Lithocolletiella esculenta monitoring and predicting system

Abstract

The invention is suitable for the technical field of pest prevention and control, and provides a Lithocolletis esculenta monitoring and predicting system, which comprises a meteorological data acquisition module, a target pest sex luring automatic monitoring device, an Internet of Things module, a far-end display device and a data processing control module, and is characterized in that the data processing control module comprises a YOLOv8sAHSS model and a BP prediction model; the data processing control module is used for calculating the number of the Lithocollettiae at this time on the basis of a YOLOv8sAHSS model and predicting the number of the Lithocollettiae at the next time on the basis of a BP (Back Propagation) prediction model. The system provided by the invention achieves the comprehensive monitoring and intelligent prediction of the Lithocolletiella esculenta, and provides a scientific basis for agricultural decision makers. The system shows relatively high prediction accuracy in application in different regions and years, has good applicability and stability, and provides powerful technical support for prevention and control of Lithocolletis ringoniella in agricultural places such as apple orchards and the like.

Description

Technical Field

[0001] The present invention belongs to the technical field of pest control, and particularly relates to a monitoring and prediction system for Lithocolletis ringoniella. Background Art

[0002] Lithocolletis ringoniella (Matsumura) belongs to the family Gracillariidae of Lepidoptera. It mainly damages apples, Chinese flowering crabapples, pears, peaches, plums, cherries, and hawthorns. The larvae of Lithocolletis ringoniella mine the mesophyll from the back of the host leaves, forming oval insect spots, causing the epidermis on the back of the leaves to shrink and the leaves to bend backward. On the front of the leaves, yellowish-green mesh-like insect spots appear, with black insect excrement inside. The insect spots often occur at the edges of the leaves and, when severe, cover the entire leaf. In the main apple-producing areas of Shaanxi, Shandong, Shanxi, Liaoning, Jilin, Heilongjiang, etc. in China, the damage caused by Lithocolletis ringoniella has been increasing year by year, showing a trend of spreading into disasters. In some orchards, due to the high density of pests and severe damage, a large number of fruit trees shed their leaves prematurely, resulting in insufficient nutrition, weak tree vigor, and reduced economic benefits of fruit tree production.

[0003] With the development of technical means, using image recognition technology to automatically analyze the captured pictures or video materials to monitor the presence and damage degree of pests has become a new effective method. However, how to establish a comprehensive monitoring system and predict the possible pest situations based on the monitoring results to provide strong data support for subsequent pest control work is an important direction worthy of in-depth study at present. For this reason, the present invention proposes a monitoring and prediction system for Lithocolletis ringoniella. Summary of the Invention

[0004] The purpose of the present invention is to provide a monitoring and prediction system for Lithocolletis ringoniella, aiming to solve the problems raised in the above background art.

[0005] The purpose of the present invention is achieved through the following technical solutions:

[0006] A monitoring and prediction system for Lithocolletis ringoniella, comprising:

[0007] A meteorological data collection module for collecting meteorological data of the area to be monitored;

[0008] A target pest sex pheromone automatic monitoring device deployed in the area to be monitored, the target pest sex pheromone automatic monitoring device comprising a collection module and a camera module for collecting the quantity information of Lithocolletis ringoniella;

[0009] A data processing and control module, which is connected to a meteorological data acquisition module and a target pest sex pheromone automatic monitoring device. The data processing and control module is used to receive meteorological data from the meteorological data acquisition module and the number information of Lithocolletis ringoniella from the target pest sex pheromone automatic monitoring device. The data processing and control module includes a YOLOv8s_AHSS model and a BP prediction model. The data processing and control module calculates the number of Lithocolletis ringoniella this time based on the YOLOv8s_AHSS model, and the data processing and control module predicts the number of Lithocolletis ringoniella next time based on the BP prediction model.

[0010] An Internet of Things module, which is connected to the data processing and control module. The data processing and control module is used to transmit the prediction result to the Internet of Things module.

[0011] A remote display device, which is connected to the Internet of Things module. The Internet of Things module is used to transmit the prediction result to the remote display device.

[0012] Further, the meteorological data includes temperature, rainfall, and relative humidity.

