Remote disease and pest monitoring system

Abstract

The invention discloses a remote disease and pest monitoring system, and the system comprises a data obtaining module which is used for obtaining environment data and disease and pest image information of a target region; the analysis module is used for analyzing the image information based on a pre-trained deep learning model so as to improve the pest and disease identification precision; the communication early warning module is used for transmitting the recognition result to the remote terminal, displaying the recognition result on the remote terminal and carrying out early warning, the remote disease and pest monitoring system can remarkably improve the recognition precision and the system stability, the image definition and the detail capturing capability under different illumination conditions are enhanced, and the remote disease and pest monitoring system is high in practicability. The method can more effectively cope with the complex ecological environment of the target area, reduces the interference of weeds and non-target crops, improves the recognition accuracy, improves the precision of pest and disease damage monitoring, protects the environment, provides data support for pest and disease damage prevention and control, and reduces the damage of pest and disease damage to garden crops.

Description

Technical Field

[0001] The invention relates to the field of garden information technology, and in particular to a remote pest and disease monitoring system. Background Art

[0002] In the modern garden industry, remote pest and disease monitoring technology has gradually been widely used as a new garden industry management tool. This technology usually uses a combination of remote sensing technology and ground monitoring equipment to achieve real-time monitoring of garden pests and diseases. By collecting data related to garden crop pests and diseases, such as spectral information, plant growth status images, etc., and combining weather conditions and geographic information, based on the Internet of Things technology, data analysis is performed in the background to identify data change patterns, thereby achieving timely warnings and making prevention and control recommendations. This method can effectively improve the accuracy of garden industry management, reduce the use of drugs, protect the environment, and reduce the impact of pests and diseases on crops.

[0003] However, in practical applications, remote pest and disease monitoring systems face the problem of low recognition accuracy. This problem often arises from multiple aspects: first, the low quality of collected images is one of the main factors, especially when the lighting conditions are poor or the equipment performance is limited, the image clarity and detail capture ability are weak; second, the pest and disease recognition algorithm is not advanced enough or cannot adapt to the complex ecological environment of the garden industry. For example, interference from weeds and non-target crops will lead to a decrease in recognition rate. In addition, due to the wide variety of pests and diseases, similar appearance characteristics, and the influence of complex factors such as ambient lighting and crop growth status, the system's ability to accurately distinguish target objects is limited. These problems significantly affect the application effect and scope of remote pest and disease monitoring methods. Summary of the invention

[0004] The purpose of the present invention is to provide a remote pest monitoring system to solve the problem of low recognition accuracy faced by the prior art remote pest monitoring system.

[0005] To achieve the above object, the present invention provides the following technical solution: a remote pest monitoring system, the system comprising:

[0006] Data acquisition module, used to obtain environmental data and pest image information of the target area;

[0007] An analysis module, used to analyze the image information based on a pre-trained deep learning model to improve the accuracy of pest and disease identification;

[0008] The communication warning module is used to transmit the recognition results to the remote terminal and display the recognition results on the remote terminal and issue a warning.

[0009] Preferably, the data acquisition module acquires the target area environmental data and pest image information including:

[0010] The sensor array deployed in the target area collects garden environment parameters in real time, wherein the parameters include at least humidity, temperature, and light intensity;

[0011] Start an unmanned aerial vehicle equipped with a high-definition camera and control it to fly to a predetermined height above the garden crops;

[0012] Capture images of crops in the garden according to preset shooting frequency or trigger conditions;

[0013] The environmental parameters collected by the sensor are combined with the crop images captured by the high-definition camera as the acquired garden environment data and pest and disease image information.

[0014] Preferably, the analysis module analyzes the image information based on a pre-trained deep learning model to improve the accuracy of pest identification, including:

[0015] Preprocess the acquired pest and disease image information, including scaling, denoising and brightness normalization;

[0016] Use the labeled pest and disease sample data to conduct preliminary training on the convolutional neural network model and fine-tune its parameters;

[0017] Applying the fine-tuned deep learning model to perform reasoning analysis on the pest and disease image to be identified to determine whether there are signs of a specific type of disease or pest;

[0018] Based on the characteristic values ​​of the identified disease spots or insect bodies, the distribution density and health of the insects in the field are evaluated to quantify the severity of the disease;

[0019] The identification labels and confidence indicators processed by the model are output as the key basis for identifying pests and diseases.

