IEEE Geoscience and remote sensing letters
Dual-Path Network-Based Hyperspectral Image Classification
Abstract: for pixel-wise HSI classification, the CNN-based methods may not effectively extract the discriminative features from the complex scene in HSIs.
1. Introduction
ResNet enables feature reusage while DenseNet enables new feature exploration.
2 Proposed method
a. Dual-path Network
Residual learning enables the network to be extended to hundreds of layers by using skipping connections to discover deep features. However, the ResNet can reuse features implicitly, but it is not good at exploring new features. On the contrary, the densely connected network can explore new features continuously but causes feature redundancy. By designing two paths to combine two kinds of networks, the proposed HSDPN method can overcome both shortcomings and achieve a good classification performance.
- Path of the Residual Network
- Path of the Convolutional Network
- Dual-path Network
b.Hsi classification Based on DPN
这个图左半部分像残差连接,右半部分想密集连接,残差连接是把bottleneck layer放到残差线上,主线是没有结构的,密集连接是密集连接结构上没有任何结构。
3. Experiments
Data set
Salinas Valley 512*217*204 16
Houston image 349*1905*144 15
Three parameters the size of image patch,the band number of the dimension-reduced data,the number of groups in each grouped convolution layer
评价指标
OA,AA,k