Spatial-semantic interaction guided network for mapping Honghe Hani Rice Terraces via remote sensing

A novel spatial-semantic interaction-guided network (SSIGNet) achieves unprecedented accuracy in mapping the UNESCO-listed Honghe Hani Rice Terraces, with an intersection over union (IoU) score of 87.44%. This breakthrough addresses longstanding challenges posed by fragmented terrace morphology and complex boundaries, which have hindered precise remote sensing analysis. The SSIGNet integrates a spatial-semantic interaction fusion module (SSIFM) and a boundary guidance module (BGM), enabling high-resolution delineation of terraces across varying scales and morphological features. Outperforming nine other methods on the Yuanyang Terraces dataset, this approach offers critical technical support for heritage protection, ecological management, and agricultural sustainability in one of China's most iconic cultural landscapes.

Zheng, S., Huang, L., Su, B., Zhang, B., & Tang, B. (2026). Spatial-semantic interaction guided network for mapping Honghe Hani Rice Terraces via remote sensing. Npj Heritage Science, 14(1). https://doi.org/10.1038/s40494-026-02767-2