Big data-based grassland spodoptera frugiperda identification processing method and system, and medium

Abstract

The invention belongs to the field of agricultural pest prevention and control, and particularly relates to a grassland spodoptera frugiperda recognition processing method based on big data, and the method specifically comprises the steps: dividing a farmland into a plurality of grids, and obtaining and preprocessing an egg mass image, an environment, a vibration signal and population density data in each grid; grid node features are defined, dynamic edge weights are calculated, the edge connection probability is predicted based on the pre-trained network, the connection state of the graph is updated according to updating rules of the connection probability and the edge connection state, and a dynamic space-time heterogeneous graph is obtained; inputting the space-time dynamic heterogeneous graph into a pre-trained graph convolutional network, and outputting density growth rate and migration probability of spodoptera frugiperda; and performing risk level assessment and early warning based on the density growth rate and the migration growth rate. In addition, the invention also discloses a system and a medium for executing the method. According to the method, efficient prediction of density increase and migration probability of the grassland spodoptera exigua is realized, and the problems of low identification precision and inaccurate prevention and control in prevention and control of the grassland spodoptera exigua are effectively solved.

Description

Technical Field

[0001] The present invention belongs to the field of agricultural pest control and involves big data analysis and artificial intelligence technology, and specifically provides a method, system and medium for identifying and processing fall armyworm based on big data. Background Art

[0002] The fall armyworm (Spodoptera frugiperda) is a highly destructive agricultural pest, and its monitoring and control are crucial for ensuring food security. Current technologies for its practical application face numerous challenges. Traditional pest identification methods rely primarily on single-image acquisition, which is susceptible to complex field conditions. Light fluctuations and leaf obstruction often make larvae difficult to identify. While adult insect monitoring can be performed using sex attractant traps to count insects, real-time analysis of population characteristics is not possible.

[0003] In terms of migration route prediction, existing technologies are mostly based on a single meteorological model or historical data statistics, which makes it difficult to capture the synchronization of pest vibration signals between different regions. There is a lack of accurate modeling of the local migration behavior of pests in small-scale farmland, resulting in prediction results often lagging behind actual insect changes.

[0004] At the same time, in the formulation and implementation of prevention and control strategies, traditional methods mostly rely on manual inspection experience or extensive pesticide application, which may cause low-risk areas to waste resources and pollute the environment due to excessive pesticide application, and high-risk areas may miss the best prevention and control opportunities due to untimely response, resulting in low overall prevention and control efficiency. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a fall armyworm identification and processing method, system and medium based on big data, which are used to solve the technical problems of low identification accuracy and inaccurate prevention and control in the prevention and control of fall armyworm.

[0006] To achieve the above objectives, the first aspect of the present invention provides a method for identifying and processing fall armyworm based on big data, comprising:

[0007] S1. Data acquisition and preprocessing: Divide the farmland into multiple grids, and acquire and preprocess the egg mass image data, environmental data, vibration signal data, and insect population density data in each grid;

[0008] S2. Construct a dynamic spatiotemporal heterogeneous graph: Define grid node features, calculate dynamic edge weights, and predict edge connection probabilities based on a pre-trained network. Update the graph's connection state based on the connection probability and edge connection state update rules to obtain a dynamic spatiotemporal heterogeneous graph.

[0009] S3. Graph Convolutional Network Prediction: Input the spatiotemporal dynamic heterogeneous graph into the pre-trained graph convolutional network and output the fall armyworm density growth rate and migration probability for each grid;

[0010] S4. Risk assessment and early warning: Conduct risk level assessment and early warning based on density growth rate and migration growth rate.

[0011] Preferably, the grid node features defined in step S2 are specifically as follows: each node in the grid is defined to contain the egg mass image feature I i (t), environmental data vector E i (t), vibration signal characteristics V i (t) and symbiotic effect factor C i (t), where the symbiotic effect factor is calculated based on insect population density data, and the calculation formula is as follows:

[0012]

[0013] D max =α×NDVI×S

[0014] Where i represents the grid node index, k represents the pest index, t represents time, and P k (i, t) represents the density data of competing pests, D FAW (i, t) represents the density data of fall armyworm, α represents the crop resource coefficient, S represents the grid area, D max represents the environmental carrying threshold, λ k represents the inhibition coefficient of pest k on fall armyworm, and NDVI represents the normalized difference vegetation index of farmland.

[0015] Preferably, the dynamic edge weight in step S2 is calculated as follows:

[0016]

[0017] Where w ij (t) represents the edge weight between grid node i and grid node j at time t, exp() represents the exponential function, E i (t), E j (t) represents the environmental data vectors of grid node i and grid node j, which are composed of temperature, humidity, wind speed, and rainfall; σ represents the standard deviation of the historical environmental data vector, which is obtained by normalizing the historical environmental data vector; S ij (t) represents the cross-correlation value of the vibration signals between nodes, which is used to quantify the synchronization of the vibration signals between grid nodes i and j, reflecting the spatial correlation of pest activities.

[0018] Preferably, the mutual correlation value S of the vibration signal between nodes ij (t) is calculated as follows:

[0019]

[0020] Where Δτ represents the sampling point offset of the time delay; K represents the maximum allowable sampling point offset, that is, the search range of the time delay. d represents the length of the grid edge, v max represents the maximum flight speed of the pest; n represents the sampling point index within the time window T, represents the vibration signal characteristics of node i at time t after preprocessing, Represents the vibration signal characteristics of node j at time t after preprocessing.

[0021] Preferably, the edge connection probability in step S2 is predicted by a pre-trained GRU network, wherein the training and verification process of the pre-trained GRU network includes:

[0022] (1) Collect historical migration path data of fall armyworms according to preset time intervals, and collect historical environmental data and historical vibration signal data according to grid nodes;

[0023] (2) Preprocess the historical environmental data and historical vibration signal data of each grid node, time-align them with the historical migration path data, and then splice them with the preprocessed data at the same time to obtain the node feature vector;

[0024] (3) The edge connection relationship between grid nodes is annotated according to the historical migration path data of fall armyworm, and the edge connection annotation labels between nodes are obtained. Among them, the annotation label is 0, which means that there is no direct migration record between grid node i and grid node j, and the annotation label is 1, which means that there is a migration record between grid node i and grid node j;

[0025] (4) Match the node feature vectors and annotation labels one by one according to time, and divide the data into training set, validation set and test set according to the preset ratio;

[0026] (5) Input the training set and validation set into the constructed GRU network for iterative training and validation, and output the edge connection probability;

[0027] (6) The network loss is calculated using the edge connection probability and the annotation label. When the loss value does not decrease in multiple consecutive rounds, or the verification accuracy reaches the preset accuracy threshold, the training and verification of the GRU network are completed, and the pre-trained GRU network is obtained.

