Two-stage sparse coding of region covariance via Log-Euclidean kernels to detect saliency

A bearing fault diagnosis scheme with statistical-enhanced covariance matrix and Riemannian maximum margin flexible convex hull classifier

To achieve more appropriate fault feature representation for bearing, a statistical-enhanced covariance matrix (SECM) is proposed to extract the global-local features and the interaction of them. Besides, three statistical parameters are introduced to SECM to enhance its statistical characteristics. For fully mining the Riemannian geometric information embedded in SECMs, a Riemannian maximum margin flexible convex hull (RMMFCH) classifier with Log-Euclidean metric (LEM) is designed, where a set of Riemannian kernel mapping functions map SECMs to a higher-dimensional Hilbert space. In this space, the RMMFCH can be directly solved, which reduces the extra computation cost. Hence, we design a fault diagnosis scheme of bearing with SECM and RMMFCH. Experiment results prove the promising performance of our method for bearing fault diagnosis.