基于铁谱图像异类特征融合的磨损类型识别方法

Wear Type Recognition Method Based on Heterogeneous Feature Fusion of Iron Spectrum Images

针对铁谱磨粒图像识别中存在特征单一、异类特征的综合利用率低等问题,提出一种磨粒图像多特征的异类信息融合识别方法。首先,对在线铁谱图像预处理基础上提取磨粒纹理(ASM、熵、相关、对比度)、颜色(均值、方差、斜度)、几何(7个不变矩)3种统计特征;其次,对提取特征数据进行[0,1]归一化处理,采用超球心间距法确定核参数,运用超球多类SVM实现基于单种特征的多类磨损识别;最后,在单种特征识别基础上通过后验概率构造3种特征所需的软判决基本概率赋值(BPA)函数,运用超球多类SVM与D-S证据理论结合法实现异类特征融合的铁谱图像识别。特征融合方法识别最高识别率达到了96.1%,与单一特征识别结果相比,识别准确度更高,且实现了不同特征的互补。

Aimed at the problems of single feature and low comprehensive utilization of the features of iron spectrum abrasive image recognition,a heterogeneous information fusion recognition method for multi-features of abrasive image was proposed.Firstly,based on the pretreatment of the online ferrography images,three statistical features of abrasive grain texture(ASM,entropy,correlation,contrast),color(mean,variance,slope),and geometry(7 invariant moments)were extracted.Secondly,the feature data was extracted and normalized by[0,1],the kernel parameters were determined by hypersphere spacing method,and multi-class SVM based on single feature was used to realize multi-class wear recognition based on single feature.Finally,based on single feature recognition,the soft decision basic probability assignment(PBA)function required to three features was constructed through posterior probabilities,by using hypersphere multi-class SVM and DS evidence theory combination method,the heterogeneous feature fusion of ferrography image recognition was achieved.The highest recognition rate of the feature fusion method reaches 96.1%,and compared with the single feature recognition result,it has higher recognition accuracy and can realize the complementary of different features.