[0028] Preferably, the prediction process of the pre-trained GRU network is as follows: according to the update period ΔT of the edge connection state, {t now -ΔT,t now The node feature vector processed within the time window is input into the pre-trained GRU network, and the predicted edge connection probability p is output. ij(t), and then update the edge connection state of the spatiotemporal heterogeneous graph according to the connection probability and the update rule of the edge connection state, where t now The update rules for the edge connection status at the current time include:

[0029] If p ij (t)>first probability threshold, and w ij When (t)>the first weight threshold, the edge connection relationship is retained;

[0030] If the second probability threshold < p ij (t)≤the first probability threshold, and the geographical distance d between nodes ij < the preset distance threshold, an edge connection relationship between grid node i and grid node j is added, where the geographical distance between nodes is calculated based on the Euclidean distance, 0 < the second probability threshold < the first probability threshold < 1;

[0031] If p ij (t)≤the second probability threshold, the edge connection relationship between grid node i and grid node j is deleted.

[0032] Preferably, the input of the graph convolutional network includes: node feature matrix X i =[I i (t),E i (t),V i (t),C i (t)] and the adjacency matrix A, where I i (t) represents the egg mass image features extracted by inputting the egg mass image data into the pre-trained YOLOv8 network, E i (t) represents the environmental data vector consisting of temperature, humidity, wind speed, and rainfall, V i (t) represents the signal feature, and the element A of the adjacency matrix A ij =w ij , t represents time;

[0033] The output of the graph convolutional network is the fall armyworm density growth rate and migration probability of each grid. The fall armyworm density growth rate and migration probability of each grid are calculated as follows:

[0034]

[0035]

[0036] Where r i (t) represents the density growth rate of fall armyworm, Sigmoid() represents Sigmoid function, and MLP() represents multi-layer perceptron; Represents the high-order feature H output of the Lth layer of the graph convolutional network node i (L), and L is the last layer before applying the sigmoid function; β represents the competitive inhibition coefficient, which is used to adjust the symbiotic effect factor C i The overall effect of (t) was determined through field control experiments to control the global impact of the entire competition effect on the density growth rate; p migrate (i, t) is the migration probability; α1, α2, and α3 represent the weight coefficients of each item, which are determined by fitting the corresponding historical data and satisfy α1+α2+α3=1; ||W(t)|| represents the wind speed, θ represents the preset migration threshold, and D max Indicates the environmental carrying threshold.

[0037] Preferably, in step S4, a weighted summation operation is used to obtain a comprehensive risk value based on the density growth rate and the migration growth rate, and then a Kriging interpolation method is used to generate a farmland risk heat map, and the comprehensive risk value is divided into three levels;

[0038] In addition, the risk level in step S4 is defined as low, medium, and high. The method for assessing the risk level based on the comprehensive risk value is as follows:

[0039] If the comprehensive risk value is less than the first risk threshold, it is defined as low risk and a level 1 signal is sent, and no action is taken;

[0040] If the first risk threshold is less than or equal to the comprehensive risk value and less than the second risk threshold, it is defined as medium risk and a secondary signal is sent. Staff will formulate a regular monitoring plan and use drones to conduct inspections to detect vegetation growth status, where the first risk threshold is less than the second risk threshold.

[0041] If the second risk threshold is ≤ the comprehensive risk value, it is defined as high risk, a level 3 signal is sent, and the coordinates are sent according to the signal to automatically trigger the spraying system.

[0042] A second aspect of the present invention provides a fall armyworm identification and processing system based on big data, which is used to execute a fall armyworm identification and processing method based on big data, including:

[0043] Data acquisition module: used to divide the farmland into multiple grids, acquire and pre-process the egg mass image data, environmental data, vibration signal data and insect population density data in each grid;

[0044] Data processing module: pre-processes the egg mass image data, environmental data, vibration signal data, and insect population density data in each grid, where the pre-processing includes standardizing the time frequency of multi-source data acquisition and feature extraction;

[0045] Spatiotemporal heterogeneous graph module: Based on preprocessed data, it defines grid node features, calculates dynamic edge weights, and predicts edge connection probabilities based on a pretrained network. It then updates the graph's connection state based on the connection probability and edge connection state update rules to obtain a dynamic spatiotemporal heterogeneous graph.

[0046] Data prediction module: This module inputs the spatiotemporal dynamic heterogeneous graph into a pre-trained graph convolutional network and outputs the fall armyworm density growth rate and migration probability for each grid.

[0047] Risk assessment module: Based on density growth rate and migration growth rate, risk level assessment and early warning are carried out.

[0048] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, a method for identifying and processing fall armyworm based on big data can be implemented.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] 1. The present invention integrates field cameras, sensors and other equipment to collect multi-source data such as egg mass images, environmental parameters, vibration signals and insect population density, and combines image recognition, signal processing and other technologies to extract key features, effectively integrating multimodal information such as visual, environmental and biological information, and significantly improving the accuracy and reliability of pest identification. At the same time, it also introduces symbiotic effect factors to quantify the impact of competing pests and environmental resources on the fall armyworm population, scientifically depicting the dynamic relationship between interspecific competition and environmental carrying capacity, and providing an analytical model that is more in line with the actual field ecology for accurately assessing the risk of insect pest occurrence.

