Insect pest prediction method based on multi-dimensional data deep learning
By using deep learning methods based on multidimensional data, convolutional networks and recurrent components are used to capture the periodic dependence of pests, solving the problem of inaccurate pest prediction, achieving accurate medium- and long-term pest prediction, and improving control efficiency and agricultural economic benefits.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-10
- Publication Date
- 2026-04-10
AI Technical Summary
Existing technologies are not very accurate in pest prediction, making it difficult to achieve precise control. Furthermore, human experience-based predictions have large errors, leading to increased agricultural production costs and greater pressure on the ecological environment.
We employ a deep learning approach based on multidimensional data, using convolutional networks and recurrent components to extract patterns and local dependencies in the temporal dimension. We combine GRU and ReLU functions to compute hidden states, and use a cyclic skip structure to capture periodic dependencies. We also incorporate autoregressive linear models to improve prediction accuracy.
It enables accurate medium- and long-term forecasting of pests, improves the efficiency and effectiveness of pest control, reduces labor and time costs, and maintains the balance of the agricultural ecological environment.
Smart Images
Figure CN121835964A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical field of agricultural monitoring, applying deep learning to pest prediction. Background Technology
[0002] my country is a major agricultural country, and pests and diseases can have a devastating impact on crops. In recent years, extreme weather events have increased significantly, leading to a rise in the number of pests and diseases, increasing the pressure and costs of crop cultivation. Pest control measures mainly involve using insecticides, pesticides, and traps to kill pests. However, due to my country's vast arable land and complex terrain, common control measures often require significant time and labor costs. Improving the efficiency and effectiveness of pest control measures will help maintain the balance of the agricultural ecological environment and increase agricultural economic benefits.
[0003] Pest populations often exhibit cyclical patterns with seasonal variations and are strongly correlated with temperature and humidity. Different pest populations require different control densities. Only by planning control strategies in advance based on pest forecasts can the effectiveness of control measures be maximized.
[0004] While artificial intelligence has been applied to various sectors of social production, it has not received much attention in pest prediction. Current technologies are not very accurate in predicting pests, and are highly dependent on chance. Some methods even rely solely on production experience to manually estimate the scale of pests, making it difficult to implement precise pest control. Summary of the Invention
[0005] The purpose of this invention is to provide a medium- to long-term pest prediction method based on deep learning of multidimensional data, to assist in the precise prevention and control of pests.
[0006] The following is the technical solution of the present invention to solve the above problems:
[0007] Step 1: Obtain the data needed to train the model and form a multidimensional variable as the input matrix.
[0008] Step two involves feeding the input matrix into a convolutional network without pooling to extract short-term patterns and local dependencies between variables over time.
[0009] Step 3: The output of the convolutional layer is fed back to the recurrent component and the recurrent skip component. The recurrent component is a recurrent layer composed of GRUs, which uses the ReLU function as the hidden update activation function to calculate the hidden state of the recurrent unit on the time series.
[0010] Step four: Use fully connected layers to combine the results of the recurrent component and the recurrent skip component to train the prediction model.
[0011] Step 5: Input 60 days of observation data and output pest prediction data for the next 60 days.
[0012] Furthermore, the data used for model training consists of the number of pests and diseases, the number of days in the year, the highest temperature, and the lowest temperature as raw data, which are then processed into an input matrix with time as the sequence coordinate during the preprocessing process.
[0013] Furthermore, using ReLU instead of the tanh function as the activation function makes gradients easier to backpropagate.
[0014] Furthermore, a loop jump component is used, with a time period length p set as the number of hidden units to be skipped, to capture periodic dependencies.
[0015] Furthermore, the traditional autoregressive linear model is combined with a neural network component to increase the sensitivity over time. Attached Figure Description
[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0017] Figure 1 This is a flowchart of a pest prediction method provided in an embodiment of the present invention. Detailed Implementation
[0018] The following description, in conjunction with the accompanying drawings, illustrates specific embodiments of the present invention, but not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.
