菌落形态的计算机识别法用于菌种的分离筛选

SCREENING FOR STRAINS BY COLONY AUTOMATIC RECOGNITION METHOD

菌落形态是鉴别和分类菌种的重要特征之一。以分形和多重分形理论为基础，以计算机图像识别技术为手段，考察霉菌（绿僵菌）菌落形态的定量描述，分别测定各菌落样本的分形特征（覆盖维）和多重分形特征（多重分形谱）。研究表明，多重分形特征与菌种性能的相关性更大。以多重分形特征ａ－ｒｉｇｈｔ、ａ－ｗｉｄｔｈ和ｆ－ｓｔａｒｔ为依据设计的分类器，可以用于优良菌种的自动识别，速度快，与人工分离筛选的实验数据相吻合。

The morphology of microorganism colony is proven to be an important feature to classify and screen for strains. Based on Fractal and Mulhtifractal Theory, the quantitative description of the morphology images of mold (Metarhizium sp.) colonies was investigated by means of Computer Image Analysis Technique in this Paper. The fractal features (box dimension) and multifractal features (multifractal spectrum) were measured for each sample. The multifractal characteristics extracted were proven to be more related with the biotransformation activities of the strains. Moreover, we showed that the classifier designed depending on multifractal spectrum characteristics a-right. a-width and f-start, can be used for automated screening for fine strains. The result is found to be in good agreement with the experimental data of artificial screening with the much faster screening speed.