基于贝叶斯优化XGBoost的真空包装玉米均匀性检测

Bayesian Optimization-Based XGBoost for Uniformity Detection in Vacuum-Packed Corn

摘要: [目的]针对真空包装玉米外观均匀性人工检测效率低、主观性强的问题，提出基于机器视觉的自动化检测方法。[方法] 首先采集真空包装玉米图像，预处理后提取颜色、纹理和空间分布特征，然后采用贝叶斯优化(BO)的XGBoost算法进行特征筛选并建立均匀性预测模型，最后结合SHAP方法分析关键特征的非线性效应。[结果]实验表明，用贝叶斯优化的XGBoost模型在测试集上R²达到0.864，RMSE为3.16，显著优于传统方法，相较于原始XGBoost、RF随机森林、SVM支持向量机、LASSO最小绝对收缩和选择算子模型精度均有不同程度提升；通过关键特征分析发现GLCM纹理熵、L通道四分位距、色差均值、色调标准差、相异性等特征是影响均匀性的主要因素。[结论]所提方法能定量检测真空包装玉米均匀性，为玉米质量自动化分级提供技术支持。

Abstract: [Objective] To address the low efficiency and high subjectivity of manual inspection for uniformity in vacuum-packed corn, this study proposes an automated detection method based on machine vision. [Methods] Images of vacuum-packed corn were first captured, preprocessed to extract color, texture, and spatial distribution features, then subjected to feature selection using the Bayesian Optimization (BO)-based XGBoost algorithm to establish a uniformity prediction model. Finally, the SHAP method was applied to analyze the nonlinear effects of key features. [Results] Experiments demonstrate that the Bayesian-optimized XGBoost model achieves an R² of 0.864 and an RMSE of 3.2 on the test set, significantly outperforming traditional methods. Compared to the original XGBoost, RF Random Forest, SVM Support Vector Machine, and LASSO Least Absolute Shrinkage and Selection Operator models, it exhibits varying degrees of accuracy improvement. Key feature analysis revealed that GLCM texture entropy, L-channel interquartile range, mean chromaticity difference, hue standard deviation, and dissimilarity are primary factors influencing uniformity. [Conclusion] The proposed method enables accurate and rapid detection of uniformity in vacuum-packed corn, providing technical support for automated quality grading of vacuum-packed corn.