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

Abstract As a World Heritage site and a Globally Important Agricultural Heritage site, Honghe Hani Rice Terraces urgently need efficient spatial distribution mapping for heritage protection, ecological management and agricultural sustainability. However, fragmented morphology and complex boundaries lead to confusion of terraces with the background in remote sensing imagery, while accurate boundary delineation is difficult to achieve. Therefore, a spatial-semantic interaction-guided network (SSIGNet) is proposed for high-resolution remote sensing images. A spatial-semantic interaction fusion module (SSIFM) is constructed, and a boundary guidance module (BGM) is introduced to extract terraces with different scales and morphological features. Compared with nine representative methods, SSIGNet achieves the optimal performance on the Yuanyang Terraces dataset, with the intersection over union (IoU) measure reaching 87.44%. The results of this study are expected to provide technical support for the use and conservation of Honghe Hani Rice Terraces.