module: performs corresponding early warning operations according to the alarm signal.

[0075] Compared with the prior art, the present invention has the following advantages:

[0076] 1. This application uses data such as the real-time growth status of the plant to adjust the relevant parameters of the collection equipment, making the collected data more accurate, and dynamically adjusts the fusion weights between multimodal data based on the adjusted data quality and task requirements, so that the monitoring of the current results is more accurate and closer to actual needs. Subsequently, the severity of the plant is adaptively calculated through parameters such as leaf severity and leaf position, making the monitoring of tomato plants more accurate; at the same time, the future plant pests are predicted based on environmental factors, which can be prevented in a timely and effective manner, thereby improving the accuracy and efficiency of the tomato leafminer identification and monitoring method.

[0077] 2. This application takes into account the growth conditions and environmental conditions of different plants when collecting data. The relevant parameters of the equipment will be different. By adjusting the model based on tomato parameters and environmental data as well as pre-trained equipment parameters, the parameter values ​​of the relevant equipment are dynamically adjusted to improve the accuracy of the collected data, thereby providing solid and accurate data support for the subsequent identification and prediction of tomato leafminers.

[0078] 3. This application continuously obtains two frames of adjusted monitoring data during monitoring, and dynamically adjusts the image weight corresponding to the adjusted monitoring data of the subsequent frame with the adjusted monitoring data of the previous frame, so that the subsequent frame can better reflect the importance of task requirements when performing image fusion, and the fused image has more effective features for monitoring and identification, providing accurate data support for the accurate judgment of current monitoring results and future prediction results.

[0079] 4. This application uses a pre-trained multi-task result generation model to dynamically predict future plant pest conditions based on current monitoring results and environmental data as well as future environmental data, allowing users to take corresponding preventive measures in advance, reduce the occurrence of product damage, and improve the comprehensiveness of the method. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0081] Figure 1 This is a flow chart of a method for identifying and predicting tomato leafminers based on image recognition in this application;

[0082] Figure 2 Generate a flow chart for the alarm signal of this application;

[0083] Figure 3 This is a framework diagram of a tomato leafminer identification and prediction system based on image recognition in this application. DETAILED DESCRIPTION

[0084] The following will clearly and completely describe the technical solutions of this application in conjunction with the embodiments. Obviously, the embodiments described are only a part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0085] See also Figure 1 The first embodiment of the present application provides a method for identifying and predicting tomato leafminers based on image recognition, comprising:

[0086] S1. Obtain tomato parameters, environmental data, and future weather forecast data.

[0087] In this application, tomato parameters include tomato root value, tomato height value and tomato detection image, etc.; environmental data include temperature, humidity, light intensity, etc.; weather forecast data includes temperature, humidity, rainfall, light intensity, etc.

[0088] It should be noted that the future weather forecast data specifically refers to the next number of days, which is set by the staff according to actual needs, such as the next 3 days, the next 7 days, etc.

[0089] S2. Input the tomato parameters and environmental data into the equipment parameter adjustment model to obtain several equipment adjustment parameters.

[0090] In this application, device adjustment parameters refer to the relevant parameters that need to be adjusted for the relevant equipment when collecting data, including camera height, shooting angle, etc.

[0091] In this application, the device parameter adjustment model is constructed using an artificial intelligence model. The specific construction steps are as follows:

[0092] S21. Obtain several historical tomato parameters and historical environment data as well as corresponding historical equipment adjustment parameters.

[0093] S22. Divide the historical tomato parameters, historical environment data, and corresponding historical equipment adjustment parameters into training data, verification data, and test data.

[0094] S23. Perform data preprocessing on the training data, verification data, and test data to obtain a training set, a verification set, and a test set, wherein the preprocessing includes data denoising, data enhancement, etc.

[0095] In this application, the ratio between the training set, test set and validation set is 7:2:1.

[0096] S24. Select an artificial intelligence model as a basic model, wherein the selected artificial intelligence model is a neural network model.

[0097] S25. Use the training set to train the basic model, and adjust the learning rate on the validation set to obtain the pre-trained model.

[0098] It should be noted that in addition to the learning rate, other hyperparameters can also be adjusted, and staff can set them according to actual needs when using it.

[0099] S26. Use the test set to test the pre-trained model. If the test criteria are met, the device parameter adjustment model is obtained; otherwise, return to step S25 and retrain until the test criteria are met.

[0100] The equipment parameter adjustment model obtained in this application has tomato parameters and environmental data as input parameters, and several equipment adjustment parameters as output parameters.

[0101] It should be noted that this application does not impose specific restrictions on test standards, and staff can set them according to actual needs when using it.

