Pest monitoring method and system based on artificial intelligence

Abstract

The present invention discloses an artificial intelligence-based insect monitoring method and system, the method includes data collection, preprocessing, establishing an insect monitoring model and insect monitoring. The present invention belongs to the field of insect monitoring, specifically refers to an artificial intelligence-based insect monitoring method and system, the scheme introduces residual learning to extract multi-scale features, introduces normalization processing in the remote dependent attention module, uses similarity as weight, and reduces the impact of noise monitoring data through nonlinear processing, enhances the ability to reconstruct the characteristics of insect monitoring data; thereby improving the accuracy of insect monitoring; introduces contrast loss to enhance category discrimination, optimizes and updates examples through feedback of classification results, designs metric functions based on dynamic examples and weighting mechanisms, provides more accurate detection results for insect monitoring; maintains stability in an insect monitoring environment with dynamic data changes and noise interference.

Description

Technical Field

[0001] The present invention relates to the technical field of insect monitoring, and in particular to an insect monitoring method and system based on artificial intelligence. Background Art

[0002] Pest monitoring is a method that integrates sensor technology, data analysis and machine learning, aiming to efficiently monitor and predict pest population dynamics and provide scientific support for agricultural production and ecological protection. However, general pest monitoring methods have problems such as improper feature extraction of pest monitoring data, difficulty in dealing with implicit features, and excessive sensitivity to noise data, which leads to poor stability and insufficient accuracy of pest monitoring; general pest monitoring methods are unable to effectively handle changes in the distribution of monitoring data over time, lack of discrimination of samples, and lack of feedback optimization, which leads to poor pest monitoring results and misjudgment. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an artificial intelligence-based insect monitoring method and system. In view of the problems that general insect monitoring methods have improper feature extraction of insect monitoring data, are difficult to deal with implicit features, and are overly sensitive to noise data, resulting in poor stability and insufficient accuracy of insect monitoring, this scheme introduces residual learning to perform multi-scale feature extraction, introduces normalization processing in the remote dependency attention module, uses similarity as a weight, and reduces the influence of noise monitoring data through nonlinear processing, thereby enhancing the ability to reconstruct the features of insect monitoring data; thereby improving the accuracy of insect monitoring; in view of the fact that general insect monitoring methods cannot effectively handle the changes in the distribution of monitoring data over time, lack of discrimination of samples, and lack of feedback optimization, thereby resulting in poor insect monitoring effect and the problem of misjudgment, this scheme introduces contrast loss to enhance category discrimination, optimizes and updates examples through feedback of classification results, designs metric functions based on dynamic examples and weighting mechanisms, and provides more accurate detection results for insect monitoring; and maintains stability in an insect monitoring environment with dynamic data changes and noise interference.

[0004] The technical solution adopted by the present invention is as follows: The insect monitoring method based on artificial intelligence provided by the present invention comprises the following steps:

[0005] Step S1: data collection;

[0006] Step S2: preprocessing;

[0007] Step S3: Establishing an insect monitoring model;

[0008] Step S4: Insect monitoring.

[0009] Furthermore, in step S1, the data collection is to collect historical insect monitoring data; the historical insect monitoring data includes insect pest type data, time data, space data, environment data, behavior data, distribution data, chemical composition data, weather data and insect pest risk level; the insect pest risk level includes mild risk, moderate risk and severe risk; the insect pest risk level is used as a data label; and the data set is cleaned, converted and standardized; the processed time series data is divided into a test set and a training set.

