Forest pest and disease development trend prediction method and system combined with remote sensing monitoring

Abstract

The invention discloses a forest pest and disease development trend prediction method and system combined with remote sensing monitoring, and relates to the related field of forestry monitoring data processing, and the method comprises the steps: carrying out the remote sensing monitoring of a target forest, building a remote sensing image data set, calling ground monitoring data, and carrying out the rasterization based on a time step after the interpolation; performing multi-feature joint extraction on the remote sensing image data set, and superposing extraction results to a unified grid to form a multi-layer feature map; constructing a joint feature matrix by using the rasterized data and the multi-layer feature map, and performing pest distribution identification through a classification model; performing pest development trend prediction by taking the pest distribution diagram as a basic feature and taking the combined feature matrix as an additional feature; and according to a development trend prediction result, carrying out pest abnormity early warning. The technical problem of insufficient prediction accuracy caused by insufficient monitoring efficiency and accuracy in existing forest pest development trend prediction is solved, and the technical effects of improving the monitoring efficiency and accuracy and improving the pest prediction accuracy are achieved.

Description

Technical Field

[0001] This application relates to the field of forestry monitoring data processing, and particularly to a method and system for predicting the development trend of forest pests and diseases combined with remote sensing monitoring. Background Art

[0002] With the continuous changes in global climate and ecological environment, the problem of forest pests and diseases has become increasingly severe, posing a huge threat to forest resources. Effective monitoring and prediction of forest pests and diseases are of great significance for protecting forest resources and maintaining ecological balance. Traditionally, the monitoring of forest pests and diseases mainly relies on manual ground surveys. Through manual field surveys, relevant information on forest pests and diseases is collected, such as the number of damaged trees and the degree of damage. This method is not only time-consuming and laborious but also difficult to achieve large-area and high-precision monitoring. At the same time, most of the existing remote sensing monitoring methods focus on the extraction of single features and are difficult to accurately evaluate the severity and development trend of pests and diseases.

[0003] In the current related technologies, there are technical problems in the prediction of the development trend of forest pests and diseases, such as insufficient monitoring efficiency and accuracy, resulting in insufficient prediction accuracy. Summary of the Invention

[0004] This application provides a method and system for predicting the development trend of forest pests and diseases combined with remote sensing monitoring. By remotely sensing the target forest, a remote sensing image data set is established, and ground monitoring data is collected synchronously. The data is interpolated, and based on the time-step rasterized interpolated ground monitoring data of the remote sensing image data set, multi-feature joint extraction is performed on the remote sensing image data set. The extraction results are superimposed on a unified raster to form a multi-layer feature map. A joint feature matrix is constructed with rasterized data and the multi-layer feature map. The distribution of pests is identified through a classification model, and a pest distribution map with a grade evaluation label is generated. Based on the pest distribution map as the basic feature and the joint feature matrix as the additional feature, a prediction of the pest development trend based on time series and spatial diffusion is performed to establish a prediction result. According to the prediction result of the development trend, technical means such as early warning of pest anomalies are carried out, achieving the technical effect of improving monitoring efficiency and accuracy, and further improving the accuracy of pest prediction.

[0005] The present application provides a method for predicting the development trend of forest pests and diseases in combination with remote sensing monitoring, including: conducting remote sensing monitoring on the target forest, establishing a remote sensing image data set, and synchronously invoking ground monitoring data, where the ground monitoring data includes temperature data, humidity data, rainfall data, and vegetation species distribution data. After interpolating the ground monitoring data, rasterize the interpolated ground monitoring data based on the time step of the remote sensing image data set; after jointly extracting multiple features from the remote sensing image data set, superimpose the results of the joint extraction of multiple features onto a unified grid to form a multi-layer feature map; construct a joint feature matrix with the rasterized data and the multi-layer feature map, and identify the pest distribution based on the joint feature matrix through a classification model to generate a pest distribution map, where the pest distribution map is marked with a grade evaluation identifier; use the pest distribution map as the basic feature and the joint feature matrix as the additional feature, perform pest development trend prediction based on time series and spatial diffusion, establish a development trend prediction result; and issue an early warning for pest anomalies according to the development trend prediction result.

[0006] In a possible implementation, after jointly extracting multiple features from the remote sensing image data set, superimposing the results of the joint extraction of multiple features onto a unified grid to form a multi-layer feature map, the following processing is performed: After preprocessing the remote sensing image data set, perform image feature extraction at multiple granularities to establish a multi-granularity feature extraction result; establish a feature granularity distribution based on the multi-granularity feature extraction result, where the feature granularity distribution includes a leaf damage feature granularity, a large-area pest and disease distribution feature granularity, a temperature anomaly feature granularity, and a spatial diffusion feature granularity; optimize the image extraction scale according to the feature granularity distribution, and configure the attention scale according to the image extraction scale optimization result; perform scale feature extraction of corresponding features based on the attention scale to establish a joint extraction result of multiple features.

[0007] In a possible implementation, for performing scale feature extraction of corresponding features based on the attention scale to establish a joint extraction result of multiple features, the following processing is performed: Input the attention scale and corresponding features into a scale joint extraction network to determine the main extraction scale and the auxiliary scale; perform scale feature extraction of corresponding features according to the main scale and the auxiliary scale, and perform joint analysis of the scale feature extraction results to establish a scale joint extraction result; configure a feature coupling model, and use the feature coupling model to perform cross-feature dynamic joint analysis of the scale joint extraction result to establish a joint extraction result of multiple features.