[0102] This embodiment fully takes into account the differences in growth conditions and environmental conditions among different tomato plants. When collecting data, the parameter settings of related equipment will also change accordingly. By combining tomato growth parameters, environmental data, and a pre-trained equipment parameter adjustment model, the system can dynamically optimize the parameter configuration of related equipment, thereby effectively improving the accuracy and reliability of the collected data, and providing solid and high-quality data support for the subsequent accurate identification and prediction of tomato leafminers.

[0103] S3. Adjust the device based on several device adjustment parameters, and obtain adjustment monitoring data based on the adjusted device.

[0104] In this application, the adjusted monitoring data refers to the tomato data obtained after assigning the device adjustment parameters to the corresponding device, including the total number of leaves, the positions of several leaves, and two consecutive frames of visible light images and near-infrared images corresponding to several leaves.

[0105] S4. Generate monitoring results based on the adjusted monitoring data, where the monitoring results refer to the current status of the plant, including the insect egg probability CLL, the larva probability YCL, the burrow area ratio QDMB, and the chlorophyll loss ratio YLSB.

[0106] In this application, the specific steps of generating monitoring results based on the adjusted monitoring data in step S4 include:

[0107] S41. Acquire and adjust monitoring data and monitoring labels, wherein the monitoring labels include current labels and predicted labels. The current labels refer to labels required to obtain the current state of the plant, and the predicted labels refer to labels required to obtain the estimated state of the plant in the future.

[0108] It should be noted that only the current label is used in step S4 of this application, and the predicted label is used in step S5. In addition, the monitoring label is not an empty label, but is only used to select the corresponding model branch to obtain the corresponding result. Since the multi-task result generation model used in this application is used to produce current results and predicted results, the current label and predicted label are set in this application so that the model automatically selects the corresponding model branch when used. For example, when obtaining the current label, the model will automatically select the multi-task result generation model used to generate the current plant state; when obtaining the predicted label, the model will automatically select the multi-task result generation model used to generate the estimated future plant state.

[0109] S42. Generate image weights corresponding to the visible light image and the near-infrared image based on the adjusted monitoring data, where the image weights refer to the degree of influence of the visible light image and the near-infrared image on the monitoring results when the monitoring results are obtained. The specific steps are as follows:

[0110] S421 , continuously acquiring two frames of visible light images / near infrared images and the current label.

[0111] In this embodiment, by acquiring two consecutive frames of visible light images / near-infrared images, the weights of each image in the next frame are adjusted according to the result of the image of the previous frame. This is because the state of the plant between two adjacent frames will not change significantly. Therefore, this method can be used to help generate image weights.

[0112] S422: Fusing the visible light image and the near-infrared image of the previous frame according to preset weights to obtain a fused image, and inputting the current label and the fused image into a multi-task result generation model to obtain the previous frame monitoring result corresponding to the current label.

[0113] In this application, the preset weights can be set based on experience. In the embodiment of this application, the preset weights of the visible light image and the near infrared image are set to 0.4 and 0.6 respectively.

[0114] In the embodiment of the present application, the preset weights of the visible light image and the near-infrared image can also be obtained by the following method, specifically:

[0115] (1) Extract the signal-to-noise ratio, information entropy and contrast corresponding to the visible light image / near-infrared image.

[0116] (2) The signal-to-noise ratio is normalized to obtain the normalized signal-to-noise ratio.

[0117] (3) The signal-to-noise ratio, information entropy, and contrast are calculated by weighted summation to obtain the visible light image quality and near-infrared image quality, as shown in the following formula:

[0118] KJZ=A1×a+B1×b+C1×c

[0119] JHZ=A2×a+B2×b+C2×c

[0120] Where KJZ is the visible light image quality, A1 is the signal-to-noise ratio of the visible light image, B1 is the information entropy of the visible light image, C1 is the contrast of the visible light image, JHZ is the near-infrared image quality, A2 is the signal-to-noise ratio of the near-infrared image, B2 is the information entropy of the near-infrared image, C2 is the contrast of the near-infrared image, a is the weight coefficient of the signal-to-noise ratio, b is the weight coefficient of the information entropy, and c is the weight coefficient of the contrast.

[0121] It should be noted that information entropy and contrast are numerical values ​​within the range of 0-1, so they are not normalized in this application.

[0122] The larger the weight, the greater the impact on image quality assessment. In this application, the weight coefficients between signal-to-noise ratio, information entropy, and contrast are set based on experience. In the embodiment of this application, the weight coefficients of signal-to-noise ratio, information entropy, and contrast are set to 0.3, 0.5, and 0.2, respectively.