[0013] Further, the YOLOv8s_AHSS model uses data augmentation. The activation function of the YOLOv8s_AHSS model is the Hardswish activation function, the loss function is the SIoU loss function, and the YOLOv8s_AHSS model adopts the SimAM attention mechanism.

[0014] Further, the maximum number of training times of the BP prediction model is 1000, the training target accuracy is 0.001, the learning rate is 0.01, and the learning function is the L-M function. The BP prediction model includes an input layer, a hidden layer, and an output layer. There are 12 input layer factors, the hidden layer contains 5 nodes, and the output layer is the number value of the next generation of Lithocolletis ringoniella population.

[0015] Further, the input layer factors include the average temperature in the first 16 - 20 days, the average temperature in the first 11 - 15 days, the average temperature in the first 6 - 10 days, the average temperature in the first 1 - 5 days, the precipitation in the first 16 - 20 days, the precipitation in the first 11 - 15 days, the precipitation in the first 6 - 10 days, the precipitation in the first 1 - 5 days, the pest occurrence in the first 16 - 20 days, the pest occurrence in the first 11 - 15 days, the pest occurrence in the first 6 - 10 days, and the pest occurrence in the first 1 - 5 days.

[0016] Further, the network connection weight value from the input layer to the hidden layer is W1, the node threshold is b1, and the logarithmic sigmoid function is used as the transfer function.

[0017] Further, the network connection weights from the hidden layer to the output layer are W2, the node threshold is b2, and the linear transfer function is used as the transfer function.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0019] The Lithocolletis ringoniella monitoring and prediction system provided by the present invention realizes the comprehensive monitoring and intelligent prediction of Lithocolletis ringoniella by integrating functions such as meteorological data collection, automatic monitoring of target pest sex pheromones, data processing and control, Internet of Things transmission, and remote display. The system uses high-precision image recognition technology to accurately count Lithocolletis ringoniella, and uses an intelligent prediction model to predict the next pest quantity based on the meteorological data of the previous 20 days and the real-time monitoring data of pests, providing a scientific basis for agricultural decision-makers. The system has shown a high prediction accuracy in applications in different regions and years, with good applicability and stability, providing strong technical support for the prevention and control of Lithocolletis ringoniella in agricultural sites such as apple orchards. Description of the Drawings

[0020] Figure 1 It is the system structure diagram of the present invention.

[0021] Figure 2 It is to label the target pests using the labelimg software.

[0022] Figure 3 It is for the training of the YOLOv8s model (Lithocolletis ringoniella (Matsumura) in the figure represents Lithocolletis ringoniella); among them, (a) is the training process of the model, (b) is the Precision-Confidence curve of the model, (c) is the Recall-Confidence curve of the model, (d) is the PR curve of the model, and (e) is the F1-Confidence curve of the model.

[0023] Figure 4 It is the detection result of the optimized YOLOv8s_AHSS model for the target pests.

[0024] Figure 5 It is the comparison between the actual value and the predicted value of the number of Lithocolletis ringoniella in the model training set.

[0025] Figure 6 It is the comparison between the actual value and the predicted value of the number of Lithocolletis ringoniella in the apple orchard in Yangling area in 2023.

[0026] Figure 7 It is the comparison between the actual value and the predicted value of the number of Lithocolletis ringoniella in the apple orchard in Yangling area in 2024.

[0027] Figure 8Comparison of the actual and predicted numbers of Lithocolletis ringoniella in apple orchards in Liquan area in 2024. Detailed implementation mode

[0028] For a clearer understanding of the technical features, objectives, and beneficial effects of the present invention, the technical solution of the present invention will now be described in detail below, but it should not be construed as a limitation on the scope of implementation of the present invention.

[0029] The following describes the specific implementation of the present invention in detail with reference to specific embodiments.

