利用高光谱技术对火龙果可溶性固形物含量(SSC)检测进行研究,为火龙果内部品质无损检测提供科学方法.以火龙果为研究对象,对光谱数据进行预处理,应用连续投影算法(SPA)进行特征变量的选择,通过偏最小二乘法(PLS)和前馈反向传播神经网络法(BPNN)建立预测模型,分析了火龙果果皮对SSC 模型预测精度的影响.实验结果表明：采用平滑去噪(MAS) 效果最优,PLS 模型的交叉验证相关系数(Rcv) 为0.8635,交叉验证均方根误差(RMSECV)为0.6791,可提高火龙果可溶性固形物模型精度；通过SPA 算法能够有效地对光谱数据进行降维处理,采用优选的15 个特征变量建立的BPNN 预测模型的预测相关系数(RP)为0.8411,预测均方根误差(RMSEP)为0.8171；果皮对建模结果会产生一定的影响,完整果PLS 模型的(RP)为0.8999,RMSEP 为0.7208；果肉PLS 模型的RP 为0.9304,RMSEP 为0.5291,果肉SSC 模型比完整果SSC 模型的预测能力略高.研究结果表明基于高光谱技术采集的火龙果漫反射光谱进行SSC 无损检测具有可行性.

Using hyperspectrum technology to measure the soluble solid content (SSC) of pitaya can provide a scientific method for the non- destructive measurement of interior quality of pitaya.Pretreatment methods are used to process diffuse reflectance spectroscopy.The successive projections algorithm (SPA) is used to select characteristic variables.The partial least squares model (PLS) and back propagation neural network model (BPNN) are built to predict the SSC of pitaya.The influence of the pitaya peel on the SSC model prediction accuracy is also analyzed.The result of moving average smoothing (MAS) pretreatment method is best.The correlation coefficient of cross calibration (Rcv) of the PLS model built is 0.8635,root mean square error of cross validation (RMSECV) reaches 0.6791.Based on the 15 characteristic variables,the correlation coefficient of prediction (RP) of the BPNN model for predicting SSC of pitaya is 0.8411,and the root mean square error of prediction (RMSEP) is 0.8171.The peel has influence on prediction results of models,RP of the whole pitaya model is 0.8999,and RMSEP is 0.7208;RP of the pitaya model without peel is 0.9304,and RMSEP is 0.5291.The prediction accuracy of the pitaya model without peel is higher than the whole pitaya model.The results indicate that the non-destructive measurement for SSC of pitaya based on hyperspectrum technology is feasible.