苹果内部品质分级机械手设计与试验

Design and Experiment of Apple Internal Quality Sorting Manipulator

为检测苹果内部品质,基于可见/近红外光谱检测技术并结合分拣机械手,设计了苹果内部品质分级机械手。该装置主要由夹持机构、近红外光谱采集系统、控制系统等组成。机械手稳定夹持苹果后采集苹果的近红外光谱数据,上位机软件中的预测模型对光谱数据进行分析处理,并显示光谱曲线和预测结果。为建立苹果可溶性固形物含量预测模型,基于该装置采集了苹果在650～1 100 nm波长范围内的光谱数据,通过国家标准测量法测得苹果样本的可溶性固形物含量,采用SG卷积平滑(SG-smooth)、标准正态变量变换(SNV)和多元散射校正(MSC)对光谱数据进行预处理,并结合可溶性固形物含量测量值建立偏最小二乘(PLSR)模型。结果表明,采用多元散射校正方法预处理后的建模效果最优,其预测模型的校正集和预测集相关系数分别为0. 978 2、0. 970 1,均方根误差分别为0. 274 6、0. 326 3°Brix。选取20个同品种苹果样本对该装置的稳定性和准确性进行了测试,可溶性固形物含量预测值与测量值相关系数为0. 957 3,均方根误差为0. 422 4°Brix。试验结果表明,苹果内部品质分级机械手在夹持苹果的同时可以实现对苹果可溶性固形物含量的准确预测。

In order to meet the requirements of apple internal quality inspection, the apple internal quality sorting manipulator was designed based on the visible/near infrared spectroscopy detection technology and the sorting manipulator. The device consisted of three parts: clamping mechanism, near infrared spectroscopy acquisition system and control system. The manipulator clamped the apple and collected the near-infrared spectral data of the apple. The spectral data was analyzed by the predictive model in the upper computer software, and the spectral curve and predicted result were displayed. Based on this device, the visible and near-infrared reflection spectra of apple in the range of 650~1 100 nm were collected. Totally 200 apples were used for the experiment, including 150 apples in the prediction set and 50 apples in the verification set. The soluble solids content of the apples was measured by temperature-compensated refractometer after the collection of spectral information. The collected spectra were pretreated by Savitzky Golay smooth (SG smooth), standard normal variable transformation (SNV) and multiplication scattering correction (MSC). The partial least-squares prediction model of the apple’s SSC was established with spectral data as independent variable and soluble solids as dependent variable. The prediction result that preprocessed with the multi-scattering correction (MSC) method was the best. The correlation coefficients of the calibration set and the verification set of the prediction model were 0.978 2 and 0.970 1, respectively, and the root mean square errors were 0.274 6°Brix and 0.326 3°Brix, respectively. Finally, the accuracy of models was tested. The reflect spectra of 20 samples were collected, and then the real values of these samples’ SSC were measured. The prediction model could give satisfactory results with the correlation coefficient of 0.957 3 and the root mean square error of prediction of 0.422 4°Brix. The results indicated that this device can satisfy the requirements of apple internal quality detection with high accuracy and good performance.