基于高光谱技术的籼稻霉变程度鉴别模型构建与优化

为解决快速无损鉴别籼稻霉变程度问题,利用高光谱技术采集200份霉变样本可见/近红外光谱信息,随机选取155份样本作为校正集,剩余45份作为验证集,根据预测浓度残差检验标准对校正集中异常样本进行剔除。以新校正集建立主成分线性判别分析(PCA-LDA)和簇类独立软模式法(SIMCA)模型,选用正确识别率为指标,优选最佳鉴别模型。并采用连续投影算法(SPA)提取特征波长,优化优选的最佳模型构建速度。研究结果表明,PCA-LDA对所有样本的误判总数为15,正确识别率为92.50%;SIMCA和SPASIMCA对所有样本的未能正确识别总数分别为6、2,正确识别率分别为97.00%、99.00%,并且经SPA筛选的变量数为20,仅占原始变量数的7.81%,建模时长缩短为原始变量的40.93%。因此,SPA-SIMCA鉴别效果最好,该方法在快速、准确鉴别籼稻霉变程度上具有可行性。

Establishment and optimization of identification model of the degree of moldy indica rice based on hyperspectral technology

In order to solve the problem of fast and nondestructive identification of moldy indica rice, the hyperspectral technique was used to collect the visible/near infrared spectroscopy of 200 moldy paddies, 145 samples were randomly chosen as calibration set, and 55 samples were chosen as validation set. According to the criterion of predicted concentration residual, the outlier samples of calibration set were eliminated. Then, the principal component analysis combined with linear discriminate analysis (PCA-LDA) and soft independent modeling of class analogy (SIMCA) were established by using the new calibration set, and, by comparing the correct recognition rate of the two models, the optimal model was elected. In order to improve the speed of establishing the optimal model, the successive projections algorithm ( SPA) was used to extract the characteristic wavelength. The results showed that 15 samples of the all samples were mistakenly identified by using the PCA-LDA, the correct recognition rate was 92.50%. The numbers of wrongly indentified samples of the all samples respectively were 6 and 2 by using the SIMCA and SPA-SIMCA, the correct recognition rate were 97.00% and 99.00%, respectively. 20 characteristic wavelengths were selected from 256 full wavelengths by SPA, the number of variables was dropped to 7.81% and the time of establishing model was reduced to 40.93% compared with initial variables. Therefore, the identification model established by the SPA-SIMCA was the best. The model could provide the technical support for fast and non-destructively identifying the moldy degree of the paddies, also could online identify whether the indica rice on the market mildews or not, and provide reference data for moldy degree.