库尔勒香梨感官及理化指标的定量无损检测

Quantitative nondestructive testing of sensory and physical and chemical indexes of Korla fragrant pears

中文关键词:  高光谱成像  库尔勒香梨  颜色  硬度  可溶性固形物  无损检测

英文关键词:hyperspectral imaging technology  Korla fragrant pear  color  hardness  soluble solids  non-destructive detection

中文摘要:

      目的 利用高光谱成像技术建立库尔勒香梨分级指标的快速检测方法。方法 选择采摘期香梨作为研究样本, 以颜色(a*)、硬度(带皮硬度, hardness)和可溶性固形物(soluble solids content, SSC)为研究指标, 使用高光谱成像系统采集样本900~1700 nm范围波长的漫反射光谱。提取样本感兴趣区域(region of interest, ROI)的光谱并对其进行预处理, 采用多元散射校正(muliplication scattering correction, MSC)、标准正态变量变换(standard normal variable transformation, SNV)及其分别与卷积平滑滤波法(savitzky-golay, S-G)相结合的组合处理方法。基于不同的预处理结果建立偏最小二乘回归(partial least squares regression, PLSR)预测模型, 以验证集相关系数(Rv)和均方根误差(RMSEv)对模型进行评价。为进一步优化模型, 采用竞争性自适应重加权算法(competitive adaptive reweighted sampling, CARS)筛选特征波长, 并建立PLSR模型和最小二乘支持向量机(least square-support vector machine, LS-SVM)模型对比建模效果。结果 MSC-SG多光谱数据的预处理效果最好。使用CARS筛选的特征波段建立的LS-SVM模型效果最佳, 香梨a*、硬度和SSC的模型预测值与标准理化值的相关系数分别为0.873、0.908和0.916, 均方根误差分别为0.375、0.385 kg/mm2和0.346%。结论 利用高光谱成像技术可以实现库尔勒香梨多品质参数的无损检测。

英文摘要:

      Objective To explore the rapid detection method for the classification index of Korla fragrant pears by hyperspectral imaging technology. Methods Selecting the pears of picking period as the research samples, using the color (a*), hardness (with skin), and soluble solids content (SSC) as experimental indexes, the diffuse reflectance spectra of the sample in the range of 900-1700 nm was collected by hyperspectral imaging system. The spectra of the region of interest (ROI) were extracted and preprocessed. The methods of multiple scatter correction (MSC), standard normal variable transformation (SNV), and their combined method with savitzky-golay (S-G) respectively was used. Partial least squares regression (PLSR) prediction model was established based on different preprocessing results, and the correlation coefficient of validation (Rv) and the root mean square error of validation (RMSEv) were used to estimate the performance of the models. In order to further optimize the model, the competitive adaptive reweighted sampling (CARS) algorithm was used to select the characteristic wavelengths, and the partial least squares regression (PLSR) model and least squares support vector machine (LS-SVM) model were built to compare the modeling effect. Results The combination of MSC and SG had the best preprocessing effects on spectral data. The results showed that the LS-SVM model established by the CARS algorithm had the best effect. The correlation coefficient of model predictive value and the standard values of the pear color, hardness, and SSC were 0.873, 0.908, and 0.916, respectively; and the root means square errors were 0.375, 0.385 kg/mm2, and 0.346%, respectively. Conclusion Hyperspectral imaging technology can meet the requirements for the non-destructive detection of multi-quality parameters of Korla fragrant pear.