[0051] 2. In terms of dynamic prediction and prevention of pests, the present invention captures the spatiotemporal correlation of pest migration in real time by constructing a dynamic spatiotemporal heterogeneous graph, dynamically adjusts the node connection status by combining historical migration data with real-time environmental information, and accurately depicts the spatial correlation and temporal dependence of pest activities. Subsequently, the prediction model based on the graph convolutional network further integrates multimodal features to achieve efficient prediction of the density growth and migration probability of the fall armyworm, effectively solving the problem of the traditional method's delayed response to complex environmental changes.

[0052] 3. The present invention uses a layered early warning mechanism to automatically trigger differentiated prevention and control measures for different risk levels, realizing the transformation of pest control from manual experience-driven to data intelligence-driven, greatly improving the pertinence and automation level of prevention and control strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0054] Figure 1 A schematic diagram of the process of the fall armyworm identification and processing method based on big data provided by the present invention;

[0055] Figure 2 This is a schematic diagram of the framework of the fall armyworm identification and processing system based on big data provided by the present invention. DETAILED DESCRIPTION

[0056] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0057] In order to solve the technical problems of low recognition accuracy and inaccurate prediction of migration paths of fall armyworms in the prior art, the present invention provides a fall armyworm recognition and processing method based on big data, such as Figure 1 As shown, the specific steps include:

[0058] S1. Data acquisition and preprocessing: Divide the farmland into multiple grids, and acquire and preprocess the egg mass image data, environmental data, vibration signal data, and insect population density data in each grid.

[0059] In this application, step S1 specifically includes:

[0060] S11. Multi-source data acquisition: Divide the farmland into multiple grid nodes and obtain multi-source data in each grid, including egg mass image data, environmental data, vibration signal data, and insect population density data.

[0061] In this example, egg mass image data is collected in real time from corn plant leaves and stalks using cameras deployed in the field. This data covers six different light periods daily: morning, noon, and evening, ensuring the data's adaptability to complex lighting conditions. A single acquisition density of no fewer than 50 images per hectare provides a multi-dimensional visual sample for subsequent insect identification and population analysis.

[0062] In the embodiment of the present application, environmental data is obtained through multi-parameter sensors, including temperature, humidity, wind speed and rainfall.

[0063] In the embodiment of the present application, the vibration signal data acquisition mainly focuses on the 20-50Hz frequency band, which corresponds to the wing beat frequency of the adult fall armyworm during flight. The plant vibration signal is collected by a vibration sensor at a sampling frequency of 1000Hz. This frequency band focusing strategy can effectively filter out environmental noise and accurately capture the specific vibration signals of the target pest during flight.

[0064] In the embodiment of the present application, the insect population density data includes fall armyworm density data and competing pest density data.

[0065] It should be noted that the fall armyworm adults are counted using sex-attractant traps. At least one trap is deployed per hectare and the number of traps is counted daily to reflect the dynamics of the adult population. Competing pests such as corn borers and armyworms are monitored through image recognition combined with ground sucker samples or sex-attractant traps. For example, multiple sampling points can be set up in each grid, and samples are taken at preset time intervals. The samples are then fused through Kalman filtering to generate the fall armyworm density data D. FAW (i, t) and the density data of competing pests P k (i, t), realizing dynamic quantitative analysis of field pest communities.

[0066] It should be noted that competing pests can be determined based on historical field surveys. For example, pest populations with a coexistence frequency of ≥50% with the fall armyworm can be identified as competing pests. Specifically, the competing pests are screened by counting the frequency of the two pests appearing simultaneously in the same grid in field survey data for more than three consecutive years, accounting for the proportion of the total survey frequency, to ensure that the included competing pests have a real impact on the population dynamics of the fall armyworm.

[0067] S12. Preprocessing the collected multi-source data, wherein the preprocessing includes standardizing the time frequency of multi-source data collection and feature extraction.

[0068] In an embodiment of the present application, the specific steps for standardizing the time frequency of multi-source data acquisition are: first, determining a common time interval; second, based on the common time interval, downsampling and averaging the high-frequency data and performing time interpolation and completion on the low-frequency data, so that data from different sources form an equally spaced synchronous sequence in the time dimension.

[0069] It is important to note that the collection intervals of multi-source data should be consistent. This is a key prerequisite for ensuring spatiotemporal alignment of data and achieving effective fusion analysis. If the actual collection intervals of different data types differ due to sensor characteristics or business requirements, the time scale can be uniformly calibrated through averaging. This means that based on the timestamp information of the data acquisition and a common preset time interval as a benchmark, high-frequency data is downsampled and averaged, and low-frequency data is time-interpolated and supplemented. This allows data from different sources to form a synchronous sequence with equal intervals in the time dimension. This eliminates analytical errors caused by time asynchrony and ensures strict alignment of multi-dimensional information such as environmental data, vibration signals, egg mass images, and insect population density along the time axis.

[0070] Specifically, using environmental data (10-minute intervals), vibration signals (1000 Hz high-frequency sampling), and insect population density (daily sampling) as examples, we illustrate how to achieve multi-source data synchronization by forcing time interval consistency:

[0071] 1. Determine a common time interval: Assuming that the environmental data is based on a 10-minute interval, all data must be calibrated to this time scale.

[0072] 2. Downsampling of high-frequency data (taking vibration signals as an example)

[0073] Raw data acquisition: The vibration sensor samples at 1000 Hz, which means it collects 1000 data points per second. This results in 600 seconds x 1000 Hz = 600,000 data points every 10 minutes, focusing on signals in the 20-50 Hz frequency band.

[0074] Downsampling: Divide 10 minutes into consecutive, non-overlapping time windows (600 seconds per window). Calculate the effective mean value (RMS) or average amplitude of the vibration signal data within each window as the representative value of the 10-minute interval.

[0075] Output results: 1 vibration characteristic value every 10 minutes (such as the average vibration intensity in the 20-50Hz frequency band), aligned with the time point of the environmental data.

[0076] 3. Interpolation and completion of low-frequency data (taking insect population density as an example)

[0077] Original collection: The number of traps is counted at a fixed time every day, and 1 data point is obtained per 24 hours;

[0078] Interpolation processing: Divide 24 hours into 144 10-minute intervals. At non-sampling time points (such as 08:10, 08:20, etc.), the nearest neighbor interpolation method is used to use the capture amount of the previous sampling point.

