基于多光谱图像的不同品种绿茶的纹理识别

Texture discrimination of different kinds of green tea based on multi-spectral imaging technique

为了提高茶叶加工的智能化水平,提出一种基于多光谱图像纹理分析的快速识别不同品种绿茶的方法.通过3CCD成像仪同时获得绿茶样本的红光、绿光和近红外三个通道的图像,采用灰度共生矩阵和纹理滤波相结合来提取图像纹理特征,分析了不同品种绿茶的各个通道图像的纹理特征.非监督聚类分析表明,基于组合方法提取的纹理特征优于仅依靠灰度共生矩阵得到的纹理特征.优化和筛选后得到10个特征参数作为支持向量机模型的输入,建立模式识别模型.结果表明,对于126个建模样本的识别正确率达到94.4%,对于未知64个预测样本的识别正确率达到93.8%,说明提出的组合纹理特征提取和模式识别方法能够较好地识别不同品种的绿茶.

For enhancing the intelligence level of tea processing, a method for discriminating images of four different kinds of green tea was put forward based on texture analysis. First, green tea images were obtained from three charged coupled device camera, which could simultaneously obtain three images from green, red and near-infrared channels. The statistical information derived from the combination of gray level co-occurrence matrix （GLCM） and texture filtering was used to describe the textural features of image. The results of unsupervised cluster analysis indicated that the combined textural features were more optimal than those from GLCM. Then ten optimized texture parameters were selected as input for least squares support vector machine （LS-SVM） classifier. This classifier got high recognition rate （94. 4%） for 126 samples in calibration set, and same excellent performance was obtained for discrimination of 64 unknown samples in prediction set with 93.8%. The result indicated that it is feasible to discriminate tea images of different kinds of green tea based on combined textural feature extraction and LS-SVM classifier.