[0008] In a possible implementation, for pest distribution recognition based on the joint feature matrix by means of a classification model to generate a pest distribution map, the following processing is performed: perform correlation analysis of pest generation on the joint feature matrix to establish an identification correlation coefficient; obtain a general pest classification model, perform feature migration of the general pest classification model through the identification correlation coefficient, and configure the classification model; use the joint feature matrix as input data and input it into the classification model to generate a rasterized pest distribution map.

[0009] In a possible implementation, for the correlation analysis of pest generation on the joint feature matrix to establish an identification correlation coefficient, the following processing is performed: calculate the correlation coefficient of pest generation through a formula as follows: ; where represents the identification correlation coefficient, characterizes the summary value of single-feature correlation, and single-feature correlation where characterizes the single-feature correlation strength of the th feature, characterizes the feature, is the pest label, , are respectively the and standard deviations of characterizes the summary value of multi-feature joint correlation, and multi-feature joint correlation where characterizes the weight of the feature pair, characterizes the feature , and the pest label mutual information, , are respectively the information entropies of the features , , characterizes the feature interaction correlation value, where characterizes the total number of features, , are respectively the feature indices, and , characterizes the feature interaction weight, characterizes the feature interaction function, is the sample index, ranging from 1 to , characterizes the dynamic change correlation value, where characterizes the time change amount of the feature , is a regulation coefficient, characterizes the feature of the time variation amount, , , , are respectively the weight coefficients of single feature, multi - feature, feature interaction and dynamic change.

[0010] In a possible implementation manner, taking the pest distribution map as the basic feature and the joint feature matrix as the additional feature, perform pest development trend prediction based on time series and spatial diffusion, and perform the following processing: extract the temporal pest distribution change of the pest distribution map, perform time series prediction of the diffusion and time - dependence relationship according to the extraction result of the temporal pest distribution change, and establish the diffusion prediction result at each moment; take the pest distribution map and the joint feature matrix as input features, perform local diffusion detail fitting, and establish the local spatial diffusion fitting result; perform spatio - temporal development trend prediction on the diffusion prediction result at each moment and the local spatial diffusion fitting result, and establish the development trend prediction result.

[0011] In a possible implementation manner, for pest anomaly early warning according to the development trend prediction result, perform the following processing: perform trend anomaly trigger analysis based on the development trend prediction result to establish the anomaly level, and the trend anomaly trigger analysis includes trend development speed analysis and trend development scale analysis; perform pest anomaly early warning according to the anomaly level.

[0012] This application also provides a forest pest development trend prediction system combined with remote sensing monitoring, including: a ground monitoring data rasterization module, which is used to perform remote sensing monitoring on the target forest, establish a remote sensing image data set, and synchronously call ground monitoring data, where the ground monitoring data includes temperature data, humidity data, rainfall data, and vegetation type distribution data, and after interpolating the ground monitoring data, rasterize the interpolated ground monitoring data based on the time step of the remote sensing image data set; a multi - layer feature map formation module, which is used to perform multi - feature joint extraction on the remote sensing image data set, and then stack the multi - feature joint extraction results onto a unified grid to form a multi - layer feature map; a pest distribution recognition module, which is used to construct a joint feature matrix with rasterized data and the multi - layer feature map, and perform pest distribution recognition based on the joint feature matrix through a classification model to generate a pest distribution map, and the pest distribution map is marked with a grade evaluation; a pest development trend prediction module, which is used to take the pest distribution map as the basic feature and the joint feature matrix as the additional feature, perform pest development trend prediction based on time series and spatial diffusion, and establish a development trend prediction result; a pest anomaly early warning module, which is used to perform pest anomaly early warning according to the development trend prediction result.

[0013] The method and system for predicting the development trend of forest pests and diseases combined with remote sensing monitoring proposed in this application first conduct remote sensing monitoring on the target forest, establish a remote sensing image data set, and simultaneously call ground monitoring data. The ground monitoring data includes temperature data, humidity data, rainfall data, and vegetation type distribution data. After interpolating the ground monitoring data, based on the time step rasterization of the remote sensing image data set, the interpolated ground monitoring data is then subjected to multi-feature joint extraction on the remote sensing image data set. After that, the multi-feature joint extraction results are superimposed onto a unified grid to form a multi-layer feature map. Then, a joint feature matrix is constructed with the rasterized data and the multi-layer feature map, and a pest distribution recognition based on the joint feature matrix is performed through a classification model to generate a pest distribution map. The pest distribution map has a grade evaluation identifier. Then, taking the pest distribution map as the basic feature and the joint feature matrix as the additional feature, a prediction of the pest development trend based on time series and spatial diffusion is performed to establish a development trend prediction result. Finally, pest anomaly warnings are issued based on the development trend prediction result, achieving the technical effects of improving the monitoring efficiency and accuracy, and thus improving the accuracy of pest and disease prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of this application. It should be understood that the operations before or below do not necessarily need to be performed precisely in sequence. On the contrary, as needed, various steps can be performed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or several steps can be removed from these processes.

[0015] Figure 1 It is a schematic flowchart of the method for predicting the development trend of forest pests and diseases combined with remote sensing monitoring provided by the embodiments of this application.

[0016] Figure 2 It is a schematic structural diagram of the system for predicting the development trend of forest pests and diseases combined with remote sensing monitoring provided by the embodiments of this application.

[0017] Description of the reference numerals: Ground monitoring data rasterization module 10, multi-layer feature map formation module 20, pest distribution recognition module 30, pest development trend prediction module 40, pest anomaly warning module 50. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] The above description is only an overview of the technical solutions of this application. In order to be able to understand the technical means of this application more clearly, it can be implemented according to the content of the description. And in order to make the above and other purposes, features, and advantages of this application more obvious and understandable, the following specifically illustrates the specific embodiments of this application.