[0079] Output result: 1 adult density value every 10 minutes (data during non-sampling period are consistent with the nearest sampling point).

[0080] For example, assuming that the insect population density on the previous day is 5 heads / grid, then before the insect population density data of the second day comes out, the density point data of each time period is 5 heads / grid, that is, each time period is the trapping amount.

[0081] 4. Egg Mass Image Data Alignment

[0082] Original collection: 6 times a day (e.g., 06:00, 10:00, 14:00, 18:00, 22:00, 02:00), each time with an interval of 4 hours (240 minutes), and a single collection of ≥50 images / hectare;

[0083] Time calibration: Each image acquisition time point corresponds to the nearest 10-minute reference point. If the acquisition time is a non-integer 10-minute point (such as 10:05), the data is normalized to the previous reference point (10:00) or the next reference point (10:10) to ensure alignment with other data time points.

[0084] Example of unified data timeline:

[0085]

[0086] In the embodiment of the present application, the feature extraction of the egg mass image data is specifically as follows: the egg mass image is input into the pre-trained YOLOv8 model to extract the egg mass features, and the egg mass image feature I is output. i (t), where the YOLOv8 model training data contains multiple multi-scene images, annotates the egg mass area of ​​the fall armyworm, and uses focal loss to optimize the recognition of difficult samples.

[0087] In the embodiment of the present application, the feature extraction of vibration signal data is specifically as follows: the original vibration signal is decomposed by db4 wavelet 3 layers, the energy feature of the 20-50Hz frequency band is extracted, and the vibration signal feature V is generated by calculating the effective value. i (t), to highlight the specific vibration signals of Spodoptera frugiperda adults during flight.

[0088] In the embodiment of the present application, the feature extraction of environmental data is specifically as follows: normalizing temperature, humidity, wind speed, and rainfall to [0, 1] to generate an environmental data vector E i (t).

[0089] It's also important to note that during data collection and grid division, if the grid density exceeds the device deployment density, resulting in some grid nodes not directly deploying sensors or cameras, the data collected by each device can be assigned to the corresponding grid using nearest neighbor interpolation. Specifically, for grids not directly covered by a device, the nearest device grid is found based on its geographic coordinates. The characteristic values ​​collected by the device, such as environmental data, vibration signals, and insect population density, are then directly assigned to the target grid. This ensures that each grid node receives complete multi-source data for constructing the subsequent node feature vector.

[0090] Based on the above steps, YOLOv8 is used to extract egg mass features, and vibration sensors are used to capture 20-50Hz specific wingbeat signals. Combined with environmental data and satellite remote sensing NDVI, a four-dimensional node feature containing visual, vibration, environmental, and biological factors is constructed. This solves the recognition problem of single-modal data affected by light and noise, and can effectively improve the recognition accuracy of the fall armyworm.

[0091] S2. Construct a dynamic spatiotemporal heterogeneous graph: define grid node features, calculate dynamic edge weights, and predict edge connection probabilities based on the pre-trained network. Update the connection state of the graph according to the update rules of connection probabilities and edge connection states to obtain a dynamic spatiotemporal heterogeneous graph.

[0092] In this application, step S2 specifically includes:

[0093] S21, define grid node features: define each node in the grid to contain egg mass image features I i (t), environmental data vector E i (t), vibration signal characteristics V i (t) and symbiotic effect factor C i (t), where the symbiotic effect factor is calculated based on insect population density data, and the calculation formula is as follows:

[0094]

[0095] D max =α×NDVI×S

[0096] Where i represents the grid node index, k represents the pest index, t represents time, and P k (i, t) represents the density data of competing pests, D FAW (i, t) represents the density data of fall armyworm, α represents the crop resource coefficient, S represents the grid area, D max represents the environmental carrying threshold; λ k represents the inhibition coefficient of pest k on fall armyworm, which is obtained based on historical experimental experience; NDVI represents the normalized difference vegetation index of farmland, which is obtained through satellite remote sensing.

[0097] In this application, the symbiotic effect factor C i (t) Reflects the inhibitory effect of competing pests on Spodoptera frugiperda. By integrating insect population density with environmental carrying capacity, the impact of interspecific competition on population growth is quantified.

[0098] It should be noted that since traditional pest identification and control technologies often ignore the inhibitory effect of competing pests on the population dynamics of fall armyworm and fail to fully consider the interactions between species in field ecosystems, there is a large deviation between the prediction of fall armyworm density growth and the actual insect situation. Therefore, in this application, by introducing the density data of competing pests to calculate the symbiotic effect factor, the interspecific competition relationship between competing pests and fall armyworm is quantified, and then the density growth rate of fall armyworm is corrected, so that the model can more accurately reflect the impact of the actual ecological environment in the field on the population growth of fall armyworm, and provide data support for the accurate assessment of pest risks and the formulation of scientific prevention and control strategies.

[0099] In the embodiment of the present application, the suppression coefficient λ k The results were obtained by fitting historical field experimental data. Specifically, different density gradients of competing pests were set in a controlled environment, and the changes in the population growth rate of fall armyworm were observed. The inhibition coefficient λ of each pest on fall armyworm was obtained by fitting a linear regression model. k The crop resource coefficient α is preset based on the farmland crop type (such as corn and rice) through expert experience, and the value range is 0.8-1.2.

[0100] S22. Calculate the dynamic edge weights in spatiotemporal heterogeneous graphs: Define the edge weights between nodes as shown below:

[0101]

[0102] Where w ij (t) represents the edge weight between grid node i and grid node j at time t, exp() represents the exponential function, E i (t), E j (t) represents the environmental data vectors of grid node i and grid node j, which are composed of temperature, humidity, wind speed, and rainfall; σ represents the standard deviation of the historical environmental data vector, which is obtained by normalizing the historical environmental data vector and is used to adjust the degree of influence of environmental differences on edge weights; S ij (t) represents the cross-correlation value of the vibration signals between nodes, which is used to quantify the synchronization of the vibration signals between grid nodes i and j, reflecting the spatial correlation of pest activities.

