多层次信息融合在铁谱图像磨粒识别中的应用

Application of multi-level information fusion for wear particle recognition of ferrographic images

针对铁谱图像磨粒识别中异类信息综合利用率较低的问题,提出多层次信息融合的铁谱图像磨粒识别方法。首先,在铁谱图像二值化分割的基础上进行二值滤波,结合彩色铁谱图的R、G、B三分量,实现铁谱图像的彩色滤波。其次,以实际采集的磨粒图像样本为例,提取滤波后二值图像的形态特征,以及滤波后彩色图像的颜色特征;在特征层利用PCA对异类特征进行维数约简,并结合SVM和k-fold交叉验证,实现形态特征和颜色特征的特征层融合;在决策层将异类特征的SVM概率输出结果作为D-S证据理论的基本概率分配函数,实现形态特征和颜色特征的决策层融合。通过与形态学滤波结果对比,验证了本文提出滤波方法的优越性;其次,不同层次的信息融合结果表明,与单独使用颜色特征和形态特征相比,异类信息融合后可实现优势互补,有效提高故障磨粒的识别准确率。

Aiming at the insufficient utilization of the heterogeneous information in wear particle recognition of ferrographic images, a method for wear particle recognition based on multi-level information fusion was proposed. First, the binary filtering was conducted for the binary segmented ferrograhpic image, and the red, green and blue components of color ferrographic images were extracted to obtain the color filtered ferrographic images. Then, the experimental ferrographic images were collected as processing objects, the morphological features and color features of ferrographic imagesare were extracted from filtered binary images and filtered color images, respectively. PCA was utilized to reduce dimensions, and k-fold cross-validation and Support Vector Machine were combined to fuse different information in feature-level. The probabilistic output of SVM was used as the basic probability assignment of D-S information fusion, and the morphological information and color information were fused in decision-level. The superiority of proposed filtering method was demonstrated by comparing with the morphological filtering results. In addition, the multi-level information fusion results show that, compared with the use of color features and morphological features alone, the fusion of heterogeneous information can achieve complementary advantages and effectively improve the recognition accuracy of the fault wear particles.