CN-122020181-A - Corn pest and disease damage density prediction method and system based on big data

CN122020181ACN 122020181 ACN122020181 ACN 122020181ACN-122020181-A

Abstract

The invention discloses a corn pest density prediction method and system based on big data, which relate to the technical field of agricultural information and comprise the steps of accessing and preprocessing multi-source heterogeneous agricultural data to generate a normalized field block-time sequence multi-mode characteristic data cube, inputting the normalized field block-time sequence multi-mode characteristic data cube into a general characteristic extraction network to perform characteristic coding, outputting a primary fusion characteristic vector of each field block, optimizing the general characteristic extraction network by utilizing a label-free field block-time sequence multi-mode characteristic data cube through a self-supervision pre-training task, and outputting a high-quality primary fusion characteristic vector containing crop physiological state semantic information. The method is used for extracting the specific characteristics triggered by the biological stress of the plant diseases and insect pests, and stripping the interference of water and fertilizer stress and variety difference confounding factors, so that the interpretability and reliability of the prediction result and the stability under different environments are greatly improved.

Inventors

LI HONG
LIU SHUO
LIU JINWEN
JIN SONG
WANG QIUHUA
LI MENGMAN
SUN YUTING
MOU XINTONG
ZHANG SHUXIN

Assignees

吉林省农业科学院(中国农业科技东北创新中心)

Dates

Publication Date: 20260512
Application Date: 20260224

Claims (10)