[0020] Preferably, the communication warning module transmits the recognition result to the remote terminal, displays the recognition result on the remote terminal, and issues a warning, including:

[0021] Encode the analysis and identification results in a specific format and convert them into data packets suitable for transmission;

[0022] Call the pre-defined wireless communication module (such as LTE, LoRa) to initiate the upload command according to the configuration strategy;

[0023] After being sent to the remote monitoring platform, the cloud server receives and unpacks the data to extract the raw data and identification analysis results;

[0024] The processed results are presented to the user and the current status and threat level of garden crops are displayed in a visual way;

[0025] When the set risk threshold is reached, an alarm notification is immediately sent to the designated device or application to prompt the administrator to take necessary actions.

[0026] Preferably, the analysis module analyzes the image information based on a pre-trained deep learning model to improve the accuracy of pest and disease identification, specifically including:

[0027] Use labeled image datasets for model training;

[0028] Input the acquired garden environment image into the trained model;

[0029] Calculate the probability of each category based on the recognition results output by the model; and

[0030] The recognition result is determined when the probability of a certain category exceeds the set baseline.

[0031] Preferably, the calculation of the probability of each category based on the recognition result output by the model includes:

[0032] Set the probability of each class to be initialized to zero;

[0033] According to the model prediction output, add the corresponding probability to the predicted value of each category; and

[0034] Calculate the probability score p of the predicted value and of each category; and when the score p of a certain category is higher than the set probability threshold θ, confirm that the predicted value of the category is the final pest identification result.

[0035] Preferably, when a category score p is higher than a set probability threshold θ, confirming the predicted value of the category as the final pest identification result further comprises:

[0036] Determine the class score p equal to its classification probability;

[0037] Set the relationship between the classification probability and the predefined threshold, θ represents the set threshold; and p>θ, then high confidence recognition is satisfied;

[0038] Marking the results of the pest identification and associating the corresponding category label T;

[0039] Based on the high confidence identification of the category T, the tag is uploaded to the server as a remote warning.

[0040] Preferably, the uploading of the mark to the server as a remote warning based on the high-confidence identification of the category T specifically includes:

[0041] Check whether the marked data packet has integrity M (if M is true, it is intact, otherwise it is discarded); if the marked data loss exceeds the proportion δ of the total information volume, M is not true,

[0042] Transmit valid data to the designated server address S based on the network communication protocol;

[0043] Before receiving the confirmation signal C, the current pest identification message is retained for backup; and when the server responds to the confirmation signal C, the memory can be released.

[0044] The results are finally displayed on the user interface for decision support and implementation of emergency measures.

[0045] It can be seen from the above technical solution that the present invention has the following beneficial effects:

[0046] The remote pest and disease monitoring system obtains target area environmental data and pest and disease image information through a data acquisition module. The analysis module analyzes the image information based on a pre-trained deep learning model to improve the accuracy of pest and disease identification. The communication warning module transmits the identification result to the remote terminal and displays the identification result on the remote terminal and issues a warning. It can significantly improve the recognition accuracy and system stability, enhance the image clarity and detail capture capabilities under different lighting conditions, and more effectively cope with complex ecological environments, reduce weeds, and quickly and accurately identify interference from non-target crops including diseases, cordyceps, and weeds, thereby improving the accuracy of identification. It not only improves the accuracy of pest and disease monitoring, but also effectively reduces the use of drugs, protects the environment, provides data support for pest and disease prevention and control, and reduces the damage of pests and diseases to garden crops, solving the problem of low recognition accuracy faced by remote pest and disease monitoring systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a schematic diagram of the connection of the system modules of the present invention. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] like Figure 1 As shown, a remote pest monitoring system comprises:

[0050] The data acquisition module acquires the target area environmental data and pest image information. After collecting the corresponding images and environmental data, the analysis module analyzes the image information based on the pre-trained deep learning model to improve the accuracy of pest identification. It analyzes and processes the image data based on a pre-trained deep learning network to identify the possible threat types, and continuously verifies the accuracy through machine self-improvement to improve the efficiency and accuracy of the entire monitoring system. This process can be specifically manifested as a convolutional neural network that has become very professional after being educated by tens of millions of annotated garden crop images. Automatically check the newly input corn leaf samples, it will quickly determine whether the sample has signs of downy mildew or other common pest bites and damage based on previous learning experience.

