.png){kind=icon}
.png){kind=icon}
Return
Enhanced hyperspectral image classification via Spatial-structure encoding and subpixel aggregation
.png){kind=icon}
PRE
AI
DOI：10.1016/j.optlastec.2025.114427.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Hyperspectral image classification plays a crucial role in various fields, yet extracting spatial structural features and subpixel-level details remains challenging. This paper introduces a dual-branch network that integrates Spatial-structure Encoding and Subpixel Aggregation (SESA) to address these challenges by enhancing spatial feature representation and subpixel-level detail extraction. The Spatial-structure Encoding (SSE) module captures structural context by computing self-similarity among neighboring pixels, while the Subpixel Aggregation (SPA) module enhances edge and texture details through subpixel convolution and aggregation. The outputs of both branches are weighted and fused, then fed into a fully connected layer for final prediction. Extensive experiments on four public HSI datasets (Indian Pines, Pavia University, WHU-Hi-HanChuan, and WHU-Hi-HongHu) demonstrate that SESA achieves superior performance with overall accuracies of 97.21 %, 99.16 %, 97.77 %, and 98.45 %, respectively, outperforming several state-of-the-art methods including HybridSN, SSRN, GAHT, ABLSTM, SSFTT, MASSFormer, and GLMGT, while maintaining low computational cost. The source codes are available at: https://github.com/Maike-Feng/SESA .