奇异品质大米的外观特征分布研究

针对现有分类方法不能判断散粒体的整体类别及其奇异性问题，研究大米奇异品质特征分布规律，提出基于特征分布的散粒体奇异性识别方法。以大米的颜色特征为例，分别估计纯类别样本、不同比例的二元混合样本的特征分布，作为该特征在对应混合比的目标分布，建立以特征分布为基本样本、混合类别及混合比为类标签的大米颜色特征分布数据集。对被试样本，估计它的特征分布，分别计算它与各个目标分布的误差，将其划分为取得最小误差的目标分布所对应的类别。实验结果表明，所提出的方法能够识别被试的奇异性及奇异样本的混合比，分类准确率和召回率达到100%。

Study on Distribution of Appearance Features of Rice with Quality Singularity

Current classification methods can''t judge the whole quality and singularity of tested samples. In order to solve the problem, the feature distribution law of rice quality was studied. A method based on feature distribution was proposed. Taking the color characteristics of rice as an example, the feature distribution of pure category samples and binary mixed samples with different proportions were estimated respectively. They were the target feature distributions. Taking feature distribution as basic sample, the mixing category and the mixing ratio as the class label, the classification dataset was established. For samples to be tested, estimate their feature distribution, calculate its error with each target distribution respectively, and divide it into the category corresponding to the target distribution with the minimum error. The experiment results show that the proposed method can identify the singularity and the mixture ratio of singularity, and the classification accuracy and recall rate reach 100%.