[0051] The communication warning module transmits the identification results to the remote terminal and displays the identification results and issues warnings on the remote terminal. Finally, it requires that the results of the above two steps be sent back to the remote operation interface through modern communication technology, so that relevant personnel can know the current situation and receive corresponding warning audio and video information, so as to arrange necessary prevention and control actions such as pesticide treatment or physical prevention and control in time to prevent the spread of the disaster and cause greater losses. For example, users can receive information pop-ups about suspected wheat aphid infestation at a specific location in the field in the mobile phone application, and the system will also make suggestions for the next response strategy based on the current situation. In this way, even if the garden managers are in the city office, they do not need to worry about missing important intelligence notifications and neglecting the normal development of garden production activities.

[0052] In order to overcome the traditional unsatisfactory accuracy of pest and disease detection due to technical limitations, this solution adopts the most advanced mathematical algorithm architecture of brain neural simulation, that is, the deep training model structure mentioned above. It establishes a knowledge representation mechanism with super pattern matching capabilities by imitating the information propagation and interaction between human brain cells. This special framework has strong adaptability and expansion potential, allowing us to extract key factors from large and complex real-world scenarios to construct highly generalized classification and recognition models, ensuring that the target of interest can be quickly locked even in adverse environments with severe noise interference. Therefore, it greatly improves and breaks through the bottleneck situation caused by the signal-to-noise ratio of early solutions, and realizes a one-stop automated operation mode from perception, analysis, prediction to action guidance, greatly simplifying and accelerating the speed and reliability of the process chain of the garden crop protection cycle.

[0053] According to the content of the data acquisition module, a series of environmental monitoring sensor devices are deployed in the garden, including the functions of measuring humidity levels, recording ambient temperature changes, and detecting the intensity of sunlight or artificial lighting. Such operations can realize the simultaneous monitoring of multiple influencing factors in the garden microenvironment.

[0054] The next step is to activate and operate a drone equipped with a high-resolution video camera, and guide it to a specific flight altitude above the gardening site under the guidance of a set path. This measure aims to use the flexibility and efficiency of drones to provide detailed aerial coverage of the vast fields.

[0055] Then, according to the pre-compiled work cycle table or the basis of induction activation, continuous or intermittent photography and recording of cultivated land vegetation communities are carried out from top to bottom. This step requires the application of pre-arranged image acquisition rules to the actual situation to ensure that the acquired view samples can represent the overall situation of the current growth status of garden crops.

[0056] As for the relevant data obtained in the previous two steps, including the graphic data of the environmental conditions, symptoms and pest infestation phenomena, it is summarized to form a set of integrated intelligence for analysis and comparison, which enables experts or automated software to carry out targeted decision-making in a timely manner after receiving these comprehensive materials to ensure the healthy growth of crops from being damaged by adverse factors. For example, when a certain area has experienced abnormally hot and humid weather for several weeks in a row and yellow spots appear on the leaf pictures many times, it can be alerted that a certain fungal disease is spreading, so the prevention and control drugs are sprayed in time to prevent the disaster from spreading and affecting the entire garden production.

[0057] The system can be divided into front-end identification, terminal analysis, terminal warning and solutions. According to different users, the front-end identification is divided into two types. The first is for non-professional users, such as ordinary flower and grass users, who have little knowledge of pests and diseases. They can take photos of the pest parts with their mobile phones, input the photos into the app system, and the system automatically identifies the types of pests and diseases and gives specific prevention and control plans. The second is for professional users, such as gardening planting, garden green space maintenance and forest professional maintenance. The front end uses fixed or non-fixed cameras to monitor the current area. The camera captures and transmits the data to the terminal system for identification, and issues a warning to the user.

[0058] The first step is to preprocess the acquired pest and disease image information. This is mainly to optimize the image quality input into the model and thus improve the accuracy of pest and disease detection. Specific measures may include the following image preprocessing techniques, such as resizing to ensure uniform specifications of input data, using noise reduction methods to remove information errors or redundant parts in image data caused by environmental conditions, and using brightness normalization operations to control the image brightness within a certain range to avoid result deviations caused by large differences in brightness. For example, in a real-time monitoring system for rice pests and diseases, the camera may capture different images of rice fields in various environments. After a series of preprocessing actions, the input data meets the requirements to achieve better classification results.