[0010] Furthermore, in step S2, the preprocessing specifically includes the following steps:

[0011] Step S21: Feature extraction: For feature extraction of insect monitoring, considering the scale changes of different time, location and environmental factors, a multi-scale 1D convolutional network is designed to process the time series data obtained in step S1, and the information of different scales is extracted through the expansion convolution operation. The convolution operation is expressed as: ; And adopt the residual connection technology to pass the input directly to the subsequent layer to form a skip connection. The residual learning is expressed as: ; Where F is the input feature; d is the dilation rate; k is the convolution kernel; (s) and (t) are the input position and output position in the convolution operation respectively; X is the input data; L is the number of network layers, l is the layer index; F l (·) is the feature extraction of each layer; y is the residual learning result;

[0012] Step S22: Long-range dependency attention module: A long-range dependency attention module is introduced to capture long-range dependency relationships; for noise elimination and feature extraction of insect data, non-local operations are used, which can be expressed as: ; ; ; ; Among them, y i is the output of the i-th sample after being processed by the non-local attention mechanism; C(x) is the normalization factor; x i and x j are the inputs of the ith and jth data points respectively; f(·) is the similarity; T is the transposition operation; g(·) is the linear transformation used for re-encoding, specifically; is the softmax function; is the weight matrix; i is the sample index.

[0013] Furthermore, in step S3, the establishment of the insect monitoring model specifically includes the following steps:

[0014] Step S31: Example generation: Generate initial examples c for each category k , through the feature embedding function Map the data into a high-dimensional space; first generate initial examples c for each category k , through the feature embedding function Mapping the data into a high-dimensional space, the examples of each category are represented as: ; Use contrast loss to enhance the discrimination of different categories of insect data; For each category of samples, construct positive and negative sample pairs to maximize the similarity between samples of the same category and minimize the similarity between samples of different categories; Sample similarity learning loss function L ce It is expressed as: ; Examples generated based on sample similarity learning at time nodes , calculate the classification probability of the query sample, expressed as:

[0015] ; Among them, c k (·) is the sample at time node; t is the time node; α is the smoothing parameter; is the sample set of the k-th category at time t+1; is the sample size; is a binary label, which is equal to 1 when the i-th and j-th samples belong to the same category, and equal to 0 otherwise; d(·) is a metric function; Represents the query sample x q The probability of belonging to the kth category; λ is the influence weight; It is The examples generated by each category; m is the interval threshold;

[0016] Step S32: Feedback optimization: Generate query examples through the classification probability of the query samples, expressed as: ; ; Calculate the corrected classification probability based on the query example, expressed as: ; and adjust the model parameters through the interactive loss function and gradient descent algorithm; interactive loss function It is expressed as: ;in, is the query example; M is the number of samples in each category; α i is the query weight of the i-th sample; e i It is the feature representation obtained by the multi-layer perceptron MLP; is the probability that the query sample belongs to the kth category; is the corrected classification probability; K is the analogy number; S k is the sample set of the kth analogy;

[0017] Step S33: define a metric function; measure the distance between examples and propose a metric function, which is expressed as: ;in, is the mean;

[0018] Step S34: model establishment; the established insect monitoring model includes feature extraction, remote dependency attention module, example generation and feedback optimization; the time series data obtained in step S1 is processed by feature extraction and remote dependency attention module to obtain enhanced feature representation, and the model output is obtained after example generation and feedback optimization; when the loss of the insect monitoring model on the training set converges, the training of the insect monitoring model is completed; an evaluation threshold is set in advance, and when the prediction accuracy of the trained insect monitoring model on the test set is higher than the evaluation threshold, the insect monitoring model is established; otherwise, the initial parameters are adjusted based on the particle swarm search algorithm and re-trained.

[0019] Furthermore, in step S4, the insect monitoring is based on the established insect monitoring model, and the pest type data, time data, spatial data, environmental data, behavior data, distribution data, chemical composition data and weather data are input into the insect monitoring model after data cleaning, data conversion and standardization, and the pest risk level output by the model is used as the insect monitoring result.

