CN-121998170-A - Crop yield inversion method and device

CN121998170ACN 121998170 ACN121998170 ACN 121998170ACN-121998170-A

Abstract

The specification provides a crop yield inversion method and device, the method comprises the steps of obtaining simulated crop yield obtained through a crop growth model through crop growth simulation based on a meteorological data sequence of a target geographic area, performing supervised training on a machine learning model based on training samples which are input into the meteorological data sequence and labeled as simulated crop yield, determining the trained machine learning model as a crop estimated yield model for predicting the crop yield based on the meteorological data sequence, wherein the machine learning model comprises a sequence encoder for mapping an input sequence into a vector representation sequence, a flattening layer for flattening the vector representation sequence into a global feature vector, and a fully connected layer for mapping the global feature vector into a single numerical value, wherein a sequence unit of the sequence encoder is aligned with a crop growth stage output unit of the crop growth model, and the single numerical value output by the fully connected layer is the crop yield predicted by the machine learning model.

Inventors

FANG RUIXIN
YU LEI
LIANG LEI
Zhong Liheng
XIE NAN
WU ZHAOCONG

Assignees

支付宝(杭州)数字服务技术有限公司
武汉大学

Dates

Publication Date: 20260508
Application Date: 20251229

Claims (11)

1. A method of crop yield inversion, the method comprising: Obtaining simulated crop yield obtained by performing crop growth simulation on the basis of a meteorological data sequence of a target geographic area through a crop growth model; inputting the meteorological data sequence as a model, and constructing a training sample by taking the simulated crop yield as a label; performing supervised training on a machine learning model based on the training sample, and determining the machine learning model after training as a crop estimation model for predicting crop yield based on a meteorological data sequence, wherein the machine learning model comprises a sequence encoder for mapping an input sequence into a vector representation sequence, a flattening layer for flattening the vector representation sequence into a global feature vector, and a fully connected layer for mapping the global feature vector into a single numerical value, a sequence unit of the sequence encoder is aligned with a crop growth stage output unit of the crop growth model, and the single numerical value output by the fully connected layer is the crop yield predicted by the machine learning model.
2. The method of claim 1, the crop generation model being WOFOST model.
3. The method of claim 1, the sequence encoder being a transducer encoder or an LSTM encoder.
4. The method of claim 1, wherein the machine learning model further comprises a convolutional layer for performing feature extraction on an input sequence to obtain a feature data sequence, and the feature data sequence output by the convolutional layer is the input sequence of the sequence encoder.
5. The method of claim 1, the machine learning model further comprising a spatial transformer based on an attention mechanism for weighting input vectors in a feature space, the input vectors of the spatial transformer being global feature vectors output by the flattening layer, the temporal transformer for time-sequential semantic modeling of input vectors resulting in context-optimized feature vectors, the input vectors of the temporal transformer being weighted global feature vectors output by the spatial transformer.
6. The method of claim 5, wherein the spatial Transformer is a transducer-based spatial Transformer, and/or wherein the temporal Transformer is a transducer-based temporal Transformer.
7. The method of claim 1, the method further comprising: inputting the meteorological data sequence as a model, and constructing a fine adjustment training sample by taking the actual crop yield corresponding to the meteorological data sequence as a label; freezing the sequence encoder in the crop estimation model, and performing supervised fine tuning training on the frozen crop estimation model based on the fine tuning training samples.
8. A crop yield inversion apparatus, the apparatus comprising: The acquisition module acquires simulated crop yield obtained by performing crop growth simulation on the basis of a meteorological data sequence of a target geographic area through a crop growth model; the building module is used for inputting the meteorological data sequence as a model, taking the simulated crop yield as a label and building a training sample; The training module is used for performing supervised training on a machine learning model based on the training sample and determining the machine learning model after training as a crop estimation model for predicting crop yield based on a meteorological data sequence, wherein the machine learning model comprises a sequence encoder for mapping an input sequence into a vector representation sequence, a flattening layer for flattening the vector representation sequence into a global feature vector and a fully connected layer for mapping the global feature vector into a single value, a sequence unit of the sequence encoder is aligned with a crop growth stage output unit of the crop growth model, and the single value output by the fully connected layer is the crop yield predicted by the machine learning model.
9. An electronic device, comprising: A processor; A memory for storing processor-executable instructions; Wherein the processor is configured to implement the steps of the method according to any one of claims 1 to 7 by executing the executable instructions.
10. A computer readable storage medium having stored thereon computer instructions which, when executed by a processor, implement the steps of the method of any of claims 1 to 7.
11. A computer program product comprising computer programs/instructions which, when executed by a processor, implement the steps of the method of any of claims 1 to 7.