[0059] The second step is to use the labeled pest and disease sample data to conduct the initial stage of learning for the pre-established convolutional neural network model and adjust the various parameters in the model. This work is called the initial training and tuning process of the model, and its purpose is to improve the algorithm's ability to identify specific objects. In practice, we may teach the machine to distinguish the difference between marked diseased and normal rice leaves and constantly modify its operation method to adapt to the variables that may be encountered in actual situations. After completing this process, the algorithm should have the ability to independently determine whether garden crops are threatened by diseases to a certain extent without the need for too much human assistance to make a more accurate conclusion.

[0060] The next step is to apply the deep learning model optimized and corrected in the above steps to actual object inspection activities in unknown image sets in unknown areas, that is, to use it as a pest monitoring tool to detect whether there are signs of the target category in the sample to be tested and evaluate the probability of the target category label appearing. This application practice usually refers to directly applying the trained disease model to the freshly taken images in the current pending state to find any local features or overall contours that are suspected to be the focus of the problem. When the system recognizes certain similar patterns, it will make judgments based on the information learned in the past and then output the corresponding results to prompt the user to pay attention to the possible existence of pests and diseases in the area. For example, the filter will pick out individuals with different color and morphology from normal in a group of newly photographed leaves.

[0061] Finally, the final disease condition or insect damage is calculated based on the previously mentioned work results, and the overall health status is estimated based on the specific location, number and other details given by the model, so that garden managers can understand the health status of crop groups and implement precise prevention and control strategies to avoid large-scale outbreaks. For example, if a specific color change is found on the surface of the rice plant, it may be infected by fungi and it is necessary to intervene and spray medicine as soon as possible to prevent it from spreading and damaging the entire planting area. Finally, the program can generate a diagnostic report and clearly indicate which plots have problems and what extent they need to be dealt with.

[0062] First, the data obtained by the front-end acquisition equipment is analyzed and identified using technologies such as AI, and converted into a set standard information encoding mode, and encapsulated into a series of data structure units that can be easily transmitted over the network, such as using the common JSON format or other formatted data to meet communication requirements. For example, for a picture of a specific garden crop with traces of pests and diseases, a deep learning algorithm is used to process the image and obtain the percentage of the probability of infection; at this time, the system will convert and package the obtained information into a set of data packets that are easy to transmit in accordance with specific communication specifications and requirements for the next step.

[0063] Secondly, based on the prepared file containing the content to be sent, use the already well-established wireless communication facilities, such as selecting suitable long-distance broadband or low-power wide area network to implement the transmission operation; for example, when the above packaging is completed, it will use the existing communication infrastructure in the garden to transmit this data packet into the air through a pre-set transmission plan, thereby achieving the purpose of connecting with the remote system.

[0064] After moving through space on a physical level, it finally arrives at its destination, a location with processing power and resources for shared access, which usually refers to a data center room with large-scale computing and storage capabilities. For example, after the package is received remotely, the cloud server will unpack and interpret it, and restore it to the original form of text descriptions and quantitative indicators.

[0065] Furthermore, after analyzing the specific instructions and numbers received, they will be sorted and summarized in a more direct and easy-to-perceive form to the relevant operation objects; for example, after the analysis is completed in the cloud, these results can be displayed to garden managers or garden industry experts through various client applications, so that they can understand the current specific health status of the garden and the degree of damage at a glance.

[0066] Finally, if the actual measured value is found to be beyond the expected range according to the pre-determined risk control benchmark, a warning message needs to be quickly fed back to initiate the corresponding plan action; for example, if it is detected that the current spread of pathogens has reached a certain dangerous level threshold, an emergency notification mechanism should be immediately adopted to inform relevant personnel to arrange for processing as soon as possible to prevent irreversible damage.

[0067] Use labeled image datasets for model training; this process usually refers to using a large number of images with correct labels (pest and disease types) to train deep learning models such as convolutional neural networks. During the training process, the model learns the visual features of different types of pests and diseases so that they can be effectively identified.

[0068] The acquired garden environment images are input into the trained model; once the model is correctly trained and saved through the above steps, the newly captured or historically stored information containing pests and diseases is transmitted from the garden monitoring equipment during use, and sent to the trained network through the input layer for recognition and processing.

[0069] The probability of each category is calculated based on the recognition results output by the model; this step is the model prediction part, which generates a numerical assessment of the possibility of all known types of pests and diseases after receiving the input data, and displays the probability of each potential threat existing in the current input image in the form of a numerical value.

