焦炭光学组织图像的特征提取及分类识别

Characteristics extraction and classification recognition of coke optical texture images

焦炭光学组织结构直接决定焦炭的质量。目前,焦炭光学组织含量测定主要是利用光学显微镜进行采集,操作过程较复杂,人工识别可能带来误差。基于图像分析方法,对焦炭光学组织图像中的各个显微组织结构进行特征截取。利用RGB像素值、颜色矩、LBP算法进行图像特征提取,并分别采用K-近邻、支持向量机和随机森林3种数学模型进行分类识别。精确度对比结果表明,在颜色矩提取的图像下,利用支持向量机模型计算方法识别焦炭光学组织的精度可达90%。

The optical texture of coke directly determines the quality of coke.At present,the determination of optical texture content of coke is mainly collected by optical microscope,the operation process is complicated,and the manual identification may bring errors.Based on the image analysis,each microstructure in the optical image of the coke is subjected to characteristics interception.RGB pixel value,Color Moments and Local Binary Pattern(LBP algorithm)are used for extracting their images characteristics.K-Nearest Neighbor,Support Vector Machine and Random Forest are used for classification and identification respectively.The accuracy results show that the image characteristics extraction method based on Color Moment,using the Support Vector Machine model calculation method to identify the coke optical texture can reach 90%.