Description

Crop yield inversion method and device Technical Field One or more embodiments of the present disclosure relate to the field of artificial intelligence, and more particularly, to a method and apparatus for inversion of crop yield. Background In the field of crop management, accurate estimation of crop yield is of great importance for optimizing agricultural production. On one hand, the crop yield estimation is beneficial to scientifically regulating the input of resources such as water, fertilizer and medicine, and reasonably predicting the market price trend of agricultural products, thereby providing basis for making sales strategies in advance and improving planting benefits, and on the other hand, the high-precision crop yield estimation can also effectively reduce the resource waste and improve the quality, the tillage efficiency and the overall agricultural production efficiency of the agricultural products. In addition, accurate crop yield information can optimize the crop space layout under the regional scale, strengthen production process management, and further release the production potential of land. The crop growth model is a mathematical model based on plant physiological and ecological mechanisms, can synthesize multisource environmental factors such as weather, soil, crop variety characteristics, field management measures and the like, dynamically simulate the whole process from sowing, seedling emergence, nutrition growth to reproductive development until mature harvest of crops, and reveal the inherent mechanism and response rule of the growth and development of the crops. Typically represented as WOFOST (WOrld FOod STudies) model, mechanization simulation of yield formation is achieved by quantifying key physiological processes such as photosynthesis, respiration consumption, dry matter distribution, moisture and nutrient absorption, etc. However, when such a mechanism model is extended from field or site scale to regional scale application, spatial heterogeneity limited by surface and near-surface environments is enhanced, difficulty in acquiring key input data required for the model such as soil properties, crop parameters, management information, and the like is increased, and localization of model parameters is also difficult, resulting in a significant increase in uncertainty of model simulation results. Disclosure of Invention In view of this, one or more embodiments of the present disclosure provide the following technical solutions: According to a first aspect of one or more embodiments of the present specification, there is provided a crop yield inversion method comprising: Obtaining simulated crop yield obtained by performing crop growth simulation on the basis of a meteorological data sequence of a target geographic area through a crop growth model; inputting the meteorological data sequence as a model, and constructing a training sample by taking the simulated crop yield as a label; performing supervised training on a machine learning model based on the training sample, and determining the machine learning model after training as a crop estimation model for predicting crop yield based on a meteorological data sequence, wherein the machine learning model comprises a sequence encoder for mapping an input sequence into a vector representation sequence, a flattening layer for flattening the vector representation sequence into a global feature vector, and a fully connected layer for mapping the global feature vector into a single numerical value, a sequence unit of the sequence encoder is aligned with a crop growth stage output unit of the crop growth model, and the single numerical value output by the fully connected layer is the crop yield predicted by the machine learning model. According to a second aspect of one or more embodiments of the present specification, there is provided a crop yield inversion apparatus, the apparatus comprising: The acquisition module acquires simulated crop yield obtained by performing crop growth simulation on the basis of a meteorological data sequence of a target geographic area through a crop growth model; the building module is used for inputting the meteorological data sequence as a model, taking the simulated crop yield as a label and building a training sample; The training module is used for performing supervised training on a machine learning model based on the training sample and determining the machine learning model after training as a crop estimation model for predicting crop yield based on a meteorological data sequence, wherein the machine learning model comprises a sequence encoder for mapping an input sequence into a vector representation sequence, a flattening layer for flattening the vector representation sequence into a global feature vector and a fully connected layer for mapping the global feature vector into a single value, a sequence unit of the sequence encoder is aligned with a crop growth stage output unit of the crop grow