[0070] The recognition result is determined when the probability of a certain category exceeds the set baseline. Finally, the preset probability threshold is used to determine which species (or several most likely species) have reached a level of probability sufficient to be considered to exist in the sample, and to conclusively determine which category of pests or disease activity exists in the image. For example, for a deep learning model that has been trained and has a certain classification and recognition accuracy, after a picture taken in an actual monitoring environment is processed through the above three steps, the model gives a probability score for various pests and diseases. Assuming that the probability of rice leafhoppers is 63% and the baseline is 55%, it is confirmed that rice leafhoppers were found in this detection. This not only enables the system to have fast and efficient diagnostic capabilities, but also greatly improves accuracy and reliability.

[0071] The specific calculation of the probability of each category based on the recognition results output by the model is as follows: The first step is to set the probability of each category to zero. This setting is to assign an initial probability value to each category before starting the classification task. Here, the initial value is set to zero to ensure that there is no bias in the processing process due to the prior probability that has not been corrected for the data.

[0072] Then, according to the model prediction output, the corresponding probability is added to the prediction value of each category. In this stage, for each data sample to be identified, after the model gives its probability of classifying each category, this probability value is added to the corresponding category probability in order to have a statistical representation of the overall prediction effect of each category.

[0073] The process of calculating the probability score p of the predicted value and the sum of each category is a process of comparing the sum of probabilities accumulated in each step with the sum of predicted values ​​and probabilities of all categories. In this process, a credibility measure p is obtained for whether the category identification is of this type. And when the score p of a certain category is higher than the set probability threshold θ, the predicted value of the category is confirmed as the final pest identification result. This comparison process compares the p value of the current category with the preset threshold θ to determine whether the model prediction result can meet a confidence standard for determining the classification. When p exceeds the θ value, the sample is considered to belong to the category. For example, suppose that in an example of garden pest image recognition, there are categories A to E, representing different crop pest categories. Initially, all these probabilities are zero. After model analysis, the predicted probability of a new input image for each category is obtained. Then, the probability score obtained by each category based on this model prediction is updated. Continue to analyze other samples and update the total predicted probability of each category before starting statistics to obtain the probability score. If the probability score of a certain category C finally reaches 75% and the trust limit or threshold we set is 70%, then the final classification result is determined to be disease type C.

[0074] The present invention is based on the fact that when a certain category score p is higher than a set probability threshold θ, then confirming that the predicted value of the category is the final pest identification result further includes: determining that the category score p is equal to its classification probability; setting the relationship between the classification probability and the predefined threshold, θ represents the set threshold; and p>θ, then high-confidence identification is satisfied; marking the result of the pest identification, and associating the corresponding category label T; uploading the label to the server as a remote warning based on the high-confidence identification of the category T. Specific steps. The first step is to determine that the category score p is equal to its classification probability, which is the probability result of the occurrence of different types of pests and diseases based on the analysis of garden crop images or other input data by machine learning algorithms, where the category score represents the probability value of each type given by the model. The second step involves selecting those scores among these known category scores, that is, the probability exceeds the set threshold. Assuming that we set the threshold θ to 85%, when the probability of any pest and disease being detected exceeds this percentage, the system can confirm that the probability of this type of pest and disease being identified is very high, and confirm that the category is one of the results of the identification.

[0075] Then, after the specific pests and diseases are identified, a label T is added to it based on the discovery for subsequent processing and retrieval. This label includes information that helps with analysis, such as the identification date and geographic location, and is associated with a specific species for later tracking and recording. Finally, the system uploads the identified and classified pest and disease results to the cloud database as part of the remote early warning of pests and diseases. For example, a garden uses an intelligent system to monitor the health status of plants. If the model calculates that the probability of color aphids exceeds the set threshold of 85% based on the collected data, it is considered a reliable result; at this time, the result will be marked as a color aphid infection event, and detailed information, such as the location of the occurrence and garden information, will be sent to the remote monitoring platform to issue a warning. In this way, effective pest and disease control can be implemented through early warning to ensure smooth production in the garden industry.

[0076] The high-confidence identification based on the category T of the present invention uploads the mark to the server as a remote warning, specifically including: verifying whether the marked data packet has integrity M (if M is established, it is intact, otherwise it is discarded). This process means that when the system captures marked data about a certain type of specific pests and diseases, the captured data is comprehensively checked to ensure its integrity, that is, there is no information missing; the check here will be determined based on a predetermined information integrity threshold δ. For example, when more than δ proportion of information is lost or damaged, the entire data packet will be deemed unqualified and discarded and not uploaded.