[0103] In this application, the cross-correlation value S of the vibration signal between nodes ij (t) is calculated as follows:

[0104]

[0105] Where Δτ represents the sampling point offset of the time delay; K represents the maximum allowable sampling point offset, that is, the search range of the time delay. d represents the length of the grid edge, v max represents the maximum flight speed of the pest; n represents the sampling point index within the time window T, represents the vibration signal characteristics of node i at time t after preprocessing, Represents the vibration signal characteristics of node j at time t after preprocessing.

[0106] S23, edge connection state prediction and update: According to the update cycle ΔT of the edge connection state, {t now -ΔT,t now The node feature vector processed within the time window is input into the pre-trained GRU network, and the predicted edge connection probability p is output. ij (t), and then update the edge connection state of the spatiotemporal heterogeneous graph according to the connection state update rule, where t now The update rules for the edge connection status at the current time include:

[0107] If p ij (t)>first probability threshold, and w ij When (t)>the first weight threshold, the edge connection relationship is retained;

[0108] If the second probability threshold < p ij (t)≤the first probability threshold, and the geographical distance d between nodes ij When the distance is less than the preset threshold, a new edge connection relationship between grid node i and grid node j is added, where the geographical distance between nodes is calculated based on the Euclidean distance;

[0109] If p ij (t)≤the second probability threshold, the edge connection relationship between grid node i and grid node j is deleted.

[0110] In the present application, 0<second probability threshold<first probability threshold<1.

[0111] It should be noted that the update period ΔT in this application is a preset value and may be the same as the preset time interval during the training process.

[0112] For example, when the first probability threshold is 0.8, the second probability threshold is 0.5, the first weight threshold is 0.6, and the preset distance threshold is 5000m, the update rule for the edge connection state may be:

[0113] If p ij (t)>0.8 and w ij(t)>0.6, indicating that there is a stable migration association between nodes, and the connection is retained;

[0114] If 0.5<p ij If (t)≤0.8 and the geographical distance between nodes is <5000m, new connections are added to capture potential migration paths;

[0115] If p ij If (t)≤0.5, invalid connections are deleted to maintain graph sparsity.

[0116] In this application, the training and verification process of the pre-trained GRU network includes:

[0117] (1) Collect historical migration path data of fall armyworms at preset time intervals, and collect historical environmental data and historical vibration signal data based on grid nodes. The historical migration path can be monitored by radar, and combined with a tag release and recovery experiment, the direct migration events between nodes can be verified;

[0118] (2) The historical environmental data and historical vibration signal data of each grid node are preprocessed, time-aligned with the historical migration path data, and then spliced ​​with the preprocessed data at the same time to obtain the node feature vector, as shown in the following formula:

[0119] Z i (t)=[E i1 (t),E i2 (t),E i3 (t),E i4 (t),V i-energy (t)]

[0120] Where, E i1 (t)~E i4 (t) represents the temperature data, humidity data, wind speed data and rainfall data of grid node i at time t, V i-energy (t) represents the energy value of the vibration signal of grid node i at time t in the preset frequency band;

[0121] (3) The edge connection relationship between grid nodes is annotated according to the historical migration path data of fall armyworm, and the edge connection annotation labels between nodes are obtained. Among them, the annotation label is 0, which means that there is no direct migration record between grid node i and grid node j, and the annotation label is 1, which means that there is a migration record between grid node i and grid node j;

[0122] (4) Match the node feature vectors and annotation labels one by one according to time, and divide the data into training set, validation set and test set according to the preset ratio;

[0123] (5) Input the training set and validation set into the constructed GRU network for iterative training and validation, and output the edge connection probability;

[0124] (6) The network loss is calculated using the edge connection probability and the annotation label. When the loss value does not decrease in multiple consecutive rounds, or the verification accuracy reaches the preset accuracy threshold, the training and verification of the GRU network are completed, and the pre-trained GRU network is obtained.

[0125] It should be noted that the historical migration path data in this application refers to the migration path, which is used to calibrate the connection relationship between nodes.

[0126] In this application, the edge connection status is predicted based on the GRU network, and the edge weight is dynamically adjusted in combination with the environmental differences and the cross-correlation value of the vibration signal. The spatiotemporal correlation of the migration of the fall armyworm is captured in real time, which can improve the prediction accuracy of the subsequent migration probability.

[0127] It should be noted that the test set in this application is used to test the accuracy of the pre-trained GRU network model, which belongs to the existing technology. Therefore, the test set is not specifically introduced when building the model in this application.

[0128] S3. Graph convolutional network prediction: Input the spatiotemporal dynamic heterogeneous graph into the pre-trained graph convolutional network and output the fall armyworm density growth rate and migration probability of each grid.

[0129] In this application, step S3 specifically includes:

[0130] S31. Generate an adjacency matrix A.

[0131] In the embodiment of the present application, the adjacency matrix A is generated based on the edge connection state predicted by the environmental data and the vibration characteristics. Specifically, when a new edge connection is added, the dynamic weight of the new edge is obtained according to the dynamic edge weight calculation formula; when an edge connection is deleted, the dynamic weight of the edge is set to 0, thereby obtaining the element A of the adjacency matrix A. ij =w ij .

[0132] S32, the node feature matrix X i And the adjacency matrix A is input into the graph convolutional network, and the heterogeneous features are mapped to a unified latent space through the modal feature projection layer to obtain the projected feature H m (m∈{E,V,C,I}), where the node feature matrix X i As shown in the following formula:

[0133] X i =[I i (t),E i (t),V i (t),C i (t)]

[0134] Where, I i (t) represents the egg mass feature extracted by inputting the egg mass image data into the pre-trained YOLOv8 network, i.e., the egg mass image feature; E i (t) represents the environmental data vector consisting of temperature, humidity, wind speed, and rainfall, V i (t) represents the vibration signal characteristics, t represents time, C i (t) represents the symbiotic effect factor.

[0135] In this application, the environmental data projection is: H E =E i (t)W E , the vibration signal projection is: H V =V i (t)W V , the symbiotic effect factor projection is: H C =C i (t)W C , the image feature projection is: H I =I i (t)W I Among them, W E 、W V 、W C 、W I Represents a trainable weight matrix.

