基于深度信念网络与混合波长选择方法的蓝莓糖度近红外检测模型优化  

Optimization of Near-Infrared Detection Model of Blueberry Sugar Content Based on Deep Belief Network and Hybrid Wavelength Selection Method

作　　者：朱金艳 朱玉杰[1] 冯国红[1] 曾明飞 刘思岐 ZHU Jin-yan;ZHU Yu-jie;FENG Guo-hong;ZENG Ming-fei;LIU Si-qi(College of Engineering and Technology,Northeast Forestry University,Harbin 150040,China)

机构地区：[1]东北林业大学工程技术学院,黑龙江哈尔滨150040

出　　处：《光谱学与光谱分析》2022年第12期3775-3782,共8页Spectroscopy and Spectral Analysis

基　　金：黑龙江省自然科学基金项目(LH2020C050);国家自然科学基金面上项目(31870537);中央高校基本科研业务费专项资金项目(2572020BL01)资助。

摘　　要：利用近红外光谱技术结合组合区间偏最小二乘(SiPLS)、竞争性自适应重加权(CARS)、连续投影算法(SPA)、无信息变量消除(UVE)特征提取方法,运用深度信念网络(DBN)建立蓝莓糖度的通用检测模型,实现蓝莓糖度在线无损快速检测。采集了“蓝丰”和“瑞卡”共280个蓝莓样本的近红外光谱,采用手持折光仪测定其糖度;首先利用联合X-Y的异常样本识别方法(ODXY)检测到蓝丰和瑞卡蓝莓分别有2个和4个样本呈现异常,剔除该6个异常样本,对其余274个样本利用光谱-理化值共生距离算法(SPXY)以3∶1的比例划分出训练集和测试集;其次,对比分析卷积平滑(S-G平滑)、中心化、多元散射校正等预处理对蓝莓原始光谱的改善效果,运用SiPLS对光谱降维,筛选特征波段,利用CARS,UVE和SPA方法对特征波段进行二次筛选,以最优的特征波长建立DBN和偏最小二乘回归(PLSR)模型。结果表明,蓝莓糖度近红外检测模型的最优预处理方法为S-G平滑,SiPLS方法挑选的蓝莓糖度最优波段为593~765和1458~1630 nm,UVE算法从SiPLS筛选的346个变量中优选出159个最佳波长。建立蓝莓糖度DBN模型时,分析了不同隐含层数对检测模型的影响,并以交互验证均方根误差(RMSECV)作为适应度函数,利用粒子群算法(PSO)对各隐含层神经元个数在[1,100]之间寻优,发现隐含层为3层且隐含层节点数为67-43-25时,DBN模型的RMSECV达到最小,为0.3977。无论是以全光谱还是特征波长建模,蓝莓糖度近红外DBN模型均优于常规PLSR方法;尤其以UVE方法二次筛选的特征波长建立的模型大大减少了建模变量,且模型精度更高,蓝莓糖度最优的PLSR模型测试集相关系数(R_(P))为0.8875,均方根误差(RMSEP)为0.3959,最优DBN模型R_(P)为0.9542,RMSEP为0.3105。研究表明,利用SiPLS-UVE进行特征提取,结合深度信念网络方法建立的蓝莓糖度检测模型可以更好地完成蓝莓糖度在线精准分析,该方Using near-infrared spectroscopy technology combined with synergy interval partial least square(SiPLS),competitive adaptive reweighted sampling(CARS),successive projection algorithm(SPA)and uninformative variable elimination(UVE)feature extraction methods,the universal detection models of blueberry sugar content were established by deep belief network(DBN)to achieve on-line non-destructive rapid detection.The near-infrared spectra of 280 blueberry samples of“Bluecrop”and“Reka”were collected,and the sugar content of blueberries was determined by a hand-held refractometer.Firstly,using the outlier samples detection based on joint X-Y distances(ODXY)method to detect abnormal samples,there were 2 and 4 abnormal samples from Bluecrop and Reka respectively.After eliminating 6 abnormal samples,the remaining 274 samples were divided into a training set and test set in a ratio of 3∶1 by the sample set partitioning based on the joint X-Y algorithm(SPXY).Secondly,Compared and analyzed the improvement effect of Savitzky-Golay smoothing(S-G smoothing),centralization,multiplicative scatter correction and other pretreatment methods on the original spectrum of blueberry.Using SiPLS to reduce the spectral dimension and filter the characteristic band and using SPA,UVE and CARS,choose characteristic wavelengths again.We established partial least square regression(PLSR)and DBN models with the optimal characteristic wavelengths.The results showed that the optimal pretreatment method of the blueberry sugar content near-infrared detection model was S-G smoothing,the optimal band of blueberry sugar screening by SiPLS method were 593~765 and 1458~1630 nm,and the UVE algorithm was used to select 159 optimal wavelengths from 346 variables screened by SiPLS.When establishing the DBN model of blueberry sugar content,we analyzed the influence of different hidden layer numbers on the detection model,and the root means square error of cross-validation(RMSECV)as a fitness function,the particle swarm optimization(PSO)was used to optimiz

关 键 词：近红外光谱 蓝莓 糖度 无信息变量消除 深度信念网络 无损检测 

分 类 号：O657.3[理学—分析化学]