[0123] (4) Based on the visible light image quality and the near infrared image quality, the preset weights of the visible light image and the near infrared image are calculated as shown in the following formula:

[0124]

[0125] JHTQ=1-KJTQ

[0126] Where KJTQ is the preset weight of visible light image, KJZ is the visible light image quality, JHTQ is the preset weight of near-infrared image, JHZ is the near-infrared image quality; g is the sharpness coefficient, g>0.

[0127] It should be noted that the specific value of the sharpness coefficient g is set based on experience. In the embodiment of the present application, g is set to 2.

[0128] S423: Generate task sensitivity based on the previous frame monitoring result.

[0129] In this application, task sensitivity refers to the degree of monitoring tasks when monitoring plant status. For example, monitoring insect eggs may require more near-infrared information, while observing leaf damage may require visible light details. Task sensitivity directly affects the size relationship between image weights.

[0130] In the embodiment of the present application, the specific steps of generating task sensitivity based on the previous frame monitoring result include:

[0131] (1) Extract the egg probability CLL and larva probability YCL from the previous frame monitoring results.

[0132] (2) Based on the nonlinear relationship between the egg probability, larva probability and task sensitivity, a sensitivity function MDF(CLL, YCL) is constructed, where the sensitivity function MDF(CLL, YCL) is shown as follows:

[0133]

[0134] The sensitivity function is a Sigmoid function that controls the task sensitivity within the range of 0-1. In this example, eggs are weighted more heavily than larvae because eggs are small and hidden, requiring higher sensitivity for monitoring. Larvae, on the other hand, are easily detected by near-infrared images, so their motion characteristics should be suppressed.

[0135] (3) Substitute the egg probability CLL and the larva probability YCL into the sensitivity function MDF(CLL, YCL) to calculate the task sensitivity RMD.

[0136] S424 , extracting the visible image information entropy KJS and the near-infrared information entropy HWS corresponding to the visible light image and the near-infrared image of the next frame.

[0137] S425. Construct a visible weight function KQF (KJS, HWS, RMD) based on the task sensitivity RMD in step S423, the visible image information entropy KJS, the near-infrared information entropy HWS in step S424, and the nonlinear relationship between them and the visible light weight coefficient. The visible weight function KQF (KJS, HWS, RMD) is shown in the following formula:

[0138]

[0139] Where k is the task gain coefficient, h is the spectral difference attenuation rate, and both k and h are greater than 0.

[0140] In this application, k is set to control the intensity of the task requirements on the weight curve, and h is set to adjust the attenuation rate of the influence of the information entropy difference between the infrared information entropy and the visible image information entropy.

[0141] In the embodiment of the present application, k is set to 2 and h is set to 0.5. If k is set to 1 and h is set to 0.8, robust multispectral image fusion is achieved.

[0142] It should be noted that if the difference between coefficients k and h is large, the difference in the final value of the weight will be large. If robustness is required, the difference between the two weight coefficients should not be too large.

[0143] S426. Substitute the visible image information entropy, the near-infrared information entropy, and the task sensitivity into the visible weight function to calculate the visible image weight, and perform a difference operation between the sum of the weight coefficients and the visible image weight to obtain the near-infrared image weight. The calculation formula for the near-infrared image weight HQX is as follows:

[0144] HQX=1-KQF

[0145] In this embodiment, the sum of the weight coefficients is set to 1.

[0146] In image fusion, different tasks may require different information emphases. Task requirements directly affect which image features are more critical. Therefore, using task sensitivity as an influencing parameter for dynamically adjusting image weights can improve monitoring accuracy.

[0147] This application provides direction for the allocation of task requirements for subsequent frames by analyzing the monitoring results of the previous frame, so that the subsequent frame can better reflect the monitoring task requirements when calculating the image weight, so that when generating the monitoring results, the corresponding image features can be reasonably used to improve the accuracy of the monitoring results.

[0148] S43 , combining the image weight with the visible light image / near infrared image of the next frame to obtain analysis data.

[0149] It should be noted that the analysis data is a fused image obtained by fusing two images according to weights.

[0150] S44. Input the current label and analysis data into the multi-task result generation model to obtain the monitoring result corresponding to the current label, wherein the multi-task result generation model is constructed through a machine learning model.

[0151] During the monitoring process, this application uses data from two consecutive frames for adjustment, and dynamically optimizes the corresponding image weight in the next frame through the monitoring results of the previous frame, ensuring that when performing image fusion, the next frame can more accurately reflect the importance of task requirements, so that the fused image contains more effective features that are helpful for monitoring and identification. This not only makes the current monitoring results more accurate, but also provides reliable data support for future prediction results, enhances the accuracy of judgment, effectively improves the performance of the entire monitoring system, and ensures accurate assessment of the status of tomato plants and potential pests.

