基于近红外高光谱成像的单籽粒小麦品种分类研究

Classification of single wheat grain varieties based on near-infrared hyperspectral imaging

以22个品种的小麦为研究对象,利用近红外高光谱成像系统获取小麦籽粒的高光谱图像数据,采用图像分割技术确定小麦高光谱图像的感兴趣区域(ROI),提取小麦900~1700 nm的光谱区间共256个波段的光谱数据。利用遗传算法(GA)对光谱数据进行5次优化,分别建立256、114、57、35、26、13维的特征向量矩阵。利用未归一化与归一化后的数据分别建立灰狼优化算法(GWO)和支持向量机(SVM)的分类模型。结果表明:用1100个训练样本和440个测试样本对模型进行训练和检验分类,准确率分别为87.50%、93.18%;利用近红外高光谱成像技术对单籽粒小麦品种进行分类鉴别是可行的。

With 22 varieties of wheat as the research object,the near-infrared hyperspectral imaging system was used to obtain the hyperspectral image data of wheat grains.The region of interest(ROI)of the wheat hyperspectral image was determined by image segmentation technology,and the spectral data of 256 wave bands were extracted from the 900-1700 nm spectral interval of wheat.Genetic algorithm(GA)was used to optimize the spectral data for five times,and 256,114,57,35,26 and 13 dimensional eigenvector matrices were established respectively.The classification models of grey wolf optimization(GWO)algorithm and support vector machine(SVM)were established using the non normalized and normalized data.The results showed that the accuracy of training and testing the model with 1100 training samples and 440 test samples was 87.50%and 93.18%,respectively.It was feasible to classify and identify single grain wheat varieties by using near-infrared hyperspectral imaging technology.