[0019] Specifically, this embodiment of the invention provides a pest prediction method based on deep learning, the method comprising the following steps:
[0020] S1: Data preprocessing is performed on the collected agricultural data, using the number of pests and diseases, the number of days in the current year, the highest temperature, and the lowest temperature as multidimensional variables in the input matrix. The input matrix of the deep neural network is...
[0021] S2: X T The input is fed into a convolutional network without pooling to extract short-term patterns and local dependencies between variables in the temporal dimension. The convolutional layer consists of 100 filters with a width of 6 and a height of 2, which, after computation, produce a vector h. k The calculation formula is h k =RELU(Wk *X T +b k And by examining matrix X T To zero-padded to the left of each vector h k The length is T.
[0022] S3: The output of the convolutional layer is simultaneously fed into both the Recurrent and Recurrent-skip components, forming a Recurrent and Recurrent-skip Layer. The recurrent component is a recurrent layer with a gated recurrent unit (GRU) and uses the ReLU function as the hidden update activation function. The hidden state of the recurrent unit is calculated using the following formula: r t =σ(x t W xr +h t-1 W hr +b r ) u t =σ(x t W xu +h t-1 W hu +b u ) c t =RELU(x t W xc +r t ·(h t-1 W hc )+b c ) h t =(1-u t )·h t-1 +u t ·c t In the above formula, · represents the element-wise product, σ is the sigmoid function, and x t This is the input of this layer at time t. The output of this layer is the hidden state at each time stamp, and ReLU is used instead of the original tanh function as the activation function to make its gradient easier to backpropagate.
[0023] S4: To capture long-term relevance, this scheme adopts a time-skip structure to extend the time span of the information flow, optimizing the update process into the following formula: r t =σ(x t W xr +h t-p Whr +b r ) u t =σ(x t W xu +h t-p W hu +b u ) c t =RELU(x t W xc +r t ·(h t-p W hc )+b c ) h t =(1-u t )·h t-p +u t ·c t The input to this layer is the output of the convolutional layer, and p is the number of hidden units skipped. The relative merit value of 0.2 was selected after multiple experimental tests.
[0024] S5: Finally, a fully connected layer is used to combine the outputs of the Recurrent and Recurrent-skip components. The output is calculated using the following formula: In the above formula, h t R It is the hidden state of the Recurrent component at timestamp t; h represents the p hidden states of the Recurrent-skip component from timestamp t-p+1 to t; t D The value is the prediction result of the neural network at timestamp t.
Claims
1. A pest prediction method based on deep learning of multidimensional data, characterized in that, Includes the following steps: Step 1: Obtain the data needed to train the model and form a multidimensional variable as the input matrix. Step 2: Pass the input matrix into a convolutional network without pooling to extract short-term patterns and local dependencies between variables in the time dimension. Step 3: The output of the convolutional layer is fed back to the recurrent component and the recurrent skip component. The recurrent component is a recurrent layer composed of GRUs, which uses the ReLU function as the hidden update activation function to calculate the hidden state of the recurrent unit on the time series. Step 4: Use fully connected layers to combine the results of the recurrent component and the recurrent skip component to train the prediction model. Step 5: Input 60 days of observation data and output pest prediction data for the next 60 days.
2. The pest prediction method based on multidimensional data deep learning according to claim 1, characterized in that... The data used for model training consists of the number of pests and diseases, the number of days in the year, the highest temperature, and the lowest temperature as raw data. During the preprocessing process, an input matrix with time as the sequence coordinate is formed.
3. The pest prediction method based on multidimensional data deep learning according to claim 1, characterized in that... Using ReLU instead of the tanh function as the activation function makes gradients easier to backpropagate.
4. The pest prediction method based on multidimensional data deep learning according to claim 1, characterized in that... Use the loop jump component, set the time period length p as the number of hidden units to skip, and capture periodic dependencies.
5. The pest prediction method based on multidimensional data deep learning according to claim 1, characterized in that... Finally, the traditional autoregressive linear model is combined with the neural network part to increase the sensitivity on the time scale.