[0030] As Figure 1 shown, the present invention provides a monitoring and prediction system for Lithocolletis ringoniella, and the system includes:

[0031] A meteorological data acquisition module for acquiring meteorological data of the area to be monitored;

[0032] A target pest sex pheromone automatic monitoring device deployed in the area to be monitored, and the target pest sex pheromone automatic monitoring device includes an acquisition module and a camera module for collecting information on the number of Lithocolletis ringoniella;

[0033] A data processing and control module, the data processing and control module is connected to the meteorological data acquisition module and the target pest sex pheromone automatic monitoring device, and the data processing and control module is used to receive meteorological data from the meteorological data acquisition module and information on the number of Lithocolletis ringoniella from the target pest sex pheromone automatic monitoring device; the data processing and control module includes a YOLOv8s_AHSS model and a BP prediction model, the data processing and control module calculates the number of Lithocolletis ringoniella this time based on the YOLOv8s_AHSS model, and the data processing and control module calculates the predicted number of Lithocolletis ringoniella next time based on the BP prediction model;

[0034] An Internet of Things module, the Internet of Things module is connected to the data processing and control module, and the data processing and control module is used to transmit the prediction result to the Internet of Things module;

[0035] A remote display device, the remote display device is connected to the Internet of Things module, and the Internet of Things module is used to transmit the prediction result to the remote display device.

[0036] Among them:

[0037] First, the data processing and control module realizes automatic recognition and counting of Lithocolletis ringoniella monitoring images based on the YOLOv8s_AHSS model to calculate the number of Lithocolletis ringoniella this time.

[0038] 1. Data collection;

[0039] By using a target pest sex pheromone automatic monitoring device (including a collection module and a camera module), photos of Lithocolletis ringoniella pests were collected in apple orchards in Yangling and Liquan in 2023. A total of 511 images with a pixel size of 3072×4096 were collected for the dataset required for subsequent model training.

[0040] 2. Data processing;

[0041] 2.1 Data annotation;

[0042] Through data annotation ( Figure 2 ), rich training samples are provided for the learning model, enabling it to learn the features, patterns, and relationships in the images. This data annotation information can help the model better understand the image content, thus achieving more accurate object detection tasks.

[0043] 2.2 Dataset division;

[0044] Select the images with target pests among the 511 collected pictures as the dataset (Table 1), and randomly divide them into a training set, a validation set, and a test set according to the ratio of 7:2:1. Among them, there are 358 images in the training set, 102 images in the validation set, and 51 images in the test set. The training set is mainly used to train the model, the validation set is used to evaluate the model performance, and the test set is mainly used to test the final model effect.

[0045] Table 1 Division of the training set, validation set, and test set after annotating the dataset pictures

[0046]

[0047] 2.3 Data augmentation;

[0048] Data augmentation is an important means to increase the model training data in the case of a small amount of data. The data augmentation used in this invention includes hsv_h: 0.015 (hue), hsv_s: 0.7 (saturation), hsv_v: 0.4 (brightness), translate: 0.1 (image translation), scale: 0.5 (image scaling), fliplr: 0.5 (image flipping), and mosaic: 1.0 (multi-image splicing).

[0049] 3. Model evaluation metrics;

[0050] True positive (TP) indicates that the model successfully identifies the target pest; false positive (FP) indicates that the model identifies a non-target pest as a target pest; true negative (TN) indicates that the model correctly excludes non-target pests; false negative (FN) indicates that the model fails to identify the target pest.

[0051] Precision represents the accuracy of the model, that is, the proportion of true samples in the detected positive class samples.

[0052]

[0053] The recall rate indicates the ratio of the number of positive class samples detected by the model to the total number of positive class samples.

[0054]

[0055] Average Precision, that is, with the recall rate as the X-axis and the accuracy rate as the Y-axis, calculates the area under the PR curve.

[0056]

[0057] F1-Score is the harmonic mean of the accuracy rate and the recall rate. It comprehensively considers the precision rate and the recall rate and is used to balance the relationship between the two. The higher the value of F1-Score, the better the comprehensive performance of the model in terms of both the accuracy rate and the recall rate.

[0058]

[0059] 4. Test environment configuration and parameter settings;

[0060] Table 2 Environment configuration

[0061]

[0062] Parameter settings: The number of training epochs is 500, the input image resolution is 640×640, and the Stochastic Gradient Descent (SGD) method is used for optimization in all cases, with a momentum of 0.937 and a learning rate of 0.01.