[0152] S5. Generate prediction results based on monitoring results, environmental data, and future weather forecast data, where the prediction results refer to the estimated status of plants in the future, including the probability of pests and the level of spread trend.

[0153] In this application, the specific steps of generating the prediction results based on the monitoring results, environmental data and future weather forecast data in step S5 include:

[0154] S51, obtaining current monitoring results and environmental data as well as future weather forecast data;

[0155] S52, obtaining a number of historical monitoring results and historical environmental data and prediction labels;

[0156] S53, integrating the current monitoring results, environmental data, and future weather forecast data as well as a number of historical monitoring results and historical environmental data into a prediction sequence in chronological order;

[0157] S54, input the prediction label and the prediction sequence into a multi-task result generation model to obtain a prediction result corresponding to the prediction label, wherein the multi-task result generation model is constructed by a machine learning model.

[0158] By introducing the pre-trained multi-task result generation model, the embodiment can comprehensively integrate the current monitoring data and the current and future environmental information, dynamically predict the possible pest situation of the plant in the future, enable the user to take corresponding preventive measures in advance of the occurrence of the pest, effectively reduce the risk of crop damage, and improve the initiative and foresight of the disease and pest control, thereby enhancing the comprehensiveness and practicality of the whole monitoring and early warning method.

[0159] In the present application, the multi-task result generation model is constructed by a machine learning model, and the specific construction steps include:

[0160] (1) obtaining a current label and a plurality of historical analysis data and historical monitoring results corresponding thereto, and a prediction label and a plurality of historical prediction sequences and historical prediction results corresponding thereto.

[0161] (2) dividing the current label and the plurality of historical analysis data and historical monitoring results corresponding thereto into training data, verification data and test data corresponding to the current label, and performing data preprocessing on the training data, the verification data and the test data corresponding to the current label to obtain a training set, a verification set and a test set.

[0162] In the embodiment of the present application, the ratio between the training set, the test set and the verification set is 7:2:1.

[0163] (3) dividing the prediction label and the plurality of historical prediction sequences and historical prediction results corresponding thereto into training data, verification data and test data corresponding to the prediction label, and performing data preprocessing on the training data, the verification data and the test data corresponding to the prediction label to obtain a training set, a verification set and a test set.

[0164] In the embodiment of the present application, the ratio between the training set, the test set and the verification set is 7:2:1.

[0165] (4) selecting two machine learning models as the base models of the current label and the prediction label.

[0166] In the present application, a neural network model is selected as the base model of the current label, and an LSTM model is selected as the base model of the prediction label.

[0167] (5) training the respective base models with the respective training sets, and adjusting the learning rate and other hyperparameters on the respective verification sets to obtain respective pre-trained models.

[0168] (6) the respective test set is tested on the respective pre-trained model, if the test standard is met, the multi-task result generation model corresponding to the respective label is obtained; otherwise, return to step (5) to retrain until the test standard is met and stop.

[0169] In the present application, the multi-task result generation model corresponding to the current label inputs the current label and the analysis data corresponding thereto as parameters, and outputs the monitoring result corresponding to the current label as a parameter; the multi-task result generation model corresponding to the predicted label inputs the predicted label and the prediction sequence corresponding thereto as parameters, and outputs the prediction result corresponding to the predicted label as a parameter

[0170] It should be noted that the test standard is not specifically limited in the present application, and can be set by the staff according to actual needs.

[0171] S6, generating an alarm signal according to the monitoring result and the prediction result and performing a corresponding warning operation, the alarm signal including a plant pest infestation serious alarm signal, a plant slight pest infestation alarm signal, a plant pest contamination risk signal and a pest infestation serious alarm signal.

[0172] Please refer to Figure 2 In the present application, the specific steps of generating an alarm signal according to the monitoring result and the prediction result include:

[0173] S61, obtaining the monitoring result and the prediction result corresponding to a plurality of leaves.

[0174] S62, extracting the pest infestation probability and the infestation trend level in the prediction result, and determining whether to generate a corresponding alarm signal, the specific determination process being:

[0175] (1) determining whether the pest infestation probability is greater than a probability threshold, if yes, generating a plant pest contamination risk signal; otherwise, not doing any operation.

[0176] In the present application, the probability threshold is set according to experience. In the present application, the probability threshold is set to 0.5.

[0177] (2) determining whether the infestation trend level is greater than a level threshold, if yes, generating a pest infestation serious alarm signal; otherwise, not doing any operation, wherein the infestation trend level is in the range of 0-1.

[0178] In the present application, the level threshold is set according to experience. In the present application, the level threshold is set to 0.4.

