A surface defect detection method based on feature and morphology reconstruction was proposed, aiming at the problems of high misdetection rate of water stain residue and low efficiency of manual re-inspection in machine vision detection of defects in shaft parts surface. The high-speed industry line scanning image of the shaft part was preprocessed, and image segmentation was completed based on the improved threshold iteration algorithm. Defect image was extracted by removing background, noise, and interference. A surface defect feature model of shaft parts based on envelope contour of curve cluster was established. Combining the area, the proportion of area and coarseness of each connected area in the segmented image, a logistic regression classifier was trained to recognize the typical surface defects of pits, cracks, and pits of shaft parts. Combining with image depth information, defect morphology reconstruction was carried out to eliminate pseudo-defects such as water stain, so as to improve the robustness of surface defect detection of shaft parts. The experimental results show that this method is effective for surface defect detection of shaft parts with high defect recognition rate and robustness. The proposed method has an average recognition time of 3.69 seconds and a weighted recognition rate of 98.86%, which can accurately identify three kinds of typical defects and pseudo-defects.