机构地区:[1]华中农业大学工学院,湖北武汉430070 [2]农业部长江中下游农业装备重点实验室,湖北武汉430070
出 处:《光谱学与光谱分析》2022年第6期1985-1992,共8页Spectroscopy and Spectral Analysis
基 金:国家自然科学基金项目(31871863,32072302);湖北省重点研发项目(2020BBB072);扬州市科技计划项目(YZ2020047)资助。
摘 要:质构作为一种重要的品质参数,能够显著地影响出缸期皮蛋的凝胶品质,目前没有有效的快速无损检测方法。应用高光谱成像实现对出缸期皮蛋质构特性的无损预测及不同品质皮蛋的检测分级。以不同品质鸭皮蛋为研究对象,采集优质蛋、合格蛋与不合格蛋样本的高光谱数据,对比测定其凝胶质构硬度和弹性参数,对原始光谱进行单一和组合变换(多元散射校正MSC、包络线去除CR、一阶求导FD、包络线去除一阶求导组合CR-FD),分析一维光谱数据与质构硬度和弹性的相关性,发现经CR-FD变换后的光谱反射率与凝胶质构硬度、弹性的相关性最大,分别在波长683和715 nm达到最大值为0.882和0.865;基于最优的光谱变换和二维相关光谱法,以凝胶质构硬度和弹性作为扰动因子,探寻皮蛋硬度和弹性的最优研究区域,结果显示:以硬度为扰动因子时,波长476,539,647,672,728和851 nm处存在自相关峰;在波长483,572,657,739和826 nm处的光谱信号对弹性值较敏感,故最终选择476~851和483~826 nm两个敏感波段分别作为凝胶硬度和弹性的研究区域;对比连续投影算法(SPA)、竞争性自适应重加权算法(CARS)、遗传算法(GA)、粒子群优化算法(PSO)、无信息消除法(UVE)5种不同变量选择方式,发现使用粒子群优化算法结合PLSR模型的检测精度最高:预测硬度的R^(2)_(p)和均方根误差RMSEP为0.826和0.874,RPD为2,预测弹性的R^(2)_(p)和均方根误差RMSEP为0.886和0.402,RPD为1.9。使用3种不同分类器对不同品质皮蛋进行预测,发现对优质蛋、合格蛋与不合格蛋分类的准确率分别达到了97%,92%和100%,基于预测结果的混淆矩阵和ROC曲线对模型进行评估,PLS-DA模型的精度和泛化能力优于BP和RF模型。综上表明,使用高光谱技术可以实现皮蛋质构特性的预测及对不同品质皮蛋的无损分级。As an important quality parameter,texture can significantly affect the gelatin quality of preserved eggs.There is no effectively rapid detection method at present.In this study,hyperspectral imaging technology was used to predict preserved eggs’textural characteristics and classify different qualities.Hyperspectral data of high-quality eggs,qualified eggs and unqualified eggs were collected.The original spectra were transformed by single and combined transformation to analyze the correction between one-dimensional spectral and textural hardness and springiness.It was found that the spectral reflectance after CR-FD transformation was most correlated with the hardness and springiness of the gel texture,and the maximum values were 0.882,0.865 at 683,715 nm,respectively;The hardness and springiness were taken as disturbance factors to explore the optimal research area of the hardness and springiness,the results showed that:When hardness was used as disturbance factor,autocorrelation peaks existed at 476,539,647,672,728 and 851 nm,Spectral signals at 483,572,657,739 and 826 nm were more sensitive to springiness value.Therefore,two sensitive bands,476~851 and 483~826 nm,were finally selected as the study regions for gel hardness and springiness,respectively.Comparing five different variable selection methods(SPA,CARS,GA,PSO and UVE),it was found that PSO-PLSR model had the highest detection accuracy:the R^(2)_(p)and RMSEP for predicted hardness were 0.826 and 0.874 with an RPD of 2.The R^(2)_(p)and RMSEP for predicted springiness were 0.886 and 0.402 with an RPD of 1.9.Three different classifiers were used to predict preserved eggs,and the classification accuracy of high-quality eggs,qualified eggs and unqualified eggs reached 97%,92%and 100%,respectively.The accuracy and generalization ability of the PLS-DA model were better than the BP and RF model based on the confusion matrix and ROC curves of prediction results.In conclusion,the hyperspectral technique can be used to predict preserved eggs’textural characteris
分 类 号:TS253.7[轻工技术与工程—农产品加工及贮藏工程]
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...