[0063] 5. Training process;

[0064] Figure 3 In (a), it shows the change process of each index generated by the model during the training process. During the training process, as the number of training times increases, various loss values continuously decrease and tend to be stable, while the values of Precision, Recall, and AP gradually increase with the continuous improvement of the model performance and finally converge. Figure 3 In (b), it is the Precision-Confidence curve of the model. It can be seen that the confidence level of the model continuously increases as the accuracy rate increases. Figure 3 In (c), it is the Recall-Confidence curve of the model. When the curve shows a high recall rate at a high confidence level, it indicates that the algorithm can accurately predict the presence of the target during object detection and can still maintain a high recall rate after filtering out the prediction boxes with low confidence levels. This reflects the good performance of the algorithm in the object detection task. Figure 3In (d) is the PR curve of the model, where the abscissa is the recall rate and the ordinate is the precision. The closer the curve is to the upper right, the higher the precision and recall rate the model can ensure simultaneously during prediction, that is, the prediction results are more accurate. Figure 3 In (e) is the F1-Confidence curve of the model, which shows the change of its F1 score value under different confidence levels.

[0065] 6. Comparison of training results of different models;

[0066] From the training results of different models (Table 3), the comprehensive performance of YOLOv8s on the constructed pest dataset is significantly better. While maintaining a high level of recall rate and F1 value, its training time is significantly shorter, which can save a large amount of computer resources. Therefore, from all aspects, YOLOv8s is selected as the baseline model for further optimization of the subsequent model (Table 4). It can be seen from Table 4 that the total number of images used in the selected YOLOv8s baseline model is 358, a total of 22,692 Lithocolletis ringoniella are labeled, the average precision of the model is 82.6%, the recall rate is 87.0%, and the detection speed (FPS, Frames Per Second, that is, the detection speed) is 4 frames per second.

[0067] Table 3 Comparison of training results of different models

[0068]

[0069] Table 4 Final training results of the baseline model

[0070]

[0071] Using activation functions can introduce non-linear factors to neurons, enabling the neural network to approximate any non-linear function arbitrarily and making the expression ability of deep neural networks more powerful. By replacing the original SiLU in the baseline model with different activation functions, it is found that the Hardswish activation function has the most significant improvement in model performance, with its AP value increasing by 1.1% and the frames per second (FPS) increasing from 4 to 10 (Table 5).

[0072] Table 5 Training effects of the baseline model YOLOv8s under different activation functions

[0073]

[0074] The error between the detection box and the target box (i.e., the IoU of the bounding box loss) is a key indicator for evaluating the training effect of the model. By using different loss functions to obtain the best evaluation criteria required for the model under the self-built dataset, it is found that after replacing the original CIoU loss function of the model with SIoU, the AP value and recall rate are increased by 1.1% and 2.2% respectively, but there is also a certain loss in the detection speed (Table 6).

[0075] Table 6 Training effects of the baseline model YOLOv8s under different loss functions

[0076]

[0077] The addition of the attention mechanism can enable the model to focus on some more critical target pest features and reduce the extraction of those redundant features. After adding different attention mechanisms to the backbone network of YOLOv8s, it is found that adding the SimAM attention mechanism can increase the AP value and recall rate of the model by 1.6% and 1.2% respectively (Table 7), and the detection speed (FPS) is improved.

[0078] Table 7 Adding attention mechanisms at the backbone position of the model

[0079]

[0080] As can be seen from Table 8, by combining different modules, it is found that the model (named YOLOv8s_AHSS model) after simultaneously using data augmentation (Augmentation), Hardswish activation function, SIoU loss function, and adding the SimAM attention mechanism to the backbone network performs best on the self-built Lithocolletis ringoniella pest dataset. Compared with the original YOLOv8s model, its average precision and recall rate are increased by 6.0% and 6.4% respectively, and the detection speed (FPS) decreases slightly.

[0081] Table 8 Ablation experiments of the model under different modules

[0082]

[0083]

[0084] 7. Detection results of pest pictures;

[0085] From the detection results ( Figure 4 ), the optimized model is better than the original YOLOv8s model in both detection accuracy (false detection) and recall rate (missed detection). Even when detecting pest images with different densities, the model can still show good performance, and there is an obvious improvement in the model generalization and robustness compared with the original model.