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following will further describe this application in detail with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of this application.

[0020] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments. However, it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments and can be combined with each other without conflict. The terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application.

[0021] The embodiments of this application provide a method for predicting the development trend of forest pests and diseases in combination with remote sensing monitoring, as Figure 1 shown. The method includes: Step S100, conduct remote sensing monitoring on the target forest, establish a remote sensing image data set, and synchronously call ground monitoring data. The ground monitoring data includes temperature data, humidity data, rainfall data, and vegetation species distribution data. After interpolating the ground monitoring data, the ground monitoring data after rasterized interpolation based on the time step of the remote sensing image data set.

[0022] Specifically, a remote sensing platform such as a satellite or a drone is used to perform regular or continuous image acquisition on the target forest to form a remote sensing image dataset. These image datasets contain spectral information in different bands and can reflect the growth status, health status, etc. of forest vegetation. Meteorological data such as temperature, humidity, and rainfall of the target forest are collected synchronously, as well as vegetation species distribution data. These data can be obtained through means such as ground meteorological stations, soil moisture meters, and vegetation surveys. Since the ground monitoring data are collected at a limited number of observation points, spatial interpolation methods (such as Kriging interpolation, inverse distance weighted interpolation, etc.) are needed to extend these discrete data to the entire study area to form a continuous distribution map. The interpolated ground monitoring data are matched with the time steps of the remote sensing image dataset, and then all the data are rasterized. Rasterization is to divide the continuous spatial data into a series of regular grid cells (i.e., rasters), each grid cell representing a certain geographical range and being assigned corresponding attribute values. In this way, the remote sensing images and the ground monitoring data can be compared and analyzed on the same spatial and temporal scales.

[0023] Step S200, after jointly extracting multiple features from the remote sensing image dataset, the joint multiple feature extraction results are superimposed on a unified grid to form a multi-layer feature map.

[0024] Specifically, multiple features are extracted from the remote sensing image dataset, such as vegetation indices (values calculated using the spectral reflectance of different bands in the remote sensing image, used to reflect the growth status, health status, etc. of vegetation), texture features (information describing local patterns or structures in the image, such as roughness, directionality, etc.), spectral features, etc. These features can reflect the growth status, health status, water status, etc. of vegetation. The multiple extracted features are superimposed on a unified grid to form a multi-layer feature map. Each layer of the feature map represents a specific type of information or attribute, such as a vegetation index layer, a texture feature layer, etc.

[0025] In a possible implementation, for the multi-feature joint extraction of the remote sensing image dataset, step S200 further includes step S210. After preprocessing the remote sensing image dataset, image feature extraction at multiple granularities is performed to establish multi-granularity feature extraction results. Specifically, preprocessing the remote sensing image dataset includes steps such as denoising, enhancing contrast, and correction to improve the image quality and lay a foundation for feature extraction. On the preprocessed remote sensing image, feature extraction is performed using windows or filters of different scales. These scales can cover different ranges from the leaf level to the entire forest area to capture pest and disease characteristics at different granularities. Specifically, smaller windows or filters are used to focus on leaf-level details and extract features related to leaf damage, such as color changes and texture abnormalities; larger windows or filters are used to cover a larger geographical range and extract features related to the distribution of pests and diseases in the entire target forest, such as changes in vegetation indices and abnormalities in color distribution; by combining temperature data in the ground monitoring data and using the thermal infrared band in the remote sensing image, features related to temperature anomalies are extracted, and these anomalies may indicate the activities or impacts of pests and diseases; the diffusion patterns of pests and diseases in space are analyzed, and features related to the diffusion speed and direction are extracted to predict the future development trends of pests and diseases.

[0026] Step S220, establish a feature granularity distribution according to the multi-granularity feature extraction results. The feature granularity distribution includes leaf damage feature granularity, large-area pest and disease distribution feature granularity, temperature anomaly feature granularity, and spatial diffusion feature granularity. Specifically, according to the multi-granularity features extracted in step S210, a feature granularity distribution is established. The feature granularity distribution is a model or representation that describes the distribution of different granularity features in the remote sensing image.

[0027] Step S230, optimize the image extraction scale according to the feature granularity distribution, and configure the focus scale according to the image extraction scale optimization result. Specifically, machine learning algorithms (such as support vector machines, random forests, etc.) are used to evaluate the scale of the feature granularity distribution established in step S220 to determine which scale features are the most important for pest distribution recognition. According to the scale evaluation results, the optimal image extraction scales are selected, that is, those scales that can best reflect the pest distribution characteristics. These scales are used as the focus scales for subsequent feature extraction.

[0028] Step S240: Based on the concerned scale, perform scale feature extraction of corresponding features and establish a multi-feature joint extraction result. Specifically, at the concerned scale determined in step S230, perform feature extraction on the remote sensing image, that is, at the concerned scale, use image processing algorithms (such as edge detection, texture analysis, color histogram, etc.) to extract features related to the pest distribution. Combine the features extracted at different concerned scales to form a feature set containing various granularity information for pest distribution identification and development trend prediction. This implementation method can determine which scale features are the most important for pest distribution identification by establishing a feature granularity distribution and performing scale optimization, thereby optimizing the feature extraction process and improving the identification accuracy.