[0136] S33. Use spatial graph convolution to aggregate domain information and obtain spatial features The expression is as follows:

[0137]

[0138] Where, is an adjacency matrix with self-connection, A(t) represents the connection matrix A at time t, I N Represents the identity matrix, with dimensions N×N, and diagonal elements are 1, and off-diagonal elements are 0. It adds the connection features that each node will connect to itself, which originally only represents the external connection relationship between nodes, ensuring that the aggregated neighbor information retains its own characteristics; express The degree matrix is ​​used to avoid the uncertainty of the node matrix values ​​caused by different readings during the aggregation process; represents the spatial convolution weight of the lth layer, Represents the features after projection of the l-1th layer.

[0139] S34, using expansion rate of 2 l(l represents the layer index) Dilated Causal CNN performs temporal convolution to capture temporal dependencies and obtain temporal features The expression is as follows:

[0140]

[0141] Where, represents the temporal features of the lth layer, Represents the spatial features of the lth layer.

[0142] S35. Perform a residual connection between the time series features and the previous layer projection features to complete a spatiotemporal convolution. The expression is as follows:

[0143]

[0144] S36, after multiple layers of spatiotemporal convolution, the cross-modal attention fusion layer is used to further integrate multimodal information to obtain high-order features H (L) , the expression is as follows:

[0145] H (L) =∑ m∈{E.V.C.I} α m H m V(H m )

[0146]

[0147] Where, α m is the output feature H of multi-layer spatiotemporal convolution m The attention weight of H m The feature dimension of H; Q, K are the attention query function and key function, which are used to convert H m Projected into query vector and key vector; V represents the attention value function, which is used to transform H m Projection is a vector of values.

[0148] S37. Based on high-order features, obtain the fall armyworm density growth rate and migration probability for each grid. The expressions are as follows:

[0149]

[0150]

[0151] Where r i (t) represents the density growth rate of fall armyworm, Sigmoid() represents Sigmoid function, and MLP() represents multi-layer perceptron; Represents the high-order feature H output of the Lth layer of the graph convolutional network node i (L), and L is the last layer before applying the sigmoid function; β represents the competitive inhibition coefficient, which is used to adjust the symbiotic effect factor C i The overall effect of (t) was determined through field control experiments to control the global impact of the entire competition effect on the density growth rate; p migrate (i, t) is the migration probability; α1, α2, and α3 represent the weight coefficients of each item, which are determined by fitting the corresponding historical data and satisfy α1+α2+α3=1; ||W(t)|| represents the wind speed, θ represents the preset migration threshold, and D max Indicates the environmental carrying threshold.

[0152] It should be noted that α1 is determined by fitting the historical fall armyworm density data, α2 is determined by fitting the historical fall armyworm density growth rate data, and α3 is determined by fitting the historical wind speed data.

[0153] It should be noted that the preset migration threshold in the embodiment of the present application is determined through experimental experience.

[0154] In this application, according to the density growth rate calculation formula and the migration probability calculation formula, the density growth rate and migration probability of grid node i at time t are obtained, that is, the density growth rate and migration probability of the i-th grid at time t, to reflect the impact of the current environment on the population growth and migration of fall armyworm.

[0155] S4. Risk assessment and early warning: Conduct risk level assessment and early warning based on density growth rate and migration growth rate.

[0156] In this application, step S4 is specifically as follows: based on the density growth rate and the migration growth rate, a weighted summation operation is used to obtain a comprehensive risk value, and then the Kriging interpolation method can be used to generate a farmland risk heat map, and the comprehensive risk value is divided into three levels, where the comprehensive risk value calculation formula is as follows:

[0157] Fx i =ψ1×r i (t)+ψ2×pm igrate (i, t)

[0158] Where, Fx i represents the comprehensive risk value of the i-th grid, ψ1 and ψ2 are the weight coefficients of each item.

[0159] It should be noted that the Kriging interpolation method is used in this application to generate the farmland risk heat map in order to more intuitively show the risks of each grid to the staff.

[0160] In the embodiment of the present application, the weight coefficients ψ1 and ψ2 of each grid are the same, are preset values, and are obtained through practical experience.

[0161] In this application, risk levels are defined as low, medium, and high. The method for assessing risk levels based on comprehensive risk values ​​is as follows:

[0162] If the comprehensive risk value is less than the first risk threshold, it is defined as low risk and a level 1 signal is sent, and no action is taken;

[0163] If the first risk threshold is less than or equal to the comprehensive risk value and less than the second risk threshold, it is defined as medium risk and a secondary signal is sent. Staff will formulate a regular monitoring plan and use drones to conduct inspections to monitor vegetation growth.

[0164] If the second risk threshold is ≤ the comprehensive risk value, it is defined as high risk, a level 3 signal is sent, and the coordinates are sent according to the signal to automatically trigger the spraying system.

[0165] Preferably, the first risk threshold is less than the second risk threshold.

[0166] For example, the first risk threshold is 0.4, and the second risk threshold is 0.7. When the comprehensive risk value is less than 0.4, it is defined as low risk, and a first-level signal is sent without any processing; when 0.4≤comprehensive risk value<0.7, it is defined as medium risk, and a second-level signal is sent. The staff will formulate a regular monitoring plan and use drones for inspections to detect the growth status of vegetation; when 0.7≤comprehensive risk value, it is defined as high risk, and a third-level signal is sent. The coordinates are sent according to the signal, and the spraying system is automatically triggered.

[0167] If a grid node is assessed as high-risk multiple times in a row, early warning information can be pushed to staff via SMS or APP, including the specific grid location and insect population density value, providing basic data for staff's targeted prevention and control. The risk level of the grid node will be reassessed every time the edge connection status is updated.

[0168] Introducing the density of competing pests and the environmental carrying threshold to calculate the symbiotic effect factor and quantifying the inhibitory effect of interspecific competition on the growth of fall armyworm can effectively reduce the error of population density growth rate and provide data support for field ecological regulation.

[0169] Some of the data in the above formula are calculated by removing the dimensions and taking their numerical values. The formula is a formula that is closest to the actual situation obtained by software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained through simulation of a large amount of data.

