CAJ | 학술논문

目的：依托于Matlab环境，初步实现BP（Back Propagation）神经网络对尿沉渣图像中有形成分的自动识别与分类。方法首先应用灰度化、直方图增强、邻域滤波和中值滤波等方法对图像进行预处理；再利用Canny算子和Sobel算子叠加处理进行边缘检测，通过膨胀腐蚀和孔洞填充等操作得到有形成分的连通域信息，提取出每个连通域的周长、面积、长宽比、矩形度、圆形度等12个特征值作为BP神经网络的输入；最后利用BP神经网络创建学习训练过程，对每个连通域即有形成分进行分类。结果采用该自动分类方法得到了尿沉渣图像中有形成分的种类和数目。结论该方法分类准确，可实现尿沉渣图像中有形成分的自动识别与分类。
목적：의탁우Matlab배경，초보실현BP（Back Propagation）신경망락대뇨침사도상중유형성분적자동식별여분류。방법수선응용회도화、직방도증강、린역려파화중치려파등방법대도상진행예처리；재이용Canny산자화Sobel산자첩가처리진행변연검측，통과팽창부식화공동전충등조작득도유형성분적련통역신식，제취출매개련통역적주장、면적、장관비、구형도、원형도등12개특정치작위BP신경망락적수입；최후이용BP신경망락창건학습훈련과정，대매개련통역즉유형성분진행분류。결과채용해자동분류방법득도료뇨침사도상중유형성분적충류화수목。결론해방법분류준학，가실현뇨침사도상중유형성분적자동식별여분류。
Objective To classify the tangible components in urinary sediment images automatically through application of BP (Back Propagation) neural network on basis of Matlab. Methods The urinary sediment images were preprocessed with the methods of graying, histogram enhancement, neighborhood ifltering, median ifltering and so on. Then, the Canny and Sobel operators were applied to perform edge detection. The information of connected domains for the tangible components were obtained through expansion corrosion and hole iflling, from which 12 characteristic values including the perimeter, area, aspect ratio, rectangle and round degree were extracted as the input of BP neural network to classify the tangible components in urinary sediment images. Results The type and quantity of the tangible components in urinary sediment images were obtained with the application of this automatic classification method. Conclusion The automatic classification method made it possible to precisely identify and classify the tangible components in urinary sediment images.