[0020] The artificial intelligence-based insect monitoring system provided by the present invention comprises a data acquisition module, a preprocessing module, an insect monitoring model establishment module and an insect monitoring module;

[0021] The data acquisition module collects historical insect monitoring data and sends the data to the preprocessing module;

[0022] The preprocessing module introduces residual learning to extract multi-scale features, introduces normalization processing in the remote dependency attention module, uses similarity as a weight, and implements preprocessing based on nonlinear processing; and sends the data to the insect monitoring model establishment module;

[0023] The insect monitoring model establishment module introduces contrast loss to enhance category differentiation, optimizes and updates examples through feedback of classification results, and designs a metric function based on dynamic examples and weighting mechanism; thereby establishing an insect monitoring model; and sending data to the insect monitoring module;

[0024] The insect monitoring module realizes insect monitoring based on the established insect monitoring model.

[0025] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0026] (1) In view of the problems that general insect monitoring methods have inappropriate feature extraction of insect monitoring data, difficulty in dealing with implicit features, and excessive sensitivity to noise data, resulting in poor stability and insufficient accuracy of insect monitoring, this scheme introduces residual learning for multi-scale feature extraction, introduces normalization processing in the long-range dependency attention module, uses similarity as weight, and reduces the impact of noisy monitoring data through nonlinear processing, thereby enhancing the ability to reconstruct the features of insect monitoring data; thereby improving the accuracy of insect monitoring.

[0027] (2) In view of the fact that general insect monitoring methods cannot effectively handle the changes in the distribution of monitoring data over time, lack the ability to distinguish samples, and lack feedback optimization, which in turn leads to poor insect monitoring results and the problem of misjudgment, this scheme introduces contrast loss to enhance category discrimination, updates examples through feedback optimization of classification results, and designs metric functions based on dynamic examples and weighted mechanisms to provide more accurate detection results for insect monitoring; it also maintains stability in an insect monitoring environment with dynamic data changes and noise interference. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A schematic diagram of the process of the insect monitoring method based on artificial intelligence provided by the present invention;

[0029] Figure 2 A schematic diagram of an artificial intelligence-based insect monitoring system provided by the present invention;

[0030] Figure 3 It is a schematic diagram of the process of step S3.

[0031] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0032] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0033] In the description of the present invention, it is necessary to understand that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the referred system or element must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as a limitation on the present invention.

[0034] Example 1, see Figure 1 The present invention provides an artificial intelligence-based insect monitoring method, which comprises the following steps:

[0035] Step S1: data collection, collecting historical insect monitoring data;

[0036] Step S2: Preprocessing, introducing residual learning for multi-scale feature extraction, introducing normalization processing in the long-range dependency attention module, taking similarity as weight, and implementing preprocessing based on nonlinear processing;

[0037] Step S3: Establish an insect monitoring model, introduce contrast loss to enhance category discrimination, optimize and update examples through feedback of classification results, and design a metric function based on dynamic examples and weighting mechanism; thereby establishing an insect monitoring model;

[0038] Step S4: Insect monitoring: implementing insect monitoring based on the established insect monitoring model.

[0039] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the historical insect monitoring data includes pest type data, time data, spatial data, environmental data, behavioral data, distribution data, chemical composition data, weather data and pest risk level; the pest type data includes insect type, biological characteristics and life cycle; the time data includes monitoring time, high incidence period of pests, and pest activity time period; the spatial data includes geographical location, habitat type, regional climate conditions and crop type; the behavioral data includes flight activity, foraging behavior and reproductive behavior; the pest risk level includes mild risk, moderate risk and severe risk; the pest risk level is used as a data label; and the data set is cleaned, converted and standardized; the processed time series data is divided into a test set and a training set.

