This study investigates temple decorative patterns in the Huaihe River Basin and evaluates a morphology-based digital classification workflow that can support heritage archiving and design translation. Image samples of wood, stone, brick, tile, and painted motifs were collected from representative temple buildings in Anhui Province and organized into five mutually exclusive pattern classes according to the dominant visual motif. Six descriptors – shape factor, complexity, expansion, eccentricity, solidity, and shape ratio – were extracted after image preprocessing and used as inputs to a Levenberg-Marquardt-optimized back-propagation neural network. The model achieved an overall recognition rate of 90.88% on the reported test split, while descriptor analysis also showed stable responses under translation and rotation. Beyond classification, the study clarifies how motif digitization can support pattern archiving, textile and surface design translation, and regionally grounded cultural-product development.
Xiao et al. (Tue,) studied this question.