PatchOut: A novel patch-free approach based on a transformer-CNN hybrid framework for fine-grained land-cover classification on large-scale airborne hyperspectral images

Renjie Ji, Kun Tan, Xue Wang, Shuwei Tang, Jin Sun, Chao Niu, Chen Pan

Abstract: Highlights•A semantic segmentation model is proposed for airborne HSI classification.•A Transformer and CNN hybrid approach is explored to enhance feature extraction.•A reduced Transformer is used to fuse multi-scale spatial-spectral features.•A new hyperspectral dataset, Qingpu-HSI, is introduced for HSI classification.