[0040] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S2, the preprocessing specifically includes the following steps:

[0041] Step S21: Feature extraction: For feature extraction of insect monitoring, considering the scale changes of different time, location and environmental factors, a multi-scale 1D convolutional network is designed to process the time series data obtained in step S1, and the information of different scales is extracted through the expansion convolution operation. The convolution operation is expressed as: ; And adopt the residual connection technology to pass the input directly to the subsequent layer to form a skip connection. The residual learning is expressed as: ; Where F is the input feature; d is the dilation rate; k is the convolution kernel; (s) and (t) are the input position and output position in the convolution operation respectively; X is the input data; L is the number of network layers, l is the layer index; F l (·) is the feature extraction of each layer; y is the residual learning result;

[0042] Step S22: Long-range dependency attention module: A long-range dependency attention module is introduced to capture long-range dependency relationships; for noise elimination and feature extraction of insect data, non-local operations are used, which can be expressed as: ; ; ; ; Among them, y i is the output of the i-th sample after being processed by the non-local attention mechanism; C(x) is the normalization factor; x i and x j are the inputs of the ith and jth data points respectively; f(·) is the similarity; T is the transposition operation; g(·) is the linear transformation used for re-encoding, specifically; is the softmax function; is the weight matrix; i is the sample index.

[0043] By performing the above operations, in view of the problems that general insect monitoring methods have improper feature extraction of insect monitoring data, difficulty in dealing with implicit features, and excessive sensitivity to noise data, resulting in poor stability and insufficient accuracy of insect monitoring, this scheme introduces residual learning for multi-scale feature extraction, introduces normalization processing in the long-range dependency attention module, uses similarity as weight, and reduces the impact of noise monitoring data through nonlinear processing, thereby enhancing the ability to reconstruct the characteristics of insect monitoring data; thereby improving the accuracy of insect monitoring.

[0044] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S3, establishing an insect monitoring model specifically includes the following steps:

[0045] Step S31: Example generation: Generate initial examples c for each category k , through the feature embedding function Map the data into a high-dimensional space; first generate initial examples c for each category k , through the feature embedding function Mapping the data into a high-dimensional space, the examples of each category are represented as: ; Use contrast loss to enhance the discrimination of different categories of insect data; For each category of samples, construct positive and negative sample pairs to maximize the similarity between samples of the same category and minimize the similarity between samples of different categories; Sample similarity learning loss function Lce It is expressed as: ; Examples generated based on sample similarity learning at time nodes , calculate the classification probability of the query sample, expressed as:

[0046] ; Among them, c k (·) is the sample at time node; t is the time node; α is the smoothing parameter; is the sample set of the k-th category at time t+1; is the sample size; is a binary label, which is equal to 1 when the i-th and j-th samples belong to the same category, and equal to 0 otherwise; d(·) is a metric function; Represents the query sample x q The probability of belonging to the kth category; λ is the influence weight; It is The examples generated by each category; m is the interval threshold;

[0047] Step S32: Feedback optimization: Generate query examples through the classification probability of the query samples, expressed as: ; ; Calculate the corrected classification probability based on the query example, expressed as: ; and adjust the model parameters through the interactive loss function and gradient descent algorithm; interactive loss function It is expressed as: ;in, is the query example; M is the number of samples in each category; α i is the query weight of the i-th sample; e i It is the feature representation obtained by the multi-layer perceptron MLP; is the probability that the query sample belongs to the kth category; is the corrected classification probability; K is the analogy number; S k is the sample set of the kth analogy;

[0048] Step S33: define a metric function; measure the distance between examples and propose a metric function, which is expressed as: ;in, is the mean;

[0049] Step S34: model establishment; the established insect monitoring model includes feature extraction, remote dependency attention module, example generation and feedback optimization; the time series data obtained in step S1 is processed by feature extraction and remote dependency attention module to obtain enhanced feature representation, and the model output is obtained after example generation and feedback optimization; when the loss of the insect monitoring model on the training set converges, the training of the insect monitoring model is completed; an evaluation threshold is set in advance, and when the prediction accuracy of the trained insect monitoring model on the test set is higher than the evaluation threshold, the insect monitoring model is established; otherwise, the initial parameters are adjusted based on the particle swarm search algorithm and re-trained.