[0179] S63, generating a plant severity according to the monitoring result, and determining whether to generate a corresponding alarm signal according to the plant severity, wherein the plant severity refers to the quantification of the plant pest infestation, and the specific determination process being:

[0180] Determine whether the severity of the plant is greater than the severity threshold. If so, generate a serious plant pest alarm signal; if not, determine whether the severity of the plant is greater than D times the severity threshold. If so, generate a minor plant pest alarm signal, otherwise generate a plant health prompt signal.

[0181] In this application, the degree threshold is set based on experience. In the embodiment of this application, the degree threshold is set to 0.65.

[0182] In this application, D is a proportional coefficient, D∈(0,1), and the specific value is set based on experience. In the embodiment of this application, D is set to 0.6.

[0183] This embodiment can generate multiple alarm signals in a timely manner by analyzing the current monitoring results and the current prediction results, so that users can take timely countermeasures and avoid product damage caused by insect pests and waste of resources.

[0184] In the embodiment of the present application, the specific steps of generating the severity of the plant according to the monitoring results include:

[0185] S631. Obtain monitoring results corresponding to a number of blades, blade positions, and the total number of blades, wherein the blade positions include top blades, middle blades, and bottom blades.

[0186] S632. Extract the egg probability CLL, larva probability YCL, subsurface area ratio QDMB, and chlorophyll loss ratio YLSB from the monitoring results.

[0187] S633. Based on the nonlinear relationship between the egg probability CLL, the larval probability YCL, the duct area ratio QDMB, the chlorophyll loss ratio YLSB, and the leaf severity, a leaf severity function YCF(CLL, YCL, QDMB, YLSB) is constructed. The leaf severity function is shown in the following formula:

[0188] MDF(CLL,YCL,QDMB,YLSB)=max(CLL,YCL)×QDMB+μ×YLSB

[0189] Where μ represents the loss contribution coefficient, μ∈(0,1).

[0190] It should be noted that the specific value of the loss contribution coefficient μ is set based on experience. In the embodiment of this application, μ is set to 0.3. In other embodiments, μ can also be set to 0.4 or 0.5. The setting of μ reflects the contribution weight of the leaf physiological damage to the leaf severity. The larger μ is, the greater the impact of the leaf physiological damage on the leaf severity.

[0191] S634. Substitute the egg probability, larva probability, duct area ratio, and chlorophyll loss ratio into the leaf severity function to calculate the leaf severity YYD of the plurality of leaves.

[0192] S635 , obtaining the number of severe leaves YS whose severity is greater than the severe threshold of the leaves, and calculating the severe ratio YB based on the ratio between the severe number and the total number of leaves.

[0193] The leaf severity threshold is set based on experience. In the embodiment of the present application, the leaf severity threshold is set to 0.6.

[0194] S636. Construct a plant severity function ZCF(YYD, YW, YB) based on the nonlinear relationship between the leaf severity YYD, leaf position YW, severity ratio YB, and plant severity of the plurality of leaves. The plant severity function is shown in the following formula:

[0195]

[0196] Where i is the number of the blade, WQ i is the leaf position weight of the i-th leaf, is the severity of the plant of the i-th leaf in its corresponding YW; ω is the diffusion penalty coefficient, ω∈(0,1).

[0197] It should be noted that the specific value of the diffusion penalty coefficient ω is set based on experience. In this embodiment, ω is set to 0.1, and in other embodiments, μ can also be set to 0.2. The setting of ω reflects the impact of the proportion of severe leaves on the severity of the entire plant. The larger ω is, the greater the impact of the proportion of severe leaves on the severity of the entire plant.

[0198] In addition, the leaf position weights are also set based on experience. In this embodiment, the leaf position weights corresponding to the top leaf, middle leaf, and bottom leaf are set to 2.0, 1.5, and 0.8, respectively. For example, if the i-th leaf is the top leaf and its corresponding leaf severity YYD is 0.6, then Expressed as 2.0×0.6.

[0199] S637. Substitute the leaf severity, leaf position, and severity ratio corresponding to the plurality of leaves into the plant severity function to calculate the plant severity.

[0200] This application quantifies the pest conditions of monitored leaves, and takes into account that different positions of leaves and the number of severely infested leaves will have different impacts on the overall pest conditions of the plant. It adaptively quantifies the pest conditions of the plant, making the overall condition of the plant more accurately expressed and providing accurate data support for the generation of alarm signals.

[0201] In summary, the application optimizes and adjusts the related parameters of the collection device according to the real-time growth state of the plant and other data, ensures that the collected data is more accurate, and dynamically adjusts the fusion weight between multi-modal data according to the optimized data quality and task demand and other parameters, so that the monitoring result is not only more accurate, but also more in line with the actual demand; then the overall damage degree of the plant is adaptively calculated by analyzing the severity and location of the leaf damage and other parameters, thereby improving the accuracy of tomato plant monitoring; in addition, the future possible insect pests are predicted in combination with environmental factors and other factors, so as to take preventive measures in time and effectively, and thereby significantly improve the accuracy and efficiency of the tomato leaf moth identification and monitoring method.