[0029] In a possible implementation manner, for the step of performing scale feature extraction of corresponding features based on the concerned scale and establishing a multi-feature joint extraction result, step S240 further includes step S241: input the concerned scale and corresponding features into a scale joint extraction network to determine the main scale and auxiliary scales. Specifically, input the concerned scale determined in step S230 and its corresponding features (such as leaf damage features, large-area pest and disease distribution features, etc.) into the scale joint extraction network. The scale joint extraction network is a deep learning model for processing multi-scale feature extraction tasks. The scale joint extraction network processes the input features through internal mechanisms (such as convolutional layers, pooling layers, etc.) and automatically determines a main scale (that is, the scale that can most effectively represent the feature scale) and several auxiliary scales (that is, scales that provide additional information or context support) according to the significance and importance of the features. This process can be achieved through attention mechanisms, multi-scale fusion strategies, or feature importance evaluation algorithms.

[0030] Step S242: According to the main scale and the auxiliary scales, perform scale feature extraction of corresponding features and execute joint analysis of the scale feature extraction results to establish a scale joint extraction result. Specifically, according to the main scale and auxiliary scales determined in step S241, perform scale feature extraction on the input features, including performing operations such as convolution and pooling on the image at different scales to extract multi-scale features related to pests and diseases. Through methods such as feature fusion, feature stitching, and attention mechanisms, perform joint analysis on the extracted multi-scale features to capture the correlation and complementarity between different scale features and generate a feature representation that integrates multi-scale information.

[0031] Step S243: Configure a feature coupling model and use the feature coupling model to perform cross-feature dynamic joint analysis on the scale joint extraction results to establish a multi-feature joint extraction result. Specifically, configure a feature coupling model, which is a deep learning network for processing cross-feature dynamic joint analysis tasks. It can fuse and comprehensively analyze features from different scales and different feature types to generate a more comprehensive and accurate feature representation. Use the feature coupling model to perform cross-feature dynamic joint analysis on the scale joint extraction results generated in step S242, including steps such as feature fusion, feature transformation, and attention mechanism, to capture the interactions and dependencies between different features. Finally, generate a multi-feature joint extraction result containing multi-scale and multi-feature information. This implementation method effectively utilizes multi-scale information by automatically determining the main scale and the auxiliary scale, performing feature extraction and joint analysis on them, improving the accuracy and efficiency of feature extraction. Through cross-feature dynamic joint analysis, features from different scales and different feature types are fused and comprehensively analyzed to generate a more comprehensive and accurate feature representation, thereby improving the accuracy of pest distribution recognition and development trend prediction.

[0032] Step S300: Construct a joint feature matrix with rasterized data and the multi-layer feature map, and perform pest distribution recognition based on the joint feature matrix through a classification model to generate a pest distribution map with a grade evaluation identifier.

[0033] Specifically, combine the rasterized data and the multi-layer feature map to form a joint feature matrix containing various information. Each element of this matrix corresponds to a raster cell and contains all the feature values of that cell. Use a machine learning or deep learning classification model (such as random forest, support vector machine, convolutional neural network, etc.) to classify the joint feature matrix to identify the pest distribution. The classification model will distinguish the pest occurrence area and the non-pest area according to different combinations of feature values. According to the classification results, generate a pest distribution map, that is, a map representing the geographical distribution of pests. In the map, different areas are represented by different colors or symbols for different pest grades or categories. At the same time, add a grade evaluation identifier to each area to reflect the severity of the pests.

[0034] In a possible implementation, the pest distribution identification based on the joint feature matrix by the classification model to generate a pest distribution map, and step S300 further includes step S310 of performing correlation analysis of pest generation on the joint feature matrix to establish an identification correlation coefficient. Specifically, through statistical methods such as correlation analysis, regression analysis, or feature importance evaluation in machine learning, the correlation between each feature in the joint feature matrix (such as the leaf damage condition in remote sensing images, temperature and humidity in ground monitoring data, etc.) and pest generation is analyzed. Based on the results of the correlation analysis, a correlation coefficient is assigned to each feature, and this coefficient is a numerical index measuring the degree of association between the feature set and pest generation.

[0035] Step S320, obtain a general pest classification model, and perform feature migration of the general pest classification model through the identification correlation coefficient to configure the classification model. Specifically, the general pest classification model is a model pre-trained based on a machine learning framework, which can classify pests based on a certain feature set. Since the general pest classification model is not trained based on the current joint feature matrix, it is necessary to adjust the feature input of the model according to the identification correlation coefficient, including selecting the most important features, weighting or transforming the features, etc. After feature migration, a classification model optimized for the current joint feature matrix is obtained.

[0036] Step S330, take the joint feature matrix as input data and input it into the classification model to generate a rasterized pest distribution map. Specifically, the joint feature matrix is used as input data and passed to the configured classification model. The model calculates the probability or score of each grid cell belonging to different pest categories based on the input data. According to the output of the classification model, a pest category and a corresponding grade evaluation identifier (such as pest severity) are assigned to each grid cell, thereby generating a rasterized pest distribution map. This implementation determines the most important features through correlation analysis and uses these features to optimize the classification model, improving the accuracy of pest distribution identification.

[0037] In a possible implementation, the step of performing correlation analysis of pest generation on the joint feature matrix to establish an identification correlation coefficient, step S310 further includes step S311 of calculating the correlation coefficient of pest generation through a formula as follows: ; where represents the identification correlation coefficient, characterizes the summary value of single-feature correlation, and the single-feature correlation , where characterizes the single-feature correlation strength of the th feature, characterizes the feature, It is a pest label, and are respectively and the standard deviations of, a summary value characterizing the joint association of multiple features, and the joint association of multiple features , where represents the weight of the feature pair, represents the feature and and the pest label of the mutual information, and are respectively the feature and of the information entropy, represents the feature interaction association value, , where represents the total number of features, , are respectively the feature indices, and , represents the feature interaction weight, represents the feature interaction function, is the sample index, ranging from 1 to , represents the dynamic change association value, , where represents the feature of the time variation, is the adjustment coefficient, represents the feature of the time variation, and and and are respectively the weight coefficients of single feature, multiple features, feature interaction and dynamic change.

