MSRC-based defective nanocrystalline soft magnetic ribbon detection

Abstract

The traditional manual inspection of nanocrystalline soft magnetic materials based on metallographic samples is a time-consuming and somewhat unreliable task. It is also difficult to achieve high accuracy by simply adopting existing automatic signal processing methods as an alternative. To address the issue, a novel automatic microscopic defect recognition method for nanocrystalline soft magnetic ribbon using high-resolution optical microscopic images is proposed. The target problem is viewed as a pattern recognition problem, in which images are classified as non-defective and defective. An effective and highly efficient random feature is used to describe the structures of the nanocrystalline soft magnetic ribbons. Then the extracted features are used to recognize defects via a modified sparse representation-based classifier (MSRC). In the experiment, two well-known features, LBP (local binary pattern) and PCA (principal component analysis), and different classifiers, SVM (support vector machine) and SRC (sparse representation classifier), are compared. The experimental results demonstrate that the proposed method can provide low error rates in recognizing ribbon defects.