高阶高斯滤波和Radon变换结合的缸体特征提取

Aiming at multiple-groove-like features with obvious orientation on cylinder bore surface, a new method of groove feature extraction by combining high-order Gauss regression filter pre-processing with Radon transform is proposed. The high-order robust Gauss regression filter suppresses the boundary effect so the whole measured surface data can be applied to the analysis, and the interference of extremum caused by groove features on filtering mean surface extraction is prevented. In the Radon transform process, the whole extraction of groove features is simplified to the extraction of the centre line of grooves to avoid the difficulty of feature extraction caused by the weakening of feature orientation in dense grooves. Considering the strong correlation between the distribution of the minimum points at bottom of groove and the centre line, the centre line location is determined by fitting the minimum points at the bottom of the deep groove. Shallow grooves and slight scratches also leave minimal points to interfere with the extraction of minimal points in deep grooves, so the depth threshold is introduced to distinguish the interference. Experimental results show that the main angles of grooves are 19.5°, 91.1° and 165.6°; the overall defect rate of the surface groove reaches 35.95%, larger than that of a single groove. Compared with the surface feature extracted from the two-dimensional contour line, the three-dimensional surface feature extracted with this method contains more feature information and has stronger robustness to the local extremum of the surface.