寒地粳稻种子的拉曼光谱鉴别方法研究

针对水稻种子的品种鉴别存在检验周期长、种类少等实际问题,本研究提出一种高效、快捷、准确鉴别水稻种子的方法.以拉曼光谱技术为基础、寒地粳稻种子为研究对象,进行快速、准确鉴别.首先,利用Savitzky-Golay(SG)、一阶导(1-Der)、二阶导(2-Der)、迭代自适应加权惩罚最小二乘(AIRPLS)和均值中心化(...

Research on the Raman spectroscopic identification of cold-land Japonica rice seeds

Aiming at the practical problems of rice seed identification, such as long inspection period and few varieties, this study proposed an efficient, fast and accurate identification method for rice seeds. Based on Raman spectroscopy, japonica rice seeds in cold region were identified rapidly and accurately. First, the raw spectra are pre-processed using Savitzky-Golay (SG),first derivative (1-Der),second derivative (2-Der),Adaptive Iterative Reweighted Penalized Least Squares Method (AIRPLS), and mean centered (MC) data pre-processing methods and their combinations, to investigate the effect of on the Partial least squares discrimination analysis (PLSDA) and the Support Vector Machine (SVM) model; secondly, use sample set partitioning based on joint x-y distance (SPXY) to divide the preprocessed data; Finally, the feature bands are extracted by combining the Successive Projections Algorithm (SPA ),Stepwise Regression (SR) and Competitive Adaptive Re-weighted Sampling (CARS) to compare and analyze the difference in modeling effect and detection time between feature bands and full bands under different pre-processing. The results show that among the 13 preprocessing methods, the AIRPLS+1-Der combination preprocessed better, and the accuracy of the verification set in the PLSDA and SVM models is above 95.45%; Among the three feature extraction methods, the feature band based on the CARS method achieved 96.97% accuracy among the two identification models, and the model ran fast, which indicated that the rapid and accurate identification of japonica rice seeds in cold areas based on Raman spectroscopy was feasible.