[0050] By performing the above operations, the general insect monitoring methods are unable to effectively handle the changes in the distribution of monitoring data over time, the discrimination of samples is insufficient, and there is a lack of feedback optimization, which leads to poor insect monitoring effects and misjudgment. This scheme introduces contrast loss to enhance category discrimination, updates examples through feedback optimization of classification results, designs metric functions based on dynamic examples and weighted mechanisms, and provides more accurate detection results for insect monitoring; it maintains stability in an insect monitoring environment with dynamic data changes and noise interference.

[0051] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S4, insect monitoring is based on the established insect monitoring model. Pest type data, time data, space data, environmental data, behavior data, distribution data, chemical composition data and weather data are input into the insect monitoring model after data cleaning, data conversion and standardization. The pest risk level output by the model is used as the insect monitoring result.

[0052] Example 6, see Figure 2 , this embodiment is based on the above embodiment, and the insect monitoring system based on artificial intelligence provided by the present invention includes a data acquisition module, a preprocessing module, an insect monitoring model establishment module and an insect monitoring module;

[0053] The data acquisition module collects historical insect monitoring data and sends the data to the preprocessing module;

[0054] The preprocessing module introduces residual learning to extract multi-scale features, introduces normalization processing in the remote dependency attention module, uses similarity as a weight, and implements preprocessing based on nonlinear processing; and sends the data to the insect monitoring model establishment module;

[0055] The insect monitoring model establishment module introduces contrast loss to enhance category differentiation, optimizes and updates examples through feedback of classification results, and designs a metric function based on dynamic examples and weighting mechanism; thereby establishing an insect monitoring model; and sending data to the insect monitoring module;

[0056] The insect monitoring module realizes insect monitoring based on the established insect monitoring model.

[0057] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0058] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0059] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. The insect monitoring method based on artificial intelligence is characterized by: The method comprises the following steps: Step S1: data collection, collecting historical insect monitoring data; Step S2: Preprocessing, introducing residual learning for multi-scale feature extraction, introducing normalization processing in the long-range dependency attention module, taking similarity as weight, and implementing preprocessing based on nonlinear processing; Step S3: Establish an insect monitoring model, introduce contrast loss to enhance category discrimination, optimize and update examples through feedback of classification results, and design a metric function based on dynamic examples and weighting mechanism; thereby establishing an insect monitoring model; Step S4: Insect monitoring, implementing insect monitoring based on the established insect monitoring model; Step S3 includes step S33: defining a metric function; measuring the distance between examples, and proposing a metric function, which is expressed as: ;in, is the mean; d(·) is the metric function; is the feature embedding function; x q is a query sample; is a query example; is the sample set of the kth category; is the number of samples; i is the sample index; In step S1, the historical insect monitoring data includes insect pest type data, time data, space data, environment data, behavior data, distribution data, chemical composition data, weather data and insect pest risk level; the insect pest risk level includes mild risk, moderate risk and severe risk; the insect pest risk level is used as a data label; and the data set is cleaned, converted and standardized; the processed time series data is divided into a test set and a training set; In step S3, the establishment of the insect monitoring model further includes the following steps: Step S31: Example generation: First, generate the initial examples c of each category k , through the feature embedding function Mapping the data into a high-dimensional space, the examples of each category are represented as: ; Use contrast loss to enhance the discrimination of different categories of insect data; For each category of samples, construct positive and negative sample pairs to maximize the similarity between samples of the same category and minimize the similarity between samples of different categories; Sample similarity learning loss function L ce It is expressed as: ; Examples generated based on sample similarity learning at time nodes , calculate the classification probability of the query sample, expressed as: ; Among them, c k (·) is the sample at time node; t is the time node; α is the smoothing parameter; is the sample set of the k-th category at time t+1; is the sample size; is a binary label, which is equal to 1 when the i-th and j-th samples belong to the same category, and equal to 0 otherwise; d(·) is a metric function; Represents the query sample x q The probability of belonging to the kth category; λ is the influence weight; It is The generated examples of categories; m is the interval threshold; x i and x j are the inputs of the i-th and j-th data points respectively; Step S32: Feedback optimization: Generate query examples through the classification probability of query samples, expressed as: ; ; Calculate the corrected classification probability based on the query example, expressed as: ; and adjust the model parameters through the interactive loss function and gradient descent algorithm; interactive loss function It is expressed as: ;in, is the query example; M is the number of samples in each category; α i is the query weight of the i-th sample; e i It is the feature representation obtained by the multi-layer perceptron MLP; is the probability that the query sample belongs to the kth category; is the corrected classification probability; K is the number of categories; S k is the sample set of the kth category; Step S34: model establishment; the established insect monitoring model includes feature extraction, remote dependency attention module, example generation and feedback optimization; the time series data obtained in step S1 is processed by feature extraction and remote dependency attention module to obtain enhanced feature representation, and the model output is obtained after example generation and feedback optimization; when the loss of the insect monitoring model on the training set converges, the training of the insect monitoring model is completed; an evaluation threshold is set in advance, and when the prediction accuracy of the trained insect monitoring model on the test set is higher than the evaluation threshold, the insect monitoring model is established; otherwise, the initial parameters are adjusted based on the particle swarm search algorithm and re-trained.

