Abstract: In order to solve the problem that the traditional single image denoising algorithm based on machine vision has poor effect on the mixed noise signal processing, resulting in the inability to effectively detect and identify the surface defects of Si₃N₄ cylindrical roller used in aero-engine, a visual detection method based improved coupled denoising algorithm and multi-scale threshold segmentation algorithm is proposed. The surface defect image of Si₃N₄ cylindrical roller is denoised by the optimized wavelet threshold denoising algorithm and the improved median filter algorithm, and the multi-scale threshold segmentation algorithm is used to segment the defect image, thus identifying and extracting Si₃N₄ surface defects of cylindrical rollers. The experimental results show that the signal-to-noise ratio of surface defect images of Si₃N₄ cylindrical rollers denoised by the improved coupling denoising algorithm is more than 24.5%, and the detection and recognition accuracy rate of the multi-scale threshold segmentation algorithm for surface defect images of Si₃N₄ cylindrical rollers is more than 94%. It proves that the visual detection method has a good image denoising effect and a certain versatility.

Key words: machinevision, Si₃N₄cylindricalroller, couplingdenoising, surfacedefect, multi-scalethresholdsegmentation