Zhu Q, Deng W, Zheng Z, et al. A Spectral-Spatial-Dependent Global Learning Framework for Insufficient and Imbalanced Hyperspectral Image Classification[J]. IEEE Transactions on Cybernetics, 2021. code [pdf]

Paper:

Summary

1. a spectral-spatial dependent global learning (SSDGL) framework based on global convolutional long short-term memory (GCL) and global joint attention mechanism (GJAM) is proposed for insufficient and imbalanced HSI classification.
2. in SSDGL, the hierarchically balanced(H-B) sampling strategy and the weighted softmax loss are proposed to address the imbalanced sample problem.
3. the GCL module is introduced to extract the long-short-term dependency of spectral features to effectively distinguish similar spectral characteristics of land cover types.
4. the GJAM module is proposed to extract attention areas, learning the most most discriminative feature representations.

Problem

• to extract the deep spectral-spatial features and solve the sample problem of insufficiency and imbalance.

Two-dimensional t-SNE visualization of features. Data distributions of the labeled samples in the original feature space(the first row) and the convolutional feature space (the second row).