[0170] like Figure 2 As shown, the embodiment of the present application further provides a fall armyworm identification and processing system based on big data, which is used to perform a fall armyworm identification and processing method based on big data. The system includes:

[0171] Data acquisition module: used to divide the farmland into multiple grids, acquire and pre-process the egg mass image data, environmental data, vibration signal data and insect population density data in each grid;

[0172] Data processing module: pre-processes the egg mass image data, environmental data, vibration signal data, and insect population density data in each grid, where the pre-processing includes standardizing the time frequency of multi-source data acquisition and feature extraction;

[0173] Spatiotemporal heterogeneous graph module: Based on preprocessed data, it defines grid node features, calculates dynamic edge weights, and predicts edge connection probabilities based on a pretrained network. It then updates the graph's connection state based on the connection probability and edge connection state update rules to obtain a dynamic spatiotemporal heterogeneous graph.

[0174] Data prediction module: This module inputs the spatiotemporal dynamic heterogeneous graph into a pre-trained graph convolutional network and outputs the fall armyworm density growth rate and migration probability for each grid.

[0175] Risk assessment module: Based on density growth rate and migration growth rate, risk level assessment and early warning are carried out.

[0176] In the embodiment of the present application, the egg mass image data, environmental data, vibration signal data and insect population density data in each grid can be obtained by deploying a field camera with a resolution ≥2000×1500, a multi-parameter sensor, including monitoring temperature, humidity, wind speed, rainfall, a vibration sensor and a sex attractant trap.

[0177] An embodiment of the present application also provides a computer-readable storage medium, comprising instructions, which, when executed on a computer, enables the computer to execute any of the above methods.

[0178] An embodiment of the present application also provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute any of the above methods.

[0179] An embodiment of the present application also provides a chip, which includes a processor and an interface circuit, the interface circuit is coupled to the processor, the processor is used to run a computer program or instruction to implement the above method, and the interface circuit is used to communicate with other modules outside the chip.

[0180] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using a software program, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that a computer can access or a data storage device / module such as a server or data center that contains one or more media that can be integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive solid state disk, abbreviated as SSD).

[0181] Working principle of the present invention:

[0182] The farmland is divided into grids, and cameras, sensors and other equipment are used to obtain the egg mass images, environment, pest vibration signals and insect population density in the grids. After time alignment and feature extraction, multidimensional data including visual, environmental, vibration and biological factors are formed.

[0183] Each grid node is defined to integrate the above multi-dimensional data, and the symbiotic effect factor is calculated to quantify the inhibitory effect of competing pests on the fall armyworm; the dynamic association between nodes is determined through environmental differences and vibration signal synchronization, and the node connection status is predicted by combining the historical migration data training model to construct a real-time updated spatiotemporal association map.

[0184] The dynamic spatiotemporal graph is input into the pre-trained model, the multimodal features of the grid nodes are integrated, and the density growth rate and migration probability of the fall armyworm in each grid are output to reflect the population growth potential and migration possibility.

[0185] Based on the prediction results, a farmland risk heat map is generated, and different measures are triggered according to the risk level: automatic monitoring for low-risk areas, drone inspections for medium-risk areas, and automatic positioning and triggering of the pesticide application system for high-risk areas. At the same time, accurate prevention and control information is pushed to realize an intelligent fall armyworm monitoring and control system.

[0186] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A method for identifying and processing fall armyworm based on big data, characterized in that: include: S1. Data acquisition and preprocessing: Divide the farmland into multiple grids, and acquire and preprocess the egg mass image data, environmental data, vibration signal data, and insect population density data in each grid; S2. Construct a dynamic spatiotemporal heterogeneous graph: Define grid node features, calculate dynamic edge weights, and predict edge connection probabilities based on a pre-trained network. Update the graph's connection state based on the connection probability and edge connection state update rules to obtain a dynamic spatiotemporal heterogeneous graph. S3. Graph Convolutional Network Prediction: Input the spatiotemporal dynamic heterogeneous graph into the pre-trained graph convolutional network and output the fall armyworm density growth rate and migration probability for each grid; S4. Risk assessment and early warning: Conduct risk level assessment and early warning based on density growth rate and migration growth rate.

2. The method for identifying and processing fall armyworm based on big data according to claim 1, characterized in that: The specific definition of grid node features in step S2 is: define each node in the grid to contain egg mass image feature I i (t), environmental data vector E i (t), vibration signal characteristics V i (t) and symbiotic effect factor C i (t), where the symbiotic effect factor is calculated based on insect population density data, and the calculation formula is as follows: D max =α×NDVI×S Where i represents the grid node index, k represents the pest index, t represents time, and P k (i, t) represents the density data of competing pests, D FAW (i, t) represents the density data of fall armyworm, α represents the crop resource coefficient, S represents the grid area, D max represents the environmental carrying threshold, λ k represents the inhibition coefficient of pest k on fall armyworm, and NDVI represents the normalized difference vegetation index of farmland.

3. The method for identifying and processing fall armyworm based on big data according to claim 1, characterized in that: The calculation method of the dynamic edge weight in step S2 is as follows: Where w ij (t) represents the edge weight between grid node i and grid node j at time t, exp() represents the exponential function, E i (t), E j (t) represents the environmental data vectors of grid node i and grid node j, which are composed of temperature, humidity, wind speed, and rainfall; σ represents the standard deviation of the historical environmental data vector, which is obtained by normalizing the historical environmental data vector; S ij (t) represents the cross-correlation value of the vibration signals between nodes, which is used to quantify the synchronization of the vibration signals between grid nodes i and j, reflecting the spatial correlation of pest activities.

4. The method for identifying and processing fall armyworm based on big data according to claim 3, characterized in that: Cross-correlation value S of vibration signals between nodes ij (t) is calculated as follows: Where Δτ represents the sampling point offset of the time delay; K represents the maximum allowable sampling point offset, that is, the search range of the time delay. d represents the length of the grid edge, v max represents the maximum flight speed of the pest; n represents the sampling point index within the time window T, represents the vibration signal characteristics of node i at time t after preprocessing, Represents the vibration signal characteristics of node j at time t after preprocessing.