[0038] Specifically, a formula that comprehensively considers single-feature association, multi-feature joint association, feature interaction association, and dynamic change association is used to calculate the correlation coefficient of pest generation. This formula is in the form of a weighted sum, which contains four main parts, each of which reflects the impact of different types of feature associations on pest generation. Among them, the single-feature association part considers the association strength between each individual feature (such as temperature, humidity, etc.) and the pest label. This association strength is measured by calculating the correlation coefficient between the feature value and the pest label. The multi-feature joint association part considers the impact of multiple features acting simultaneously on pest generation, and is realized by calculating the weights between feature pairs and their mutual information with the pest label (an index that measures the amount of information shared between two variables). The feature interaction association part considers the impact of the interaction between features on pest generation, and the feature interaction is realized by calculating the feature interaction function, which can capture the complex non-linear relationship between features. The dynamic change association part considers the impact of the change of features over time on pest generation, and is realized by calculating the time change amount of the feature and performing association analysis with the pest label. By simultaneously considering single-feature association, multi-feature joint association, feature interaction association, and dynamic change association, this implementation method comprehensively captures the complex relationship between the feature set and pest generation, obtains a more accurate recognition correlation coefficient, and thus improves the accuracy of pest distribution recognition.

[0039] Step S400: Using the pest distribution map as the basic feature and the joint feature matrix as the additional feature, perform pest development trend prediction based on time series and spatial diffusion, and establish the development trend prediction result.

[0040] Specifically, use historical pest distribution data to construct a time series model (such as ARIMA, LSTM, etc.) to analyze the change trend of pests over time. Based on the spatial distribution characteristics and diffusion mechanism of pests (such as distance dependence, environmental adaptability, etc.), use a spatial diffusion model (a mathematical model used to simulate and predict the spread and diffusion of pests in space, such as cellular automata, spatial autocorrelation model, etc.) to simulate the spatial distribution of pests at future time points. Combine the results of time series analysis and spatial diffusion simulation to establish a pest development trend prediction model. This model can predict the spatial distribution and severity of pests in the future for a period of time.

[0041] In a possible implementation, taking the pest distribution map as the basic feature and the joint feature matrix as the additional feature, and performing pest development trend prediction based on time series and spatial diffusion, step S400 further includes step S410 of extracting the temporal pest distribution change of the pest distribution map, performing time series prediction of the diffusion and time dependence relationship according to the extraction result of the temporal pest distribution change, and establishing the diffusion prediction result at each moment. Specifically, arrange the pest distribution maps in a time series to form a pest distribution time series data set. Apply time series analysis techniques, such as the sliding window method, Fourier transform, etc., to extract the features of the pest distribution changing with time, such as the starting time of pest occurrence, diffusion speed, peak time, etc. Based on the extraction result of the temporal pest distribution change, use time series analysis techniques (such as ARIMA model, LSTM neural network, etc.) to predict the diffusion of pests at future time points, that is, establish a prediction model of pest diffusion changing with time.

[0042] Step S420, taking the pest distribution map and the joint feature matrix as input features, performing local diffusion detail fitting, and establishing a local spatial diffusion fitting result. Specifically, taking the pest distribution map and the joint feature matrix as input features, apply techniques such as spatial autocorrelation analysis and spatial interpolation to preliminarily analyze the spatial pattern of pest diffusion. Then, use machine learning models (such as random forest, convolutional neural network, etc.) to fit and predict the diffusion details of pests in a specific area to capture the spatial heterogeneity of pest diffusion.

[0043] Step S430, performing spatio-temporal development trend prediction on the diffusion prediction result at each moment and the local spatial diffusion fitting result, and establishing a development trend prediction result. Specifically, fuse the prediction result of the temporal pest distribution change and the fitting result of the local diffusion details to form a joint prediction result containing spatio-temporal information. Then, apply ensemble learning methods (such as weighted average, Bayesian model averaging, etc.) or deep learning models (such as spatio-temporal convolutional neural network) to further optimize and correct the joint prediction result to obtain the final development trend prediction result. This implementation method can understand the time dynamics of pest diffusion through the extraction of the temporal pest distribution change; capture the spatial heterogeneity of pest diffusion through local diffusion detail fitting. Finally, through spatio-temporal development trend prediction, it can comprehensively use spatio-temporal information to predict the future development trend of pests, improve the accuracy of pest prediction, and enhance the spatio-temporal resolution and practicality of the prediction result.

[0044] Step S500, performing pest anomaly warning according to the development trend prediction result.

