| 摘要: |
| 【目的】研究成熟期梨可溶性固形物含量的近红外漫反射光谱无损检测技术，为及时、准确地掌握成熟期梨果实的内部品质特性及田间管理、适时采收、合理储藏提供依据。【方法】基于近红外漫反射光谱检测技术分别建立了成熟期砀山酥梨可溶性固形物含量的偏最小二乘（PLS）、广义回归神经网络(GRNN)和偏最小二乘支持向量机动态预测模型（LSSVM），并综合评价了无信息变量消除法（UVE）优选有效特征波数对于简化模型、提高预测性能的影响。【结果】UVE算法能够很好地提高建模效率、有效改善GRNN和LSSVM模型预测精度，而对PLS分析模型效果不明显。3种模型中，LSSVM模型比GRNN和PLS模型具有明显优势，其中UVE LSSVM模型具有最佳预测精度和适用性，其校正相关系数（Rc）为0.988，校正均方根误差（RMSEC）为0.074，预测相关系数（Rp）为0.922，预测均方根误差（RMSEP）为0.162。【结论】基于近红外光谱技术的UVE LSSVM模型可用于成熟期梨可溶性固形物含量的无损检测。 |
| 关键词: 近红外光谱 梨 成熟期 可溶性固形物含量 偏最小二乘支持向量机 无信息变量消除法 |
| DOI： |
| 分类号: |
| 基金项目:国家自然科学基金项目（31171720） |
|
| Nondestructive detection of soluble solids contents in pears during ripening period using near infrared diffused spectroscopy |
|
WANG Minghai, GUO Wenchuan, GU Jingsi,et al
|
|
Northwest A&F University
|
| Abstract: |
| 【Objective】This study aimed to identify internal qualities of pears during ripening period timely and accurately based on nondestructive detection of soluble solids contents in pears during ripening period by near infrared diffused spectroscopy technology.【Method】Near infrared diffused spectroscopy technology coupled with dynamic models of partial least squares(PLS),generalized regression neural network(GRNN),and least squares support vector machine(LSSVM)was established to determine soluble solids contents of Dangshanshu pears during ripening period,and the effectiveness in simplifying model and improving prediction performance with optimal feature wavelengths selected by uninformative variables elimination(UVE)was evaluated.【Result】UVE algorithm had an excellent performance especially in improving modeling efficiency and prediction effectiveness of GRNN and LSSVM models,but had no obvious effect on PLS model.All three models could satisfy practical requirement,while LSSVM model performed better than PLS and GRNN. UVE-LSSVM model gave the highest accuracy and best flexibility,with the correlation coefficient for calibration of 0.998,the root mean square error for calibration of 0.074,correlation coefficient for prediction of 0.922 and root mean square error for prediction of 0.162.【Conclusion】Near infrared spectroscopy detecting technique coupled with UVE-LSSVM model can be applied in nondestructive measurement of pear soluble content in ripening period. |
| Key words: near infrared spectroscopy pear ripening period soluble solids content least squares support vector machine uninformative variables elimination algorithm |
