Objective To address the high missed detection rate of defective fruits and poor grading accuracy in current post-harvest fragrant pear sorting.Methods On the basis of an existing visual inspection system for fragrant pears, a machine vision detection system integrating an improved YOLOv8s algorithm and multi-dimensional morphological features is designed. An ECA attention module is introduced to enhance the model sensitivity to subtle defects, enabling fast and precise localization and segmentation of pears. Conventional morphological feature extraction is employed to conduct multi-dimensional shape analysis for identifying and rejecting defective fruits. Pressure-sensitive sensors are then used to grade the remaining qualified pears, realizing comprehensive quality screening and grading.Results The proposed system significantly reduces the missed detection rate and improves overall sorting efficiency. The improved YOLOv8s model achieves the single-target segmentation detection rate of 93.7%, the defect detection accuracy of 94.8%, and the grading accuracy of 97.8%.Conclusion The detection method integrating deep learning with machine vision effectively meets the practical requirements of post-harvest fragrant pear sorting.