2. The insect monitoring method based on artificial intelligence according to claim 1, characterized in that: In step S2, the preprocessing specifically includes: Step S21: Feature extraction: For feature extraction of insect monitoring, considering the scale changes of different time, location and environmental factors, a multi-scale 1D convolutional network is designed to process the time series data obtained in step S1, and the information of different scales is extracted through the expansion convolution operation. The convolution operation is expressed as: ; And adopt the residual connection technology to pass the input directly to the subsequent layer to form a skip connection. The residual learning is expressed as: ; Where F is the input feature; d is the dilation rate; k is the convolution kernel; (s) and (t) are the input position and output position in the convolution operation respectively; X is the input data; L is the number of network layers, l is the layer index; F l (·) is the feature extraction of each layer; y is the residual learning result; Step S22: Long-range dependency attention module: A long-range dependency attention module is introduced to capture long-range dependency relationships; for noise elimination and feature extraction of insect data, non-local operations are used, which can be expressed as: ; ; ; ; Among them, y i is the output of the i-th sample after being processed by the non-local attention mechanism; C(x) is the normalization factor; x i and x j are the inputs of the ith and jth data points respectively; f(·) is the similarity; T is the transposition operation; g(·) is the linear transformation used for re-encoding, specifically; is the softmax function; is the weight matrix; i is the sample index.

3. The insect monitoring method based on artificial intelligence according to claim 1, characterized in that: In step S4, the insect monitoring is based on the established insect monitoring model. The pest type data, time data, spatial data, environmental data, behavior data, distribution data, chemical composition data and weather data are input into the insect monitoring model after data cleaning, data conversion and standardization. The pest risk level output by the model is used as the insect monitoring result.

4. An artificial intelligence-based insect monitoring system, used to implement the artificial intelligence-based insect monitoring method as described in any one of claims 1 to 3, characterized in that: It includes data collection module, preprocessing module, insect monitoring model building module and insect monitoring module; The data acquisition module collects historical insect monitoring data and sends the data to the preprocessing module; The preprocessing module introduces residual learning to extract multi-scale features, introduces normalization processing in the remote dependency attention module, uses similarity as a weight, and implements preprocessing based on nonlinear processing; and sends the data to the insect monitoring model establishment module; The insect monitoring model establishment module introduces contrast loss to enhance category differentiation, optimizes and updates examples through feedback of classification results, and designs a metric function based on dynamic examples and weighting mechanism; thereby establishing an insect monitoring model; and sending data to the insect monitoring module; The insect monitoring module realizes insect monitoring based on the established insect monitoring model.

Images