摘要
近红外光谱技术在烟草化学成分检测中应用广泛,因其快速、无损、环保。针对烟草样本的多样性和复杂性,本研究采用重难点样本加权法,优化样本权重以提高模型性能。结合偏最小二乘回归模型系数的阈值筛选,去除无关特征,探究对模型性能的影响。通过比较采用样本加权和特征筛选后的模型与传统建模方法在含有重难点样本数据处理和模型性能方面的差异,评估了该方法的有效性,结果显示能提高预测精度和稳健性,对模型性能改善效果因重难点样本比例而异。
Near-infrared spectroscopy(NIRS)is widely used in tobacco chemical composition detection due to its rapidity,non-destructiveness and environmental friendliness.Aiming at the diversity and complexity of tobacco samples,this study adopts the weighting method of heavy and difficult samples to optimize the sample weights to improve the model performance.Combined with the threshold screening of partial least squares regression model coefficients,irrelevant features are removed to explore the impact on model performance.The effectiveness of this method is evaluated by comparing the differences in data processing and model performance between the model with sample weighting and feature screening and the traditional modeling method,which shows that it can improve the prediction accuracy and robustness,and the effect of improving the model performance varies depending on the proportion of the sample with critical points.
作者
王永乾
周泉
龙怡帆
Wang Yongqian;Zhou Quan;Long Yifan(Shandong Vocational College of Commerce,Jinan,China;Shandong University,Jinan,China)
出处
《科学技术创新》
2025年第17期29-32,共4页
Scientific and Technological Innovation
关键词
近红外光谱
烟草水分含量
样本权重调整
near infrared spectroscopy
tobacco moisture content
sample weight adjustment
