针对目前库尔勒香梨品质在线分级检测系统存在价格昂贵、 结构复杂等问题, 设计了库勒尔香梨内部品质在线无损检测分级系统。 基于该系统研究了不同移动速度(0.3和0.5 m·s-1)对库尔勒香梨的可溶性固形物含量(solid soluble contents, SSC)和硬度在线预测模型的影响。 不同移动速度下, 采集样品相同部位的信息, 所采集光谱存在差异。 由于采集的光谱存在差异性, 采用SG-平滑(Savitzky-Golay smooth)、 SG卷积导数、 多元散射校正(MSC)、 标准正态能量变换(SNV)、 归一化(Normalization)等多种光谱预处理方法进行处理, 基于偏最小二乘法(partial least squares, PLS), 建立移动速度为0.3 m·s-1 (S1)和0.5 m·s-1 (S2)下库尔勒香梨的SSC和硬度模型。 结果表明: 移动速度为0.5 m·s-1下, 采用SG-DER(Savitzky-Golay Derivative)处理光谱图建立SSC模型优于0.3 m·s-1, 其预测集相关系数和预测均方根误差为0.880 2和0.391 5°Bri。 而在移动速度为0.3 m·s-1下的结果, 采用SGS(Savitzky-Golay smooth)处理光谱图建立的SSC模型优于0.5 m·s-1下的结果, 其预测集相关系数和预测均方根误差分别为0.820 2和0.470 8 N。 后建立两个速度混合模型, 采用竞争性自适应重加权算法(competitive adaptive reweighted sampling, CARS)和连续投影算法(successive projections algorithm, SPA)筛选特征变量, 后采用PLS, 建立混合速度下硬度和SSC预测模型。 从建模效果来看SPA和CARS都可以有效减少建模所用变量数、 提高库尔勒香梨在线SSC和硬度检测模型的预测能力和运算速度, 增强模型的稳健性等。 采用CARS方法, 从501个光谱中筛选出24个变量, 建立了CARS-PLS模型, 建立的SSC模型较好, 其预测集相关系数和预测均方根误差分别为0.915 0和0.371 9°Bri。 采用SPA方法, 从501个光谱中筛选出32个变量, 建立硬度模型较好, 其预测集相关系数和预测均方根误差分别为0.821 0和0.492 0 N。 混合速度建立预测品质模型比单一速度建立模型稳健一些。 研究表明: 不同移动速度对建立果品品质预测模型产生不同影响, 该研究有助于果品品质在线分选提供技术支持。

With the aim of solving problems related to cost, and the complicated structure of the online grading and inspection system for detecting the quality of pears, the online non-destructive system was designed for inspecting and classification of the internal quality of pears. Based on the system, the effects of prediction models on the Soluble Solids Content (SSC) and firmness of pears were researched under the different moving speeds (0.3 and 0.5 m·s-1) . Collected spectra from the same position of the pear were discrepancy at different moving speeds. Due to the discrepancy in the collected spectra, adapting spectral pre-processing methods, as SG-smoothing, SG-convolution derivative, multiple scattering correction (MSC), standard normal energy transformation (SNV), Normalization, was to eliminate differences. Adopt Partial Least Squares (PLS), prediction models of SSC and hardness for Korla Pears were established at moving speeds of 0.3 m·s-1 (S1) and 0.5 m·s-1 (S2). The results showed that the established SSC prediction model at 0.5 m·s-1 was more effective than 0.3 m·s-1 by using SG-DER (Savitzky-Golay Derivative) processing spectrogram. The correlation coefficient of the prediction set, and the root mean square errors of prediction were to be 0.880 2 and 0.391 5°Brix respectively. However, when the moving speed was 0.3 m·s-1, established the SSC model, by adapting SGS (Savitzky-Golay Smooth) processing spectrogram, was more robust than at 0.5 m·s-1. Its correlation coefficient of the prediction set, and the root mean square errors of prediction were to be 0.820 2 and 0.470 8 N respectively . Afterwards two speed hybrid prediction models were established. Competitive adaptive re-weighted sampling (CARS) and Successive projections algorithm (SPA) were used to select the characteristic variables, and PLS was used to establish hardness and SSC prediction models at mixed speeds. In view of the perspective of the model effect, SPA and CARS effectively reduced the number of variables, improving the online prediction ability and processing data speed, and enhancing the robustness of the model. Using CARS to select 24 variables from a total of 501, then which established the CARS-PLS model. Establishing the SSC prediction model was more efficient, and its correlation coefficient of the prediction set and root mean square errors of prediction were calculated as 0.915 0 and 0.371 9°Brix respectively. Using SPA to select, 32 variables were selected from a set of 501, and a firmness model was established. The correlation coefficient of the prediction set and the root mean square errors of prediction were ascertained as 0.821 0 and 0.492 0 N respectively. Establishing predictive quality model at the mixing speed is more robust than at the single speed. The research showed that the different moving speeds have different effects on the fruit quality prediction models. The research provides technical support for on-line classification of fruit quality.<英文关键词Near-infrared spectroscopy;Korla pear;Different movement speeds;On-line inspection