【摘要】 为考察近红外光谱对玉米种子的品种识别与产地识别性能,采集了8个玉米品种波长范围为12 000~4 000 cm-1的近红外光谱数据,并基于此数据研究了基于PCA的光谱数据特征的提取方法,并探讨了神经网络(ANN)和支持向量机模型(SVM)在品种识别上的性能,进一步研究了玉米品种的产地识别技术,且比较了传统可见光图像的品种识别。研究发现:基于近红外的玉米品种识别,在6个主分量的情况下整体上性能达到90%以上;SVM算法较ANN算法稳定可靠,更适合于小样本情况下的光谱分析;基于光谱的品种识别与基于可见光图像的品种识别效果相当;另外发现同一品种在不同产地上其光谱特征差别较大,据此可以应用光谱进行产地鉴别,鉴别力达到95%以上。本研究所构建的方法对玉米品种识别和产地识别具有积极意义。更多还原

【Abstract】 In order to investigate the varieties and origin identification performance maize kernel near infrared spectroscopy,spectral data( 12 000 ~ 4 000 cm-1) of 8 maize varieties have been collected in the paper. Based on the data,feature extraction method by PCA algorithm have been studied first; the variety identification performance of neural network( ANN) and support vector machine model( SVM) algorithms have been discussed; origin identification of maize has been further studied; the traditional visible image identification of varieties have been compared. Research shows that: on the condition of 6 principal components,the recognition rate is above 90% of these algorithms based on maize’s near infrared spectral data; SVM algorithm expresses better than ANN algorithm on stability and reliability, while the former is more superior to small sample based on spectral analysis; the effect of recognition by spectral and by visible light image is almost similar. In addition,there are larger differences found on spectral of same variety from different producing areas,which can be applied for producing area identification by spectroscopy,and the discriminability reaches over 95%. The method described in the paper has a positive meaning for maize variety and origin identification.更多还原