<p>Szechuan pepper (<i>Zanthoxylum bungeanum</i>) is a traditional Asian spice whose flavor quality is closely linked to the morphology of oil glands on the pericarp. Current quality assessment relies on subjective manual inspection and expert experience, lacking a quantitative and rapid analytical approach. To address this gap, this study aims to establish a multi-dimensional feature system for recognizing and quantifying pericarp surface morphology using reconstructed 3D point clouds integrated with deep learning. A dataset of 315 red and green Szechuan pepper pericarp samples (RSPP and GSPP) from major cultivation regions was constructed through multi-view imaging combined with structure-from-motion and multi-view stereo techniques. A seven-parameter feature system was designed to quantify oil-gland morphology, including gland count, average size, coverage ratio, curvature, and spatial uniformity. Furthermore, an enhanced PointNet + + model was developed to improve segmentation performance. Quantitative morphological analysis revealed that GSPPs had more glands (14–28) of smaller size (659–1008 points), while RSPPs had fewer glands (12–20) of larger, more variable size (835–1606 points) and a less uniform spatial distribution (maximum coefficient of variation in nearest-neighbor distance: 0.502 for RSPPs vs. 0.416 for GSPPs). The enhanced PointNet + + model achieved an overall accuracy of 0.869 and a mean Intersection over Union (mIoU) of 0.673 for GSPP, with corresponding values of 0.857 and 0.595 for RSPP. These mIoU scores represent increases of 9.7% and 4.6%, respectively, over the baseline PointNet + + model. This work provides a reliable machine vision-based methodology for morphological quality evaluation and potential grading of Szechuan pepper.</p> Graphical Abstract <p></p>