1. The corn pest and disease damage density prediction method based on big data is characterized by comprising the steps of accessing and preprocessing multi-source heterogeneous agricultural data to generate a normalized field-time sequence multi-mode characteristic data cube; Inputting the normalized field-time sequence multi-mode characteristic data cube into a general characteristic extraction network to perform characteristic coding, and outputting a primary fusion characteristic vector of each field; Optimizing a general feature extraction network by using a label-free field-time sequence multi-mode feature data cube through a self-supervision pre-training task, and outputting a high-quality primary fusion feature vector containing crop physiological state semantic information; Inputting a high-quality primary fusion feature vector containing crop physiological state semantic information into a causal decoupling module, decoupling and mapping the high-quality primary fusion feature vector to a plant disease and insect pest feature subspace, an environment mixed feature subspace and a shared background feature subspace to obtain plant disease and insect pest related features, and generating a corn plant disease and insect pest density predicted value based on the decoupled plant disease and insect pest related features; And processing the predicted value of the corn pest and disease density to generate a visual early warning product.
2. The method for predicting corn pest and disease damage density based on big data of claim 1, wherein the method for accessing and preprocessing the multi-source heterogeneous agricultural data to generate a normalized field-time sequence multi-modal characteristic data cube comprises the following steps: Receiving multispectral and hyperspectral images, meteorological lattice point data, field spore capturing data and plant protection history generation standing accounts, and performing time alignment and space coordinate unification; performing space clipping and attribute extraction on the processed data by taking the field boundary as a reference to obtain an image sequence, a meteorological sequence, a spore data sequence and a ledger record of each field; Cloud detection and filling are carried out on the field image sequence, a cloud-free image sequence is generated, and a preset wave band combination and mathematical transformation rule are used for extracting a vegetation index sequence; And carrying out missing value complementation, outlier rejection and standardization on the cloud-free image sequence, the meteorological sequence, the spore data sequence, the standing book record and the vegetation index sequence, stacking and integrating according to the unified time sequence length and the characteristic dimension, and generating a standardized field-time sequence multi-mode characteristic data cube.
3. The method for predicting corn pest and disease damage density based on big data of claim 2, wherein the step of inputting the normalized field-time sequence multi-modal feature data cube into a general feature extraction network for feature coding and outputting a primary fusion feature vector of each field comprises the following steps: in the general feature extraction network, a spatial feature extraction module for extracting spatial features and a time sequence feature extraction module for extracting time sequence features are configured; Inputting a normalized tile-time sequence multi-modal feature data cube into the generic feature extraction network; The space feature extraction module performs multi-scale convolution and pooling operation on image class data components in the normalized field-time sequence multi-mode feature data cube to generate space features; the time sequence feature extraction module sequentially inputs sequence data components in the normalized field-time sequence multi-mode feature data cube into the circulating neural network unit according to time sequence, and sequence information is gradually updated and fused through a gating mechanism and state transmission in the unit to generate time sequence features; vector stitching is carried out on the space features and the time sequence features to form primary fusion feature vectors of each field.
4. The corn pest and disease damage density predicting method based on big data of claim 3, wherein the method is characterized in that a general feature extracting network is optimized by a self-supervision pre-training task by using a label-free field-time sequence multi-mode feature data cube, and a high-quality primary fusion feature vector containing crop physiological state semantic information is output, and the method comprises the following steps: The general feature extraction network comprises a spatial feature extraction module, a time sequence feature extraction module and a feature fusion module; The space feature extraction module extracts space features from the random-mask non-tag field-time sequence multi-mode feature data cube, the time sequence feature extraction module extracts time sequence features from the random-mask non-tag field-time sequence multi-mode feature data cube, the feature fusion module fuses the space features and the time sequence features, and the reconstruction head network predicts the data content of the masked part based on the fusion features; the space feature extraction module and the time sequence feature extraction module respectively extract space features and time sequence features from the unlabeled field-time sequence multi-mode feature data cube, and the space features and the time sequence features are aligned to a shared feature space through the feature fusion module; The time sequence feature extraction module extracts time sequence features from a label-free field-time sequence multi-mode feature data cube injected with an artificial synthetic abnormal mode, and the classification head network judges the occurrence position of the abnormality and the abnormality based on the time sequence feature judgment sequence; and updating parameters of the spatial feature extraction module, the time sequence feature extraction module, the feature fusion module and a related task head network by jointly optimizing the total loss function of the multi-mode mask reconstruction task, the cross-mode contrast alignment task and the time sequence abnormality detection task, and outputting a high-quality primary fusion feature vector containing crop physiological state semantic information.
5. The method for predicting corn pest and disease damage density based on big data of claim 4, wherein the step of inputting the high-quality primary fusion feature vector containing the crop physiological state semantic information to a causal decoupling module, and decoupling and mapping the high-quality primary fusion feature vector to a pest and disease damage feature subspace, an environment confounding feature subspace and a shared background feature subspace to obtain pest and disease damage related features comprises the following steps: The causal decoupling module comprises a plant disease and insect pest characteristic mapping unit, an environment confounding characteristic mapping unit and a shared background characteristic mapping unit; respectively inputting high-quality primary fusion feature vectors containing crop physiological state semantic information into a plant disease and insect pest feature mapping unit, an environment hybrid feature mapping unit and a shared background feature mapping unit, respectively generating plant disease and insect pest feature subspace vectors, environment hybrid feature subspace vectors and shared background feature subspace vectors, and forming a decoupled feature representation; And extracting the plant disease and insect pest characteristic subspace vector from the plant disease and insect pest characteristic subspace vector, the environment mixed characteristic subspace vector and the shared background characteristic subspace vector as plant disease and insect pest related characteristics.
6. The method for predicting corn pest density based on big data of claim 5, wherein generating the corn pest density prediction value based on the decoupled pest-related features comprises the steps of: Constructing a multi-layer perceptron regressor for density prediction based on the plant disease and insect pest characteristic subspace vector, and inputting plant disease and insect pest related characteristics into the multi-layer perceptron regressor; and the multi-layer perception machine regressor carries out nonlinear transformation and feature mapping on the relevant features of the plant diseases and insect pests, and predicts the density of the plant diseases and insect pests of the corn.
7. The method for predicting corn pest density based on big data of claim 6, wherein the method for processing the corn pest density predicted value to generate a visual early warning product comprises the following steps: Performing deviation correction based on historical error statistics on the corn pest density predicted value to generate a corrected corn pest density predicted value; Performing uncertainty evaluation based on Monte Carlo Dropout on the corrected corn pest density predicted value to generate a confidence interval of the corn pest density predicted value; And comparing the corrected predicted value of the corn disease and pest density with a preset economic threshold value to generate a preliminary early warning level, and generating a visualized early warning product of geospatial distribution, a confidence range and a hierarchical prevention and control suggestion by combining the confidence interval of the predicted value of the corn disease and pest density and the preliminary early warning level.
8. The corn pest density prediction system based on big data is based on the corn pest density prediction method based on big data according to any one of claims 1-7, and is characterized by comprising a processing module, a data processing module and a data processing module, wherein the processing module is used for accessing and preprocessing multi-source heterogeneous agricultural data to generate a normalized field block-time sequence multi-mode characteristic data cube; The feature coding module inputs the normalized field-time sequence multi-mode feature data cube into the general feature extraction network to perform feature coding and outputs a primary fusion feature vector of each field; The optimization module optimizes the general feature extraction network by using a label-free field-time sequence multi-mode feature data cube through a self-supervision pre-training task and outputs a high-quality primary fusion feature vector containing crop physiological state semantic information; The causal decoupling module inputs high-quality primary fusion feature vectors containing crop physiological state semantic information into the causal decoupling module, and the high-quality primary fusion feature vectors are decoupled and mapped into a plant disease and insect pest feature subspace, an environment mixed feature subspace and a shared background feature subspace to obtain plant disease and insect pest related features, and a corn plant disease and insect pest density predicted value is generated based on the decoupled plant disease and insect pest related features; And the visualization module is used for processing the predicted value of the corn pest density and generating a visual early warning product.
9. A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the computer program is characterized in that the processor realizes the steps of the corn pest density prediction method based on big data according to any one of claims 1-7 when executing the computer program.
10. A computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a processor implements the steps of the big data based corn pest density prediction method of any one of claims 1 to 7.