[0045] Specifically, according to the prediction results of the pest development trend and historical data, warning thresholds (values or conditions for triggering the warning mechanism) are set. When the prediction results exceed these thresholds, the warning mechanism is triggered. Warning information is sent to relevant personnel (such as forestry management personnel, pest control experts, etc.) via text messages, emails, App push notifications, etc. The warning information includes the type, level, predicted distribution area, possible impacts, and countermeasures of the pests. In the embodiment of the present application, remote sensing monitoring is performed on the target forest, a remote sensing image data set is established, and ground monitoring data is collected synchronously. The data is interpolated, and based on the time-step rasterized interpolated ground monitoring data of the remote sensing image data set, multi-feature joint extraction is performed on the remote sensing image data set. The extraction results are superimposed onto a unified raster to form a multi-layer feature map. A joint feature matrix is constructed with the rasterized data and the multi-layer feature map. The pest distribution is identified through a classification model, and a pest distribution map with a level evaluation label is generated. Based on the pest distribution map as the basic feature and the joint feature matrix as the additional feature, pest development trend prediction based on time series and spatial diffusion is performed, a prediction result is established, and according to the development trend prediction result, pest anomaly warning and other technical means are carried out, achieving the technical effect of improving the monitoring efficiency and accuracy, and further improving the accuracy of pest prediction.

[0046] In a possible implementation manner, for the pest anomaly warning according to the development trend prediction result, step S500 further includes step S510 of performing trend anomaly trigger analysis based on the development trend prediction result and establishing an anomaly level. The trend anomaly trigger analysis includes trend development speed analysis and trend development scale analysis. Specifically, the speed reference value or expected value of the pest development trend is extracted from historical data. The currently predicted development trend speed is compared with the reference value or expected value to evaluate whether the development speed is abnormal. If the current speed is significantly faster than the reference value or expected value, a speed anomaly is triggered. Similarly, the scale reference value or expected value of the pest development trend is extracted from historical data. The currently predicted development trend scale (including the geographical range affected by the pests, the types and quantities of damaged vegetation, etc.) is compared with the reference value or expected value to evaluate whether the development scale is abnormal. If the current scale is significantly larger than the reference value or expected value, a scale anomaly is triggered. According to the degrees of speed anomaly and scale anomaly, combined with the prediction results of the machine learning model, the anomaly degree of the pest development trend is divided into different levels (such as slight anomaly, moderate anomaly, severe anomaly, etc.).

[0047] Step S520: Conduct pest infestation anomaly warning according to the said anomaly level. Specifically, select a suitable warning method according to the established anomaly level. For minor anomalies, mark or record them in the warning system; for moderate anomalies, send text messages or emails to relevant departments or personnel; for severe anomalies, an emergency response mechanism needs to be activated, including organizing on-site investigations, taking control measures, etc. Among them, the warning information includes the anomaly level of the pest infestation development trend, the possible degree of harm, recommended control measures, etc., which are used to enable relevant departments or personnel to take corresponding measures in a timely manner. This implementation method can accurately identify the anomalies in the pest infestation development trend through trend anomaly trigger analysis, establish corresponding anomaly levels according to the degree of anomalies, which helps relevant departments or personnel take warning measures in a timely manner to prevent the further spread and harm of pest infestations. At the same time, by selecting a suitable warning method and providing detailed warning information, the accuracy and effectiveness of the warning are improved, providing strong support for pest control work.

[0048] In the above text, with reference to Figure 1 A method for predicting the development trend of forest pests and diseases combined with remote sensing monitoring according to an embodiment of the present invention was described in detail. Next, with reference to Figure 2 A system for predicting the development trend of forest pests and diseases combined with remote sensing monitoring according to an embodiment of the present invention will be described.

[0049] The system for predicting the development trend of forest pests and diseases combined with remote sensing monitoring according to an embodiment of the present invention is used to solve the technical problems of insufficient monitoring efficiency and accuracy in the existing prediction of the development trend of forest pests and diseases, resulting in insufficient prediction accuracy, and achieve the technical effect of improving monitoring efficiency and accuracy, and further improving the accuracy of pest and disease prediction. The system for predicting the development trend of forest pests and diseases combined with remote sensing monitoring includes: a ground monitoring data rasterization module 10, a multi-layer feature map formation module 20, a pest distribution identification module 30, a pest development trend prediction module 40, and a pest infestation anomaly warning module 50.

[0050] The ground monitoring data rasterization module 10 is used for remotely sensing and monitoring a target forest, establishing a remote sensing image data set, and synchronously invoking ground monitoring data. The ground monitoring data includes temperature data, humidity data, rainfall data, and vegetation type distribution data. After interpolating the ground monitoring data, rasterize the interpolated ground monitoring data based on the time step of the remote sensing image data set; The multi-layer feature map formation module 20 is used for jointly extracting multiple features from the remote sensing image data set and then superimposing the joint multi-feature extraction results onto a unified grid to form a multi-layer feature map; The pest distribution recognition module 30 is used to construct a joint feature matrix with the rasterized data and the multi-layer feature map, and perform pest distribution recognition based on the joint feature matrix through a classification model to generate a pest distribution map, and the pest distribution map is marked with a grade evaluation; The pest development trend prediction module 40 is used to use the pest distribution map as a basic feature and the joint feature matrix as an additional feature to perform pest development trend prediction based on time series and spatial diffusion, and establish a development trend prediction result; The pest anomaly warning module 50 is used to perform pest anomaly warning according to the development trend prediction result.

[0051] Next, the specific configuration of the multi-layer feature map formation module 20 will be described in detail. As described above, after jointly extracting multiple features from the remote sensing image data set, the joint multi-feature extraction results are superimposed onto a unified grid to form a multi-layer feature map. The multi-layer feature map formation module 20 may further include: The multi-granularity image feature extraction unit is used to perform image preprocessing on the remote sensing image data set and then perform image feature extraction at multiple granularities to establish a multi-granularity feature extraction result; The feature granularity distribution establishment unit is used to establish a feature granularity distribution according to the multi-granularity feature extraction result. The feature granularity distribution includes a leaf damage feature granularity, a large-area pest and disease distribution feature granularity, a temperature anomaly feature granularity, and a spatial diffusion feature granularity; The image extraction scale optimization unit is used to optimize the image extraction scale according to the feature granularity distribution and configure the attention scale according to the image extraction scale optimization result; The multi-feature joint extraction result establishment unit is used to perform scale feature extraction of corresponding features based on the attention scale to establish a multi-feature joint extraction result.