[0086] Second, the data processing and control module predicts the number of the next generation of Lithocolletis ringoniella based on the short-term BP prediction model of Lithocolletis ringoniella.

[0087] 1. Data source;

[0088] Historical data, field meteorological data of apple orchards in Yangling and Liquan in 2023 and 2024, and the survey data of the cumulative number of Lithocolletis ringoniella every 5 days were all collected by the target pest sex pheromone automatic monitoring device.

[0089] 2. Research method;

[0090] Environmental conditions such as field temperature and precipitation, and the pest population base significantly affect the biological characteristics of insects such as development and reproduction, which are important factors for predicting the pest population. Therefore, the number of Lithocolletis ringoniella investigated every 5 days (heads / trap) was used as the dependent variable, and 12 factors including the average temperature (°C) in the first 16 - 20 days, the average temperature (°C) in the first 11 - 15 days, the average temperature (°C) in the first 6 - 10 days, the average temperature (°C) in the first 1 - 5 days, the precipitation (mm) in the first 16 - 20 days, the precipitation (mm) in the first 11 - 15 days, the precipitation (mm) in the first 6 - 10 days, the precipitation (mm) in the first 1 - 5 days, the pest occurrence (heads / trap) in the first 16 - 20 days, the pest occurrence (heads / trap) in the first 11 - 15 days, the pest occurrence (heads / trap) in the first 6 - 10 days, and the pest occurrence (heads / trap) in the first 1 - 5 days were selected as independent variables to construct the model (Table 9). The historical data and the survey data (a total of 332) from the experimental sites in Liquan and Yangling in 2023 and 2024 were divided into a training set and a test set. Among them, 275 historical data were used as the training set for training the model, and 57 survey data from the experimental sites in Liquan and Yangling in 2023 and 2024 were used as the test set to detect the prediction effect of the model. Taking the occurrence of Lithocolletis ringoniella as the research object, using the relevant factor data in Table 9, a Matlab neural network algorithm program was written to establish a BP network prediction model.

[0091] In this experiment, the network connection weight from the input layer to the hidden layer was W1, the node threshold was b1, and the logarithmic sigmoid function (logsig) was used as the transfer function (tansig). The network connection weight from the hidden layer to the output layer was W2, the node threshold was b2, and the linear transfer function (purelin) was used as the transfer function. The maximum number of training times was set to 1000. During the debugging process of the BP neural network model, the trial-and-error method was used to adjust other parameters: different target accuracies (10 -9-0.9), learning functions (standard BP function traingd, BP function with momentum traingdm, resilient BP function trainrp, BP function with variable learning rate traindx, conjugate gradient BP function traincg, and L-M function trainlm), learning rate (0.01 - 0.9), and number of hidden layer nodes to train, and select the parameter settings with the best effect to train the model. During the debugging process, mean absolute error (MAE) and mean square error (MSE) are selected to evaluate the performance of the model.

[0092]

[0093] Among them, l is the number of hidden layer nodes, m is the number of input layer nodes, n is the number of output layer nodes, and α takes a constant in the range of [1 - 10]. y i is the actual survey value, is the predicted value of the BP network model, and n is the number of training samples. The smaller the MAE value, the smaller the error of the model; the smaller the MSE value, the better the stability of the model.

[0094] Table 9 Factors of the Short-Term BP Prediction Model for the Occurrence Quantity of Lithocolletis ringoniella

[0095]

[0096]

[0097] 3. Research Results;

[0098] According to the model parameter screening results in Table 10, the structure of the short-term BP prediction model for Lithocolletis ringoniella is summarized as follows:

