Multisource Remote Sensing Data Classification Based on Convolutional Neural Network

Multisource Remote Sensing Data Classification Based on Convolutional Neural Network

基于CNN的多源遥感数据分类

本文提出了two-branch CNN用于多源遥感数据,该结构通过提取HSI和other sources(LiDAR or VIS)特征,然后全连接层融合。
Multisource Remote Sensing Data Classification Based on Convolutional Neural Network
对于HIS branch:是一个dual-tunnel CNN,包含spectral(光谱)tunnel 和 spatial(空间)tunnel。光谱 tunnel 采用 1D conv ,空间 tunnel 采用 2D conv。再将两者输出特征放到全连接层,softmax分类器。
Multisource Remote Sensing Data Classification Based on Convolutional Neural Network
LiBAR or VIS branch:
采用 cascade block。
Multisource Remote Sensing Data Classification Based on Convolutional Neural Network
Multisource Remote Sensing Data Classification Based on Convolutional Neural Network
训练策略
两分支单独训练,然后融合在一起微调
Multisource Remote Sensing Data Classification Based on Convolutional Neural Network