Extraction method for edge artifact features of Si3N4 bearing roller microcracks based on two-dimensional discrete wavelet transform and Otsu threshold segmentation

Abstract

Abstract Aiming at the problem that the edge artifacts of Si3N4 bearing roller microcracks have low contrast, contain noise, and easily merge with the background, making it difficult to segment. Design a method based on a two-dimensional discrete wavelet transform and Otsu threshold segmentation. Realize the extraction of edge artifact features of Si3N4 bearing roller microcracks. Wavelet decomposition is used to remove noise, while wavelet reconstruction features are used to restore lost details. Create a discrete wavelet transform function equation in two dimensions that includes wavelet reconstruction and wavelet decomposition. Achieve contrast improvement and noise removal in edge artifact feature images. Aiming at the problem of artifacts existing at the edge of defects in images that are difficult to remove using conventional methods. A threshold segmentation function equation with the core idea of maximizing inter class variance is designed. Finish choosing the ideal threshold. In order to accomplish the goal of eliminating the edge artifact feature. The average PSNR of the image enhanced by point, line, and surface micro crack edge artifact features of the Si3N4 bearing roller is close to 62.69dB. The average SSIM is about 0.77. Increases the contrast of Si3N4 bearing roller microcrack edge artifact features in an efficient manner. Improved the effect of feature extraction of point, line, and surface micro crack edge artifacts in Si3N4 bearing rollers.