碾磨程度对LIBS鉴别稻米产地准确度的影响  

Evaluating impact of milling degree on rice origin identification by laser-induced breakdown spectroscopy

作　　者：谢玮萍 徐将[1] 魏海波 刘林林 姚明印[1] 姚小英[2] XIE Weiping;XU Jiang;WEI Haibo;LIU Linlin;YAO Mingyin;YAO Xiaoying(College of Engineering,Jiangxi Agricultural University,Nanchang 330045,China;College of Vocational Education and Technical Educational,Jiangxi Agricultural University,Nanchang 330045,China)

机构地区：[1]江西农业大学工学院,江西南昌330045 [2]江西农业大学职业师范学院,江西南昌330045

出　　处：《江西农业大学学报》2025年第2期478-485,共8页Acta Agriculturae Universitatis Jiangxiensis

基　　金：国家自然科学基金项目(32260632)。

摘　　要：【目的】稻米经碾磨加工改善感官品质,但碾磨过程会掩盖与原产地相关的信息特征,增加稻米产地辨识难度。激光诱导击穿光谱(LIBS)通过全光学方法感知稻米元素组成信息,具备稻米产地快速、绿色鉴别潜力。旨在采集不同碾磨程度的稻米样品LIBS光谱,结合随机森林(RF)和逻辑回归(LR)机器学习算法,分析碾磨程度对LIBS鉴别稻米产地准确度的影响。【方法】收集江西省内11处产地稻米,经微、轻度、中度、重度4级碾磨,再压片处理形成LIBS分析样品。对比稻米样品LIBS光谱随碾磨程度的变化,评估采用全光谱数据以及利用极限梯度提升(XGBoost)提取特征光谱时,RF、LR模型对稻米产地的鉴别准确度。【结果】(1)随着碾磨程度的提高,稻米内部矿质元素信息发生变化,LIBS可检测到的稻米成分信息减少,从而对稻米产地鉴别准确度造成一定影响。(2)全光谱数据输入时,LR模型的平均鉴别准确率为89.97%,经XGBoost特征提取后,平均鉴别准确率提升至94.85%;RF模型的鉴别准确率从85.11%提升至94.31%,这表明XGBoost特征提取能够显著增强模型鉴别性能。(3)经光谱特征变量选择后,RF模型微、轻度、中度、重度碾磨鉴别准确率分别为97.64%、93.88%、93.15%、92.55%;LR模型分别为97.76%、94.79%、94.48%、92.36%,相较于RF模型,LR模型体现出更优的鉴别性能。【结论】LIBS方法结合机器学习算法对不同碾磨程度下稻米产地均具有鉴别能力,且碾磨程度越低,鉴别准确率越高。[Objective]Rice is milled to improve its sensory quality.However,this process can mask the intrinsic characteristics indicative of the rice’s geographical origin,increasing the difficulty to identfiy the authentic provenance of the product.Laser-Induced Breakdown Spectroscopy(LIBS)is an all-optical method for sensing the elemental composition of rice,which enables rapid and green identification of rice origin.Based on the collection of LIBS spectra of rice samples with different degrees of milling,the effects of milling degree on the accuracy of identifying the origin of rice by LIBS were analyzed by combining machine learning algorithms such as Random Forest(RF)and Logistic Regression(LR).[Method]Rice samples from 11 different origins in Jiangxi Province were subjected to four distinct milling levels,that is,slight,mild,moderate,and severe,to prepare samples for LIBS analysis.The work compared the LIBS spectra of rice samples across these milling levels and evaluated the origin identification accuracy of RF and LR models using both full spectral data and data post-feature extraction via Extreme Gradient Boosting(XGBoost).[Result]The results showed that increasing milling degrees altered the internal mineral element information of rice,reducing the detectable LIBS information and consequently affecting origin identification accuracy.When the full spectrum data was input,the average identification accuracy rate of the LR model was 89.97%,while after feature extraction by XGBoost,the average identification accuracy rate was increased to 94.85%.Similarly,the RF model’s accuracy increased from 85.11%to 94.31%.Notably,XGBoost feature extraction significantly enhanced model performance.Post spectral feature selection,the RF model achieved identification accuracies of 97.64%,93.88%,93.15%,and 92.55%for slight,mild,moderate,and severe milling,respectively.The LR model demonstrated even better performance with accuracies of 97.76%,94.79%,94.48%,and 92.36%,respectively,outperforming the RF model.[Conclusion]This work i

关 键 词：激光诱导击穿光谱(LIBS) 稻米碾磨程度 产地鉴别 XGBoost 特征提取 机器学习 

分 类 号：TS212.4[轻工技术与工程—粮食、油脂及植物蛋白工程] O657.31[轻工技术与工程—食品科学与工程] S511[理学—分析化学]