[0202] Please refer to Figure 3 Another aspect of the application provides a tomato leaf moth identification and prediction system based on image recognition, comprising a data collection module, a data analysis module and a warning module, the data collection module and the data analysis module are electrically and / or communicatively connected, the data analysis module and the warning module are electrically and / or communicatively connected, wherein:

[0203] The data collection module: obtains tomato parameters, environmental data and future weather forecast data;

[0204] The data analysis module: inputs the tomato parameters and environmental data into the device parameter adjustment model to obtain a plurality of device adjustment parameters; adjusts the device based on the plurality of device adjustment parameters, and obtains adjustment monitoring data based on the adjusted device; generates a monitoring result according to the adjustment monitoring data; generates a prediction result according to the monitoring result, the environmental data and the future weather forecast data; generates an alarm signal according to the monitoring result and the prediction result;

[0205] The warning module: performs corresponding warning operation according to the alarm signal.

[0206] Some data in the above formula is calculated by removing the dimension and taking the numerical value, and the formula is obtained by software simulation of a large amount of collected data to obtain a formula closest to the real situation; the preset parameters and the preset threshold in the formula are set by the person skilled in the art according to the actual situation or obtained by a large amount of data simulation.

[0207] The working principle of this application is as follows: by obtaining tomato parameters, environmental data and future weather forecast data; inputting the tomato parameters and environmental data into the device parameter adjustment model to obtain a number of device adjustment parameters; adjusting the device based on the several device adjustment parameters, and obtaining adjustment monitoring data based on the adjusted device; generating monitoring results based on the adjusted monitoring data; generating prediction results based on the monitoring results, environmental data and future weather forecast data; generating an alarm signal based on the monitoring results and the prediction results and executing corresponding early warning operations. This application uses data such as the real-time growth status of the plant to adjust the relevant parameters of the collection equipment, making the collected data more accurate, and dynamically adjusts the fusion weights between multimodal data based on the adjusted data quality and parameters such as task requirements, so that the monitoring of the current results is more accurate and closer to actual needs. Subsequently, the severity of the plant is adaptively calculated through parameters such as leaf severity and leaf position, making the monitoring of tomato plants more accurate; at the same time, future plant pests are predicted based on environmental factors, which can be prevented in a timely and effective manner, thereby improving the accuracy and efficiency of the tomato leafminer identification and monitoring method, and avoiding the problem that the existing technology lacks consideration of the task requirements of real-time monitoring and the lack of the ability to predict future pest trends, resulting in low accuracy and efficiency in monitoring tomato leafminers.

[0208] The above embodiments are only used to illustrate the technical method of the present application and are not intended to limit it. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present application.

Claims

1. A method for identifying and predicting tomato leafminers based on image recognition, characterized in that: include: S1, obtain tomato parameters, environmental data and future weather forecast data; S2. Input the tomato parameters and environmental data into the equipment parameter adjustment model to obtain a number of equipment adjustment parameters; S3. adjusting the device based on the plurality of device adjustment parameters, and obtaining adjustment monitoring data based on the adjusted device; S4. Generate monitoring results based on the adjusted monitoring data; S5. Generate forecast results based on monitoring results, environmental data, and future weather forecast data; S6. Generate an alarm signal based on the monitoring results and prediction results and perform corresponding early warning operations.

2. The method for identifying and predicting tomato leafminers based on image recognition according to claim 1, characterized in that: Tomato parameters include tomato root and stem values, tomato height values, and tomato detection images; environmental data includes temperature, humidity, and light intensity; and device adjustment parameters refer to parameters that need to be adjusted for related equipment during data collection, including camera height and shooting angle. The device parameter adjustment model is constructed using an artificial intelligence model. The specific construction steps are as follows: S21. Obtain several historical tomato parameters and historical environmental data and corresponding historical equipment adjustment parameters; S22, dividing the historical tomato parameters and historical environment data and corresponding historical equipment adjustment parameters into training data, verification data, and test data; S23, performing data preprocessing on the training data, the validation data, and the test data to obtain a training set, a validation set, and a test set, wherein the preprocessing includes data denoising and data enhancement; S24. Selecting an artificial intelligence model as a basic model, wherein the selected artificial intelligence model is a neural network model; S25. Use the training set to train the basic model, and adjust the learning rate on the validation set to obtain the pre-trained model; S26. Use the test set to test the pre-trained model. If the test criteria are met, the device parameter adjustment model is obtained; otherwise, return to step S25 and retrain until the test criteria are met.

