Abstract: An algorithm based on Support Vector Machine（SVM） and template matching is designed to classify the urinary sediment．Firstly，visible compositions in urinary sediment are classified roughly into big object and small object following their area，the former includes epithelial cells and pipe type，the latter includes crystallization，leukocyte and erythrocyte.Secondly，the template matching method is used to identify the crystallization，and by constructing an optimal SVM classifier，leukocyte and erythrocyte are different in the most extent．Finally，these big objects are classified into epithelial cells and pipe type following narrow extent.Experiment shows that the proposed method not only can gain better recognition rate（96.7%），but also reduces 22.4% computing time.

摘要： 提出了一种综合支持向量机（Support Vector Machine，SVM）和模板匹配的尿沉渣识别算法。首先根据面积特征将有形成分粗分成大目标类和小目标类，然后对小目标类中的草酸钙结晶以模板匹配法识别，而红、白细胞采用SVM的方法进行分类，最后对大目标类中的上皮细胞和管型则根据其狭长度加以区分。实验表明，该算法在将尿沉渣识别率提高到96.7%的同时还节约了22.4%的识别时间。