The empirical ridgelet transform has the ability of direction selectivity and adaptive decomposition. 2 DPCA can directly use the original image toconstruct the covariance matrix. Combined with the advantages of Empirical ridgelet transform and 2 DPCA, anidentification method ofmetal fracture based onempirical ridgelet-2 DPCA. At the same time, the proposed method is compared with the Ridgelet-2 DPCA, Ridgelet-PCA recognition method. The experimental results show that the bidimensional intrinsic mode Function（BIMF） component has more abundant feature information than ridgelet coefficient. 2 DPCA has more complete image structure informationthan PCA. Therefore, the proposed empirical Ridgelet-2 DPCA has achieved better recognition effect than experience Ridgelet-PCA and Ridgelet-2 DPCA recognition method.