3. The method for identifying and predicting tomato leafminers based on image recognition according to claim 1, wherein: The adjusted monitoring data in step S3 refers to the tomato data obtained after assigning the device adjustment parameters to the corresponding device, including the total number of leaves, the positions of several leaves, and two consecutive frames of visible light images and near-infrared images corresponding to the several leaves; The monitoring result in step S4 refers to the current state of the plant, including the probability of insect eggs CLL, the probability of larvae YCL, the ratio of the buried channel area QDMB, and the ratio of chlorophyll loss YLSB. The specific steps of generating the monitoring result based on the adjusted monitoring data in step S4 include: S41. Acquire and adjust monitoring data and monitoring labels, wherein the monitoring labels include current labels and predicted labels. The current labels refer to labels required to obtain the current state of the plant, and the predicted labels refer to labels required to obtain the estimated state of the plant in the future. S42. Generate image weights corresponding to the visible light image and the near-infrared image based on the adjusted monitoring data; S43, combining the image weight with the visible light image / near infrared image of the next frame to obtain analysis data; S44. Input the current label and analysis data into the multi-task result generation model to obtain the monitoring result corresponding to the current label, wherein the multi-task result generation model is constructed through a machine learning model.

4. The method for identifying and predicting tomato leafminers based on image recognition according to claim 3, wherein: The specific steps of step S42 are: S421, continuously acquiring two frames of visible light images / near infrared images and the current label; S422: Fusing the visible light image and the near-infrared image of the previous frame according to preset weights to obtain a fused image, and inputting the current label and the fused image into a multi-task result generation model to obtain a previous frame monitoring result corresponding to the current label; S423, generating task sensitivity according to the previous frame monitoring result; S424, extracting the visible image information entropy KJS and the near infrared information entropy HWS corresponding to the visible light image and the near infrared image of the next frame; S425. Construct a visible weight function KQF (KJS, HWS, RMD) based on the task sensitivity RMD in step S423, the visible image information entropy KJS, the near-infrared information entropy HWS in step S424, and the nonlinear relationship between them and the visible light weight coefficient. The visible weight function KQF (KJS, HWS, RMD) is shown in the following formula: Where k is the task gain coefficient, h is the spectral difference attenuation rate, and both k and h are greater than 0; S426. Substitute the visible image information entropy, the near-infrared information entropy, and the task sensitivity into the visible weight function to calculate the visible image weight, and perform a difference operation between the sum of the weight coefficients and the visible image weight to obtain the near-infrared image weight. The calculation formula for the near-infrared image weight HQX is as follows: HQX=1-KQF 5. The method for identifying and predicting tomato leafminers based on image recognition according to claim 4, characterized in that: The specific steps of generating task sensitivity according to the previous frame monitoring result in step S423 include: (1) Extract the egg probability CLL and larva probability YCL from the previous frame monitoring results; (2) Based on the nonlinear relationship between the egg probability, larva probability and task sensitivity, a sensitivity function MDF(CLL, YCL) is constructed, where the sensitivity function MDF(CLL, YCL) is shown as follows: (3) Substitute the egg probability CLL and the larva probability YCL into the sensitivity function MDF(CLL, YCL) to calculate the task sensitivity RMD.

6. The method for identifying and predicting tomato leafminers based on image recognition according to claim 1, characterized in that: The prediction result in step S5 refers to the estimated state of the plant in the future, including the probability of pest infestation and the level of spread trend; The specific steps of generating the prediction results based on the monitoring results, environmental data and future weather forecast data in step S5 include: S51, obtaining current monitoring results and environmental data as well as future weather forecast data; S52, obtaining a number of historical monitoring results and historical environmental data and prediction labels; S53, integrating the current monitoring results, environmental data, and future weather forecast data as well as a number of historical monitoring results and historical environmental data into a prediction sequence in chronological order; S54. Input the prediction label and the prediction sequence into a multi-task result generation model to obtain a prediction result corresponding to the prediction label, wherein the multi-task result generation model is constructed by a machine learning model.

7. The method for identifying and predicting tomato leafminers based on image recognition according to claim 3 or claim 6, characterized in that: The multi-task result generation model is constructed through a machine learning model. The specific construction steps include: (1) Obtain the current label and its corresponding historical analysis data and historical monitoring results, as well as the predicted label and its corresponding historical prediction sequences and historical prediction results; (2) dividing the current label and its corresponding historical analysis data and historical monitoring results into training data, verification data, and test data corresponding to the current label, and performing data preprocessing on the training data, verification data, and test data corresponding to the current label to obtain a training set, a verification set, and a test set; (3) Dividing the prediction label and its corresponding historical prediction sequences and historical prediction results into training data, verification data, and test data corresponding to the prediction label, and performing data preprocessing on the training data, verification data, and test data corresponding to the prediction label to obtain a training set, a verification set, and a test set; (4) Select two machine learning models as the basic models for the current label and the predicted label respectively; (5) Use the respective training sets to train their corresponding basic models, and adjust the learning rate and other hyperparameters on their respective validation sets to obtain their respective pre-trained models; (6) Each test set is tested on its own pre-trained model. If the test criteria are met, a multi-task result generation model corresponding to each label is obtained; otherwise, return to step (5) and retrain until the test criteria are met.

