大米品质对标体系中多维度关键指标驱动的食味分级模型研究综述

A Review on Palatability Grading Modeling of Rice Driven by Multidimensional Key Quality Indicators within Standardized Quality Benchmarking System

摘要: 米饭食味分级是稻米育种、加工及流通全产业链品质管控的关键环节，传统人工感官品评重复性欠佳、主观误差显著，检测效率难以适配规模化标准化管控需求。为建立客观可量化的大米食味评价体系，本文综述大米品质预测方法五类主流机器学习模型，包括多元线性回归、逐步回归、偏最小二乘、人工神经网络与分类决策树，对比分析各算法在大米理化指标和食味关联建模中的适用范围、固有弊端与应用价值。文章重点阐述逐步回归联合决策树融合建模方案：借助逐步回归筛选影响米饭食味的关键理化特征，再以筛选得到的核心变量构建可解释决策树分级模型，同步输出大米食味预测分值与标准化大米品质定等分级。该建模路径形成完整的大米分级技术体系，可为稻米数字化品质评价平台搭建、大米规范化优质栽培提供理论依据与实操技术方法支撑。

Abstract: Palatability grading of cooked rice is an essential component of full-industry-chain quality control for rice breeding, processing and circulation. Traditional manual sensory evaluation exhibits poor repeatability, obvious subjective deviations and insufficient detection efficiency, making it incapable of satisfying the demands of large-scale and standardized rice quality management. To develop an objective and quantifiable evaluation system for rice eating quality, this study reviews five prevalent statistical and machine learning models commonly used for grain quality prediction, including multiple linear regression, stepwise regression, partial least squares regression, artificial neural network, and classification decision tree. The applicability, inherent limitations and practical application potential of each algorithm are systematically compared in constructing the correlation model between rice physicochemical indices and palatability traits. This study mainly proposes an integrated modeling framework that couples stepwise regression and decision tree algorithms. Firstly, stepwise regression is performed to screen pivotal physicochemical attributes affecting cooked rice palatability. The optimized core variables are subsequently utilized to build an interpretable decision tree grading model, which is capable of outputting continuous palatability prediction values as well as standardized commercial quality grades simultaneously. The established modeling system forms a comprehensive technical paradigm for intelligent rice grading, and offers reliable theoretical foundations and practical technical references for the development of digital rice quality evaluation platforms and standardized high-quality cultivation of fragrant rice.