[0077] Next is to transmit valid data to the designated server address S based on the network communication protocol. In this stage, by adopting standardized network communication frameworks and protocols such as TCP / IP or UDP, it is ensured that valid information can be smoothly sent to the expected remote monitoring system center. For example, once the marked packet is found to be intact, it will be immediately transmitted to the pest monitoring cloud service platform according to the standard protocol of the IP network to wait for its subsequent processing and generate the corresponding alarm strategy.

[0078] Then, before receiving the confirmation signal C, the current pest identification message is retained for backup; this step means that the system needs to always save the relevant detection data before successfully receiving a clear instruction from the receiving end to prevent failure to transmit correctly in the first attempt; only after the server successfully processes the information and responds to the correct response can the hardware space storing the relevant alarm information be released to record subsequent events. Assume that in an actual case, if the first transmission does not receive a response signal, the pest identification information needs to be re-stored and wait for the second transmission until a response is received.

[0079] Finally, it is displayed on the user interface for decision support and implementation of emergency measures. This process means that after completing all the above steps, the result display link is completed. At this time, the user terminal interface can present clear and reliable pest monitoring warning results, and assist users to take corresponding prevention and control actions in time according to the specific information obtained to ensure that the production of garden crops is not damaged. If the above links are successfully implemented, the garden contractor can observe the dynamics of possible pests in the field in real time through the application of the smartphone and obtain treatment suggestions immediately, so as to take necessary control methods to protect the crops from damage.

[0080] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. Output the identification labels and confidence indicators processed by the model as the key basis for identifying pests and diseases.

2. A remote pest monitoring system according to claim 1, characterized in that: The communication warning module transmits the recognition result to the remote terminal, displays the recognition result on the remote terminal, and issues a warning, including: Encode the analysis and identification results in a specific format and convert them into data packets suitable for transmission; Call the pre-defined wireless communication module (such as LTE, LoRa) to initiate the upload command according to the configuration strategy; After being sent to the remote monitoring platform, the cloud server receives and unpacks the data to extract the raw data and identification analysis results; The processed results are presented to the user and the current status and threat level of garden crops are displayed in a visual way; When the set risk threshold is reached, an alarm notification is immediately sent to the designated device or application to prompt the administrator to take necessary actions.

3. A remote pest monitoring system according to claim 1, characterized in that: The analysis module analyzes the image information based on the pre-trained deep learning model to improve the accuracy of pest identification, specifically including: Use labeled image datasets for model training; Input the acquired garden environment image into the trained model; Calculate the probability of each category based on the recognition results output by the model; and The recognition result is determined when the probability of a certain category exceeds the set baseline.

4. A remote pest monitoring system according to claim 5, characterized in that: The calculation of the probability of each category based on the recognition result output by the model includes: Set the probability of each class to be initialized to zero; According to the model prediction output, add the corresponding probability to the predicted value of each category; and Calculate the probability score p of the predicted value and of each category; and when the score p of a certain category is higher than the set probability threshold θ, confirm that the predicted value of the category is the final pest identification result.

5. A remote pest monitoring system according to claim 6, characterized in that: The method further comprises: when a category score p is higher than a set probability threshold θ, confirming the predicted value of the category as the final pest identification result. Determine the class score p equal to its classification probability; Set the relationship between the classification probability and the predefined threshold, θ represents the set threshold; and p>θ, then high confidence recognition is satisfied; Marking the results of the pest identification and associating the corresponding category label T; Based on the high confidence identification of the category T, the tag is uploaded to the server as a remote warning.

6. A remote pest monitoring system according to claim 7, characterized in that: The uploading of the mark to the server as a remote warning based on the high-confidence identification of the category T specifically includes: Check whether the marked data packet has integrity M (if M is true, it is intact, otherwise it is discarded); if the marked data loss exceeds the proportion δ of the total information volume, then M is not true, Transmit valid data to the designated server address S based on the network communication protocol; Before receiving the confirmation signal C, the current pest identification message is retained for backup; and when the server responds to the confirmation signal C, the memory can be released. The results are finally displayed on the user interface for decision support and implementation of emergency measures.

Images