[0099] There are 12 input layer factors, namely the average temperature (°C) in the first 16 - 20 days, the average temperature (°C) in the first 11 - 15 days, the average temperature (°C) in the first 6 - 10 days, the average temperature (°C) in the first 1 - 5 days, the precipitation (mm) in the first 16 - 20 days, the precipitation (mm) in the first 11 - 15 days, the precipitation (mm) in the first 6 - 10 days, the precipitation (mm) in the first 1 - 5 days, the pest occurrence quantity (number per trap) in the first 16 - 20 days, the pest occurrence quantity (number per trap) in the first 11 - 15 days, the pest occurrence quantity (number per trap) in the first 6 - 10 days, and the pest occurrence quantity (number per trap) in the first 1 - 5 days. The network connection weights from the input layer to the hidden layer are W1, the node threshold is b1, and the logarithmic sigmoid function (logsig) is used as the transfer function (tansig). The hidden layer contains 5 nodes. The network connection weights from the hidden layer to the output layer are W2, the node threshold is b2, and the linear transfer function (purelin) is used as the transfer function. The maximum number of training times is set to 1000, the training target accuracy is 0.001, the learning rate is 0.01, the learning function is the L - M function (trainlm), and the output layer is the next population quantity value of Lithocolletis ringoniella (number per trap).

[0100] Table 10 Screening Results of Parameters of Short - term BP Prediction Model for the Occurrence Quantity of Lithocolletis ringoniella

[0101]

[0102]

[0103]

[0104] 4. Prediction historical coincidence rate;

[0105] Using the BP prediction model, the historical coincidence rate is calculated for the training set of 275 historical data. The true values and predicted values of each data are shown in Figure 5 . The results show that the historical coincidence rate of the BP prediction model is 76.96%, indicating that the model has good prediction ability.

[0106] Example 1: This example provides an automatic monitoring and prediction system for Lithocolletis ringoniella, which was applied to apple orchards in Yangling area in 2023 for testing. The specific steps are as follows:

[0107] S1: Deploy the target pest sex pheromone automatic monitoring device (the sex pheromone is the sex pheromone of Lithocolletis ringoniella) in the area to be monitored;

[0108] S2: The data processing and control module receives meteorological data from the meteorological data acquisition module, including temperature, rainfall, and relative humidity;

[0109] S3: The data processing and control module receives the number information of Lithocolletis ringoniella from the automatic target pest sex pheromone monitoring device, calculates the number of Lithocolletis ringoniella next time, and the recognition and counting model is as described above.

[0110] S4: The data processing and control module calculates the predicted number of Lithocolletis ringoniella next time based on the prediction model, and the prediction model is as described above; and transmits the prediction result to the remote display device through the Internet of Things module.

[0111] The prediction result is as Figure 6 shown, and the prediction accuracy rate is 72.27%.

[0112] Example 2: This example provides an automatic monitoring and prediction system for Lithocolletis ringoniella, which was applied to apple orchards in Yangling area in 2024 for testing. The specific steps are as follows:

[0113] S1: Deploy the automatic target pest sex pheromone monitoring device (the sex pheromone is the sex pheromone of Lithocolletis ringoniella) in the area to be monitored;

[0114] S2: The data processing and control module receives the meteorological data from the meteorological data acquisition module, including temperature, rainfall, and relative humidity;

[0115] S3: The data processing and control module receives the number information of Lithocolletis ringoniella from the automatic target pest sex pheromone monitoring device, calculates the number of Lithocolletis ringoniella this time, and the recognition and counting model is as described above.

[0116] S4: The data processing and control module calculates the predicted number of Lithocolletis ringoniella next time based on the prediction model, and the prediction model is as described above; and transmits the prediction result to the remote display device through the Internet of Things module.

[0117] The prediction result is as Figure 7 shown, and the prediction accuracy rate is 70.82%.

[0118] Example 3: This example provides an automatic monitoring and prediction system for Lithocolletis ringoniella, which was applied to apple orchards in Liquan area in 2024 for testing. The specific steps are as follows:

[0119] S1: Deploy the automatic target pest sex pheromone monitoring device (the sex pheromone is the sex pheromone of Lithocolletis ringoniella) in the area to be monitored;

[0120] S2: The data processing and control module receives the meteorological data from the meteorological data acquisition module, including temperature, rainfall, and relative humidity;

[0121] S3: The data processing and control module receives the number information of Lithocolletis ringoniella from the automatic target pest sex pheromone monitoring device, calculates the number of Lithocolletis ringoniella this time, and the recognition and counting model is as described above.

[0122] S4: The data processing and control module calculates the predicted number of Lithocolletis ringoniella next time based on the prediction model, as described above; and transmits the prediction result to the remote display device through the Internet of Things module.

