基于近红外光谱的玉米水分含量快速检测研究

Research on Rapid Detection of Corn Moisture Content based on Near-infrared Spectroscopy

针对传统玉米水分检测方法测试成本高、周期长的问题,研究将近红外光谱分析技术与化学计量方法相结合构建玉米水分含量快速检测模型。通过卷积平滑结合一阶导数对光谱数据进行预处理后,得到一个基于3∶1比率Kernard-Stone方法和联合间隔偏最小二乘法(SiPLS)的校准集和验证集。并使用联合区间偏最小二乘法(SiPLS)和后向区间偏最小二乘法进行水分特征波长优选,并建立相应的偏最小二乘(PLS)回归校正模型。SiPLS优选特征谱区的建模精度最高,其验证集的决定系数、均方根误差和残余预测偏差分别为0.994、0.023和12.777,能够满足玉米水分含量快速检测的需求。

In view of the high test cost and long period of traditional corn moisture detection methods,the near-infrared spectroscopy analysis technology and chemometric methods were combined to construct a rapid corn moisture detection model.After preprocessing the spectral data by convolution smoothing and the first derivative,a calibration set and a verification set based on the 3 ∶1 ratio Kernard-Stone method and the joint synergy interval PLS(SiPLS)method were obtained.The combined synergy interval PLS(SiPLS)and backward interval partial least squares method were used to optimize the water characteristic wavelength,and the corresponding partial least squares(PLS) regression correction model was established.The SiPLS optimized feature spectrum had the highest modeling accuracy.The determination coefficient,root mean square error and residual prediction deviation of the verification set were0.994,0.023 and 12.777,respectively,which could meet the needs of rapid detection of corn moisture content.