Multimodal Remote Sensing Classification with Cascaded Attention Convolution Neural Network

EasyChair Preprint 8153

Abstract

Multimodal remote sensing classification tasks always encounter the data problem of unbalanced feature distributions from various information sources. In this paper, we adopt the attention mechanism with a cascaded multi-scale training strategy to enhance the performance feature extraction of one data source. We have utilized the hyperspectral and LiDAR data to provide the proposed algorithm's efficiency with multimodal Trento dataset. Finally, we have achieved better classification performance on the ground object categories with close similarity on height features owing to strengthening the feature extraction by our methodology.

Keyphrases: Cascaded Convolution Neural Network, Multimodal Remote Sensing Classification, attention module

@booklet{EasyChair:8153,
  author    = {Haotian Zhang and Li Ni and Min Huang},
  title     = {Multimodal Remote Sensing Classification with Cascaded Attention Convolution Neural Network},
  howpublished = {EasyChair Preprint 8153},
  year      = {EasyChair, 2022}}