[0123] The prediction result is as Figure 8 shown, and the prediction accuracy rate is 74.48%.

[0124] Summary: Through the tests of Embodiments 1-3, the automatic monitoring and prediction system for Lithocolletis ringoniella has shown high prediction accuracy rates in applications in different regions and years, which are 72.27% (Yangling in 2023), 70.82% (Yangling in 2024), and 74.48% (Liquan in 2024) respectively. This indicates that the system has good stability and applicability under different environmental conditions, and can provide a scientific basis and technical support for the prevention and control of Lithocolletis ringoniella in apple orchards.

[0125] The above is only the preferred embodiment of the present invention. It should be noted that for those skilled in the art, without departing from the concept of the present invention, several deformations and improvements can still be made, and these should also be regarded as the protection scope of the present invention, and these will not affect the implementation effect of the present invention and the practicability of the patent.

Claims

1. Golden apple moth monitoring and prediction system, characterized in that, Including: A meteorological data acquisition module for acquiring meteorological data of the area to be monitored; A target pest sex pheromone automatic monitoring device deployed in the area to be monitored. The target pest sex pheromone automatic monitoring device includes an acquisition module and a camera module for collecting the number information of Lithocolletis ringoniella; A data processing and control module. The data processing and control module is connected to the meteorological data acquisition module and the target pest sex pheromone automatic monitoring device. The data processing and control module is used to receive the meteorological data from the meteorological data acquisition module and the number information of Lithocolletis ringoniella from the target pest sex pheromone automatic monitoring device. The data processing and control module includes a YOLOv8s_AHSS model and a BP prediction model. The data processing and control module calculates the number of Lithocolletis ringoniella this time based on the YOLOv8s_AHSS model, and the data processing and control module predicts the number of Lithocolletis ringoniella next time based on the BP prediction model; An Internet of Things module. The Internet of Things module is connected to the data processing and control module. The data processing and control module is used to transmit the prediction result to the Internet of Things module; A remote display device. The remote display device is connected to the Internet of Things module. The Internet of Things module is used to transmit the prediction result to the remote display device.

2. The Lithocolletis ringoniella monitoring and prediction system according to claim 1, wherein The meteorological data includes temperature, rainfall and relative humidity.

3. The Lithocolletis ringoniella monitoring and prediction system according to claim 1, wherein, The YOLOv8s_AHSS model uses data augmentation. The activation function of the YOLOv8s_AHSS model is the Hardswish activation function, the loss function is the SIoU loss function, and the YOLOv8s_AHSS model adopts the SimAM attention mechanism.

4. The Lithocolletis ringoniella monitoring and prediction system according to claim 1, wherein The maximum number of training times of the BP prediction model is 1000, the training target accuracy is 0.001, the learning rate is 0.01, and the learning function is the L-M function. The BP prediction model includes an input layer, a hidden layer and an output layer. There are 12 input layer factors, the hidden layer contains 5 nodes, and the output layer is the number value of the next generation of Lithocolletis ringoniella population.

5. The Lithocolletis ringoniella monitoring and prediction system according to claim 4, wherein The input layer factors include the average temperature in the first 16 - 20 days, the average temperature in the first 11 - 15 days, the average temperature in the first 6 - 10 days, the average temperature in the first 1 - 5 days, the precipitation in the first 16 - 20 days, the precipitation in the first 11 - 15 days, the precipitation in the first 6 - 10 days, the precipitation in the first 1 - 5 days, the pest occurrence in the first 16 - 20 days, the pest occurrence in the first 11 - 15 days, the pest occurrence in the first 6 - 10 days, and the pest occurrence in the first 1 - 5 days.

6. The Lithocolletis ringoniella monitoring and prediction system according to claim 4, characterized in that, The network connection weight from the input layer to the hidden layer is W1, the node threshold is b1, and the logarithmic S-shaped function is used as the transfer function.

7. The Lithocolletis ringoniella monitoring and prediction system according to claim 4, characterized in that The network connection weight from the hidden layer to the output layer is W2, the node threshold is b2, and the linear transfer function is used as the transfer function.

Images