[0052] Among them, the scale feature extraction of corresponding features is performed based on the concerned scale, and a multi-feature joint extraction result is established. The multi-feature joint extraction result establishment unit may further include: an extraction scale determination subunit for inputting the concerned scale and corresponding features into a scale joint extraction network to determine a main extraction scale and an auxiliary scale; a scale feature extraction subunit for performing scale feature extraction of corresponding features according to the main scale and the auxiliary scale, and performing joint analysis of scale feature extraction results to establish a scale joint extraction result; a cross-feature dynamic joint analysis subunit for configuring a feature coupling model and using the feature coupling model to perform cross-feature dynamic joint analysis of the scale joint extraction result to establish a multi-feature joint extraction result.

[0053] Next, the specific configuration of the pest distribution recognition module 30 will be described in detail. As described above, pest distribution recognition is performed through a classification model based on the joint feature matrix to generate a pest distribution map. The pest distribution recognition module 30 may further include: an association analysis unit for performing association analysis of pest generation on the joint feature matrix to establish an identification correlation coefficient; a classification model configuration unit for obtaining a general pest classification model and performing feature migration of the general pest classification model through the identification correlation coefficient to configure a classification model; a pest distribution map generation unit for using the joint feature matrix as input data and inputting it into the classification model to generate a rasterized pest distribution map.

[0054] Among them, for the association analysis of pest generation on the joint feature matrix to establish an identification correlation coefficient, the association analysis unit may further include: a correlation coefficient calculation subunit for calculating the correlation coefficient of pest generation through a formula as follows: ; Among them, represents the identification correlation coefficient, represents the summary value of single-feature associations, single-feature association , where represents the -th single-feature association strength, represents the feature, is the pest label, , are respectively and 's standard deviations, represents the summary value of multi-feature joint associations, multi-feature joint association , where represents the weight of the feature pair, represents the feature , and the mutual information between the pest label , , are the information entropies of features and respectively, representing the feature interaction correlation value, where represents the total number of features, , are the feature indices respectively, and , represents the feature interaction weight, represents the feature interaction function, is the sample index, ranging from 1 to , represents the dynamic change correlation value, where represents the time change amount of feature is the adjustment coefficient, represents the time change amount of feature , , , are the weight coefficients of single feature, multi - feature, feature interaction and dynamic change respectively.

[0055] Next, the specific configuration of the pest development trend prediction module 40 will be described in detail. As described above, using the pest distribution map as the basic feature and the joint feature matrix as the additional feature, perform pest development trend prediction based on time series and spatial diffusion. The pest development trend prediction module 40 can further include: A moment diffusion prediction result establishment unit is used to extract the temporal pest distribution change of the pest distribution map, perform time series prediction of the diffusion and time - dependence relationship according to the extraction result of the temporal pest distribution change, and establish the moment diffusion prediction result; A local diffusion detail fitting unit is used to use the pest distribution map and the joint feature matrix as input features, perform local diffusion detail fitting, and establish the local spatial diffusion fitting result; A time - space development trend prediction unit is used to perform time - space development trend prediction on the moment diffusion prediction result and the local spatial diffusion fitting result, and establish the development trend prediction result.

[0056] Next, the specific configuration of the pest anomaly warning module 50 will be described in detail. As described above, perform pest anomaly warning according to the development trend prediction result. The pest anomaly warning module 50 can further include: An anomaly level establishment unit is used to perform trend anomaly trigger analysis based on the development trend prediction result and establish the anomaly level. The trend anomaly trigger analysis includes trend development speed analysis and trend development scale analysis; An anomaly warning unit is used to perform pest anomaly warning according to the anomaly level.

[0057] The forest pest development trend prediction system combined with remote sensing monitoring provided by the embodiments of the present invention can execute the forest pest development trend prediction method combined with remote sensing monitoring provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method.

[0058] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or server. The included individual units and modules are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of mutual distinction and do not limit the protection scope of the present invention.

[0059] The above specific embodiments do not constitute a limitation to the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present application shall be included within the protection scope of the present application. In some cases, the actions or steps recorded in the present application can be executed in a different order from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous..

Claims

1. A method for predicting the development trend of forest pests and diseases combined with remote sensing monitoring, characterized in that: The method comprises: Conduct remote sensing monitoring of the target forest, establish a remote sensing image data set, and synchronously call ground monitoring data, the ground monitoring data including temperature data, humidity data, rainfall data, and vegetation species distribution data, interpolate the ground monitoring data, and then rasterize the interpolated ground monitoring data based on the time step of the remote sensing image data set; After performing multi-feature joint extraction on the remote sensing image data set, the multi-feature joint extraction results are superimposed on a unified grid to form a multi-layer feature map; A joint feature matrix is ​​constructed using the rasterized data and the multi-layer feature map, and pest distribution identification based on the joint feature matrix is ​​performed through a classification model to generate a pest distribution map, wherein the pest distribution map has a grade evaluation mark; Using the pest distribution map as a basic feature and the joint feature matrix as an additional feature, performing pest development trend prediction based on time series and spatial diffusion, and establishing a development trend prediction result; An abnormal pest warning is issued based on the development trend prediction results.