8. The method for identifying and predicting tomato leafminers based on image recognition according to claim 1, characterized in that: The alarm signals in step S6 include a serious plant pest alarm signal, a minor plant pest alarm signal, a plant pest contamination risk signal, and a serious pest spread alarm signal; The specific steps of generating an alarm signal according to the monitoring results and the prediction results in step S6 include: S61, obtaining monitoring results and prediction results corresponding to a number of blades; S62: Extract the pest probability and spread trend level from the prediction results, and determine whether to generate a corresponding alarm signal. The specific determination process is as follows: (1) Determine whether the probability of pests is greater than the probability threshold. If so, generate a plant pest contamination risk signal; if not, do nothing; (2) determining whether the spreading trend level is greater than the level threshold, if so, generating a serious pest spreading alarm signal; if not, performing no operation, wherein the spreading trend level is within the range of 0-1; S63. Generate a plant severity level based on the monitoring results, and determine whether to generate a corresponding alarm signal based on the plant severity level. The plant severity level refers to a quantitative measure of the plant pest situation. The specific determination process is as follows: Determine whether the severity of the plant is greater than the severity threshold. If so, generate a serious plant pest alarm signal; if not, determine whether the severity of the plant is greater than D times the severity threshold. If so, generate a minor plant pest alarm signal, otherwise generate a plant health prompt signal.

9. The method for identifying and predicting tomato leafminers based on image recognition according to claim 8, characterized in that: The specific steps of generating the severity of the plant according to the monitoring results in step S63 include: S631, obtaining monitoring results corresponding to a number of blades, blade positions, and the total number of blades, wherein the blade positions include top blades, middle blades, and bottom blades; S632. Extract the egg probability CLL, larva probability YCL, subsurface area ratio QDMB, and chlorophyll loss ratio YLSB from the monitoring results; S633. Based on the nonlinear relationship between the egg probability CLL, the larval probability YCL, the duct area ratio QDMB, the chlorophyll loss ratio YLSB, and the leaf severity, a leaf severity function YCF(CLL, YCL, QDMB, YLSB) is constructed. The leaf severity function is shown in the following formula: MDF(CLL,YCL,QDMB,YLSB)=max(CLL,YCL)×QDMB+μ×YLSB Where μ represents the loss contribution coefficient, μ∈(0,1); S634, substituting the egg probability, larva probability, duct area ratio, and chlorophyll loss ratio into the leaf severity function to calculate the leaf severity YYD of the plurality of leaves; S635 , obtaining the number of severe leaves YS whose severity is greater than the severe threshold value, and calculating the severe ratio YB based on the ratio between the severe number and the total number of leaves; S636. Construct a plant severity function ZCF(YYD, YW, YB) based on the nonlinear relationship between the leaf severity YYD, leaf position YW, severity ratio YB, and plant severity of the plurality of leaves. The plant severity function is shown in the following formula: Where i is the number of the blade, WQ i is the leaf position weight of the i-th leaf, is the severity of the plant of the i-th leaf in its corresponding YW; ω is the diffusion penalty coefficient, ω∈(0,1); S637. Substitute the leaf severity, leaf position, and severity ratio corresponding to the plurality of leaves into the plant severity function to calculate the plant severity.

10. A tomato leafminer identification and prediction system based on image recognition, used to implement the tomato leafminer identification and prediction method based on image recognition according to any one of claims 1 to 9, characterized in that: include: Data acquisition module, data analysis module and early warning module, the data acquisition module is connected to the data analysis module, the data analysis module is connected to the early warning module, wherein: Data acquisition module: obtains tomato parameters, environmental data and future weather forecast data through data acquisition equipment; Data analysis module: Inputs tomato parameters and environmental data into the equipment parameter adjustment model to obtain several equipment adjustment parameters; adjusts the equipment based on the several equipment adjustment parameters, and obtains adjustment monitoring data based on the adjusted equipment; generates monitoring results based on the adjusted monitoring data; generates prediction results based on the monitoring results, environmental data, and future weather forecast data; and generates alarm signals based on the monitoring results and prediction results; Early warning module: performs corresponding early warning operations according to the alarm signal.

Images