基于显微图像模式识别的粉丝品质检测研究

Micrograph-and pattern recognition-based examination of starch-noodle quality

粉丝由植物淀粉加工而成.鉴别粉丝中淀粉的种类及成分比例对食品的营养与安全具有重要而现实的意义.目前商检中主要依靠感官评价,可靠性差.为此提出以图像处理、模式识别、人工神经网络为基础的粉丝计算机自动分类和检测新方法.以粉丝组织的显微图像为基础,运用灰度共生矩阵和分形理论提取显微图像的特征,并将提取出来的特征用做神经网络淀粉品质分类的输入,建立了粉丝中淀粉品质的自动检测系统.实验结果表明,该方法可行,结果较好.

Starch-noodle （vermicelli） is made from cereal-starches, and the quality of which rests basically with ingredients of grain and starch. To examine the ingredients and proportion of diversified starches in commercial starch-noodle is important for food safety and nourishment. At present, the inspection and classification of components in starch-noodle mainly depend on vision and sensory perception, which is not crediable. An approach based on image processing, pattern recognition and artificial neural networks is proposed to automatically inspect and classify the starch-noodles by computer system. The automatic inspection system is based on the micrograph of starch-noodle structure. The micrograph characters are extracted by grey level co-occurrence matrix and fractional theory, then the extracted characters are used as the input of neural network for starch-noodle evaluation. The result of experiment shows that it is practicable and effective.