2. The method for predicting the development trend of forest pests and diseases combined with remote sensing monitoring as claimed in claim 1, characterized in that: After performing multi-feature joint extraction on the remote sensing image data set, the multi-feature joint extraction results are superimposed on a unified grid to form a multi-layer feature map, including: After image preprocessing of the remote sensing image data set, image feature extraction under multiple granularities is performed to establish a multi-granularity feature extraction result; Establishing characteristic particle size distribution according to the multi-granularity feature extraction result, wherein the characteristic particle size distribution includes characteristic particle size of leaf damage, characteristic particle size of large-area pest distribution, characteristic particle size of temperature anomaly, and characteristic particle size of spatial diffusion; Performing image extraction scale optimization according to the characteristic particle size distribution, and configuring the focus scale according to the image extraction scale optimization result; Based on the focus scale, scale feature extraction of corresponding features is performed to establish a multi-feature joint extraction result.

3. The method for predicting the development trend of forest pests and diseases combined with remote sensing monitoring as claimed in claim 2, characterized in that: The step of extracting scale features of corresponding features based on the concerned scale and establishing a multi-feature joint extraction result includes: Inputting the concerned scale and the corresponding features into a scale joint extraction network to determine the main scale and the auxiliary scale to be extracted; Extracting scale features of corresponding features according to the main scale and the auxiliary scale, and performing joint analysis of scale feature extraction results to establish scale joint extraction results; A feature coupling model is configured, and the feature coupling model is used to perform cross-feature dynamic joint analysis of the scale joint extraction result to establish a multi-feature joint extraction result.

4. The method for predicting the development trend of forest pests and diseases combined with remote sensing monitoring as claimed in claim 1, characterized in that: The method of identifying the distribution of pests based on the joint feature matrix by using a classification model to generate a pest distribution map includes: Performing a correlation analysis of pest generation on the joint feature matrix to establish an identification correlation coefficient; Obtaining a universal pest classification model, performing feature migration of the universal pest classification model through the identification association coefficient, and configuring a classification model; The joint feature matrix is ​​used as input data and input into the classification model to generate a rasterized pest distribution map.

5. The method for predicting the development trend of forest pests and diseases combined with remote sensing monitoring as claimed in claim 4, characterized in that: The step of performing a correlation analysis on the joint feature matrix to generate pests and establishing an identification correlation coefficient includes: The correlation coefficient of pest generation is calculated by the formula as follows: ； in, represents the identification correlation coefficient, A summary value representing a single feature association, a single feature association ,in, Characterization The single feature correlation strength of the features, Characterization features, For pest labels, , They are and The standard deviation of A summary value representing the joint association of multiple features. ,in, The weights that characterize feature pairs, Characterization characteristics , Label with pests The mutual information of , Characteristics , The information entropy of Characterizes the feature interaction value, ,in, The total number of features, , are feature indices, and , Characterizes the feature interaction weights, Characterize the feature interaction function, is the sample index, ranging from 1 to , Characterizes the dynamic change of associated values, ,in, Characterization characteristics The time variation of is the adjustment coefficient, Characterization characteristics The time variation of , , , They are the weight coefficients of single feature, multiple features, feature interaction and dynamic change respectively.

6. The method for predicting the development trend of forest pests and diseases combined with remote sensing monitoring as claimed in claim 1, characterized in that: The method uses the pest distribution map as a basic feature and the joint feature matrix as an additional feature to perform pest development trend prediction based on time series and spatial diffusion, including: Extracting the time series pest distribution changes from the pest distribution map, performing time series prediction of the diffusion and time dependency relationship based on the time series pest distribution change extraction results, and establishing a moment-to-moment diffusion prediction result; Using the pest distribution map and the joint feature matrix as input features, performing local diffusion detail fitting, and establishing a local spatial diffusion fitting result; Performing a spatiotemporal development trend prediction on the momentary diffusion prediction result and the local space diffusion fitting result, and establishing a development trend prediction result.

7. The method for predicting the development trend of forest pests and diseases combined with remote sensing monitoring as claimed in claim 1, characterized in that: The abnormal warning of pests according to the development trend prediction result includes: Performing trend anomaly trigger analysis based on the development trend prediction result to establish an anomaly level, wherein the trend anomaly trigger analysis includes trend development speed analysis and trend development scale analysis; An abnormal pest warning is performed according to the abnormal level.

8. The forest pest development trend prediction system combined with remote sensing monitoring is characterized by: The system is used to implement the method for predicting the development trend of forest pests and diseases combined with remote sensing monitoring according to any one of claims 1 to 7, and the system comprises: The ground monitoring data rasterization module is used to perform remote sensing monitoring of the target forest, establish a remote sensing image data set, and synchronously call the ground monitoring data, the ground monitoring data including temperature data, humidity data, rainfall data, and vegetation type distribution data. After interpolating the ground monitoring data, the interpolated ground monitoring data is rasterized based on the time step of the remote sensing image data set; A multi-layer feature map forming module is used to perform multi-feature joint extraction on the remote sensing image data set, and then superimpose the multi-feature joint extraction results on a unified grid to form a multi-layer feature map; A pest distribution identification module, which is used to construct a joint feature matrix with rasterized data and the multi-layer feature map, and to identify pest distribution based on the joint feature matrix through a classification model to generate a pest distribution map, wherein the pest distribution map has a grade evaluation mark; The pest development trend prediction module is used to use the pest distribution map as a basic feature and the joint feature matrix as an additional feature to perform pest development trend prediction based on time series and spatial diffusion and establish a development trend prediction result; The pest abnormality warning module is used to issue an pest abnormality warning based on the development trend prediction result.

Images