A machine learning prediction method for rice grain thickness based on multi-feature fusion

By employing a multi-feature fusion machine learning approach, combined with LASSO feature selection and Stacking ensemble, the problems of low efficiency, insufficient accuracy, and high cost in rice grain thickness detection have been solved. This approach enables rapid, non-destructive, and accurate grain thickness prediction, making it suitable for high-throughput breeding and multi-variety adaptation.

CN122433952APending Publication Date: 2026-07-21HAINAN UNIVERSITY SANYA NANFAN RESEARCH INSTITUTE +2
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
HAINAN UNIVERSITY SANYA NANFAN RESEARCH INSTITUTE
Filing Date
2026-06-23
Publication Date
2026-07-21

AI Technical Summary

Technical Problem

Existing rice grain thickness detection technologies suffer from low efficiency, sample damage, high cost, insufficient accuracy, and limited model features, making it difficult to meet the needs of high-throughput breeding and rapid on-site detection.

Method used

A multi-feature fusion machine learning prediction method is adopted. By collecting feature data such as grain length, grain width, thousand-grain weight and grain area, and combining LASSO regularized feature selection and Stacking ensemble method, multiple regression models are constructed to predict grain thickness, realizing adaptive feature selection and model fusion.

Benefits of technology

It enables rapid and non-destructive detection of rice grain thickness, reduces detection costs, improves prediction accuracy and robustness, adapts to different rice varieties and breeding scenarios, and meets the needs of high-throughput breeding.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122433952A_ABST
    Figure CN122433952A_ABST
Patent Text Reader

Abstract

The present application relates to the field of agricultural information technology and intelligent breeding cross technology, and more particularly to a rice grain thickness machine learning prediction method based on multi-feature fusion. 2 The dynamic LASSO screening of the coefficient of variation threshold value eliminates weakly related characteristics such as grain length, and retains core characteristics such as grain width, thousand-grain weight, and grain area that are strongly related to grain thickness. This feature set can simultaneously adapt to the linear fitting characteristics of linear models and the nonlinear fitting characteristics of tree models, allowing the advantages of different types of base models to be fully utilized. Then, through the weighted average + Stacking double-layer integration strategy, the prediction results of each model are integrated to avoid the limitations of a single model. Compared with separately implementing feature selection or separately constructing an integrated model, the present application realizes the deep collaboration of adaptive feature selection and multi-model integration, and the collaborative effect of the two realizes the dual improvement of prediction accuracy and robustness, ensuring the prediction consistency of different batches of rice grain samples, and adapting to different rice categories and breeding scenarios.
Need to check novelty before this filing date? Find Prior Art