5. The method for identifying and processing fall armyworm based on big data according to claim 1, characterized in that: The edge connection probability in step S2 is predicted by the pre-trained GRU network. The training and verification process of the pre-trained GRU network includes: (1) Collect historical migration path data of fall armyworms according to preset time intervals, and collect historical environmental data and historical vibration signal data according to grid nodes; (2) Preprocess the historical environmental data and historical vibration signal data of each grid node, time-align them with the historical migration path data, and then splice them with the preprocessed data at the same time to obtain the node feature vector; (3) The edge connection relationship between grid nodes is annotated according to the historical migration path data of fall armyworm, and the edge connection annotation labels between nodes are obtained. Among them, the annotation label is 0, which means that there is no direct migration record between grid node i and grid node j, and the annotation label is 1, which means that there is a migration record between grid node i and grid node j; (4) Match the node feature vectors and annotation labels one by one according to time, and divide the data into training set, validation set and test set according to the preset ratio; (5) Input the training set and validation set into the constructed GRU network for iterative training and validation, and output the edge connection probability; (6) The network loss is calculated using the edge connection probability and the annotation label. When the loss value does not decrease in multiple consecutive rounds, or the verification accuracy reaches the preset accuracy threshold, the training and verification of the GRU network are completed, and the pre-trained GRU network is obtained.

6. The method for identifying and processing fall armyworm based on big data according to claim 5, characterized in that: The prediction process of the pre-trained GRU network is as follows: according to the update cycle ΔT of the edge connection state, {t now -ΔT,t now The node feature vector processed within the time window is input into the pre-trained GRU network, and the predicted edge connection probability p is output. ij (t), and then update the edge connection state of the spatiotemporal heterogeneous graph according to the connection probability and the update rule of the edge connection state, where t now The update rules for the edge connection status at the current time include: If p ij (t)>first probability threshold, and w ij When (t)>the first weight threshold, the edge connection relationship is retained; If the second probability threshold < p ij (t)≤the first probability threshold, and the geographical distance d between nodes ij < the preset distance threshold, an edge connection relationship between grid node i and grid node j is added, where the geographical distance between nodes is calculated based on the Euclidean distance, 0 < the second probability threshold < the first probability threshold < 1; If p ij (t)≤the second probability threshold, the edge connection relationship between grid node i and grid node j is deleted.

7. The method for identifying and processing fall armyworm based on big data according to claim 1, characterized in that: The input of the graph convolutional network includes: node feature matrix X i =[I i (t),E i (t),V i (t),C i (t)] and the adjacency matrix A, where I i (t) represents the egg mass image features extracted by inputting the egg mass image data into the pre-trained YOLOv8 network, E i (t) represents the environmental data vector consisting of temperature, humidity, wind speed, and rainfall, V i (t) represents the signal feature, and the element A of the adjacency matrix A ij =w ij , t represents time; The output of the graph convolutional network is the fall armyworm density growth rate and migration probability of each grid. The fall armyworm density growth rate and migration probability of each grid are calculated as follows: Where r i (t) represents the density growth rate of fall armyworm, Sigmoid() represents Sigmoid function, and MLP() represents multi-layer perceptron; Represents the high-order feature H output of the Lth layer of the graph convolutional network node i (L) , and L is the last layer before applying the sigmoid function; β represents the competitive inhibition coefficient, which is used to adjust the symbiotic effect factor C i The overall effect of (t) was determined through field control experiments to control the global impact of the entire competition effect on the density growth rate; p migrate (i, t) is the migration probability; α1, α2, and α3 represent the weight coefficients of each item, which are determined by fitting the corresponding historical data and satisfy α1+α2+α3=1; ||W(t)|| represents the wind speed, θ represents the preset migration threshold, and D max Indicates the environmental carrying threshold.

8. The method for identifying and processing fall armyworm based on big data according to claim 1, characterized in that: In step S4, based on the density growth rate and the migration growth rate, a weighted summation operation is used to obtain a comprehensive risk value, and then a Kriging interpolation method is used to generate a farmland risk heat map, and the comprehensive risk value is divided into three levels; In addition, the risk level in step S4 is defined as low, medium, and high. The method for assessing the risk level based on the comprehensive risk value is as follows: If the comprehensive risk value is less than the first risk threshold, it is defined as low risk and a level 1 signal is sent, and no action is taken; If the first risk threshold is less than or equal to the comprehensive risk value and less than the second risk threshold, it is defined as medium risk and a secondary signal is sent. Staff will formulate a regular monitoring plan and use drones to conduct inspections to detect vegetation growth status, where the first risk threshold is less than the second risk threshold. If the second risk threshold is ≤ the comprehensive risk value, it is defined as high risk, a level 3 signal is sent, and the coordinates are sent according to the signal to automatically trigger the spraying system.

9. A fall armyworm identification and processing system based on big data, used to execute the fall armyworm identification and processing method based on big data according to any one of claims 1 to 8, characterized in that: include: Data acquisition module: used to divide the farmland into multiple grids, acquire and pre-process the egg mass image data, environmental data, vibration signal data and insect population density data in each grid; Data processing module: pre-processes the egg mass image data, environmental data, vibration signal data, and insect population density data in each grid, where the pre-processing includes standardizing the time frequency of multi-source data acquisition and feature extraction; Spatiotemporal heterogeneous graph module: Based on preprocessed data, it defines grid node features, calculates dynamic edge weights, and predicts edge connection probabilities based on a pretrained network. It then updates the graph's connection state based on the connection probability and edge connection state update rules to obtain a dynamic spatiotemporal heterogeneous graph. Data prediction module: This module inputs the spatiotemporal dynamic heterogeneous graph into a pre-trained graph convolutional network and outputs the fall armyworm density growth rate and migration probability for each grid. Risk assessment module: Based on density growth rate and migration growth rate, risk level assessment and early warning are carried out.

10. A computer-readable storage medium, characterized in that The storage medium stores a computer program. When the computer program is executed by the processor, the fall armyworm identification and processing method based on big data as described in any one of claims 1 to 8 can be implemented.

Images