Spatio-Temporal Deep Learning for Crop Yield Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current crop yield estimation models face challenges in effectively handling spatio-temporal data due to over-parameterization, non-linear relationships, and collinearity, particularly in large spatial scales, and lack accurate spatio-temporal deep learning methods.
Innovation Solution
A spatio-temporal deep learning method using historical crop yield and meteorological data, incorporating preprocessing, a long short-term memory neural network, attention mechanism, and multi-task output layer to extract time-sequence and spatial features, optimized through hyperparameter tuning and training with the Adam optimization method.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If process mechanism models are used for crop yield estimation, then physiological processes can be quantitatively evaluated, but the model becomes over-parameterized and difficult to apply in large spatial scales
Solution Approach 1:
The patent extracts key meteorological factors (temperature, precipitation, humidity) from complex process mechanism models and integrates them into a deep learning framework. This extraction simplifies the model by removing unnecessary physiological process parameters while retaining the essential meteorological drivers of crop yield, thereby reducing over-parameterization and improving applicability to large spatial scales.
Solution Approach 2:
The patent transforms the model from a process mechanism approach with many physiological parameters to a data-driven deep learning model that uses meteorological parameters as inputs. By changing the parameter type from physiological process parameters to observable meteorological parameters, the model reduces complexity while maintaining estimation accuracy across large spatial scales.
2Ease of manufacture
If statistical regression models are used for crop yield estimation, then simple relationships can be captured, but the model struggles with non-linear relationships and collinearity in the data
Solution Approach 1:
The patent replaces traditional statistical regression models with a deep learning neural network model. This substitution enables the model to automatically learn and capture non-linear relationships between meteorological parameters and crop yield without requiring manual specification of functional forms. The neural network handles collinearity issues through its ability to learn complex feature representations, improving estimation accuracy while maintaining computational efficiency.
3Productivity
If conventional machine learning methods are used for crop yield estimation, then data features can be extracted, but the model lacks accurate spatio-temporal deep learning capability and cannot handle spatial heterogeneity
Solution Approach 1:
The patent introduces spatio-temporal dimensions to the machine learning model by incorporating spatial coordinates and time sequences as input features. The model processes meteorological data across multiple spatial locations and time periods, transforming the conventional 2D data table into a 4D spatio-temporal dataset. This dimensional expansion enables the model to capture spatial heterogeneity and temporal dynamics, significantly improving estimation accuracy while maintaining efficient data processing through deep learning architectures.
Data Source
AI summary
A crop yield estimation method based on spatio-temporal deep learning including: obtaining regional historical crop yield data and meteorological data, preprocessing the meteorological data and the yield data to respectively obtain meteorological parameters and a detrended yield as input and output of the crop yield spatio-temporal deep learning model; constructing the spatio-temporal deep learning model for crop yield estimation, and optimizing hyperparameters; and building a training set by taking the meteorological parameters as an input and the detrended yield as output to train the model and obtain parameters of the model; for the crop yield to be estimated, feeding meteorological parameters into the trained model, and obtaining the crop yield estimation result. The model combined temporal and spatial learning to achieve better crop yield estimation accuracy and stability at large spatial scales.