Description

Corn pest and disease damage density prediction method and system based on big data Technical Field The invention relates to the technical field of agricultural information, in particular to a corn pest and disease damage density prediction method and system based on big data. Background In the technical field of agricultural information, crop pest prediction based on multi-source data and artificial intelligence is an important research direction, the prior art generally integrates multi-source data such as satellite remote sensing, weather, sensors and the like, and after preprocessing, a deep learning model combining convolution and a cyclic neural network is utilized for feature extraction and prediction, and a prediction model is constructed through supervised learning, so that mapping from data to pest and disease levels is realized, and the development of qualitative to quantitative agricultural monitoring is promoted. The prior art faces two core bottlenecks that the model severely depends on a large amount of precisely marked pest data, the data acquisition cost is high, the samples are few, the generalization capability of the model in a new environment is limited, the abnormal crop growth is the result of multi-factor combined action in observation, the existing model is difficult to effectively distinguish pest and disease stress and other environmental stresses from mixed characteristics, false correlation relations are easy to learn, and the reliability and the interpretability of prediction are reduced. Disclosure of Invention The present invention has been made in view of the above-described problems occurring in the prior art. Therefore, the invention provides a corn pest density prediction method based on big data, which solves the problems of weak generalization capability and insufficient prediction reliability caused by high cost of training data labeling and difficulty in separating pest specificity characteristics from multi-factor coupled crop phenotypes of the existing prediction model. In order to solve the technical problems, the invention provides the following technical scheme: in a first aspect, the invention provides a corn pest density prediction method based on big data, which comprises the steps of accessing and preprocessing multi-source heterogeneous agricultural data to generate a normalized field-time sequence multi-mode characteristic data cube; Inputting the normalized field-time sequence multi-mode characteristic data cube into a general characteristic extraction network to perform characteristic coding, and outputting a primary fusion characteristic vector of each field; Optimizing a general feature extraction network by using a label-free field-time sequence multi-mode feature data cube through a self-supervision pre-training task, and outputting a high-quality primary fusion feature vector containing crop physiological state semantic information; Inputting a high-quality primary fusion feature vector containing crop physiological state semantic information into a causal decoupling module, decoupling and mapping the high-quality primary fusion feature vector to a plant disease and insect pest feature subspace, an environment mixed feature subspace and a shared background feature subspace to obtain plant disease and insect pest related features, and generating a corn plant disease and insect pest density predicted value based on the decoupled plant disease and insect pest related features; And processing the predicted value of the corn pest and disease density to generate a visual early warning product. As a preferable scheme of the corn pest density prediction method based on big data, the invention comprises the following steps: accessing and preprocessing multi-source heterogeneous agricultural data to generate a normalized field-time sequence multi-mode characteristic data cube, comprising the following steps: Receiving multispectral and hyperspectral images, meteorological lattice point data, field spore capturing data and plant protection history generation standing accounts, and performing time alignment and space coordinate unification; performing space clipping and attribute extraction on the processed data by taking the field boundary as a reference to obtain an image sequence, a meteorological sequence, a spore data sequence and a ledger record of each field; Cloud detection and filling are carried out on the field image sequence, a cloud-free image sequence is generated, and a preset wave band combination and mathematical transformation rule are used for extracting a vegetation index sequence; And carrying out missing value complementation, outlier rejection and standardization on the cloud-free image sequence, the meteorological sequence, the spore data sequence, the standing book record and the vegetation index sequence, and stacking and integrating according to the unified time sequence length and the characteristic dimension to generate a standardize