A Data Augmentation Approach using WGAN with Grey Wolf Optimizer (GWO) to Improve Deep CNN Weak Supervised Classification of Hyperspectral Images

This paper proposes a new approach to improve weak supervised classification of deep CNN architectures based on data augmentation by integrating Grey Wolf Optimizer (GWO) with Wasserstein Generative Adversarial Network (WGAN). The model is applied for hyperspectral image classification (HSI). We have chosen a weak supervised CNN classifier, where only $5 \%$ of hyperspectral training data is authentic, the remaining $95 \%$ ones being virtually generated using the model GWO-WGAN. The performance of the proposed model is evaluated using Indian Pines, Pavia University, and Salinas hyperspectral image datasets.