Survey on Remote Sensing Data Augmentation: Advances, Challenges, and Future Perspectives

EasyChair Preprint 7704

Abstract

Deep learning-based methods have shown great progress in remote sensing applications. The performance of such methods can significantly outperform traditional remote sensing methods under the condition of the availability of large datasets for training. Unfortunately, some RS tasks, such as the change detection task, lack large established datasets. This issue is due to the limited access to some remote sensing data and the absence of a sucient labeled dataset. Data augmentation techniques are generally used to tackle this issue by increasing the number of samples and enhancing the quality of the training data. These techniques have shown performance improvement for general data and have recently been applied to remote sensing data. The present survey synthesizes the recent data augmentation works contributed to the remote sensing field. It briefly describes data-level issues, existing data augmentation techniques used to address these issues, and challenges facing these techniques. This survey provides the reader with an idea about the influence of data augmentation techniques on the performances of deep learning models, especially while using a small amount of data.

Keyphrases: GANs, change detection, data augmentation, deep learning, remote sensing

@booklet{EasyChair:7704,
  author    = {Amel Oubara and Falin Wu and Abdenour Amamra and Gongliu Yang},
  title     = {Survey on Remote Sensing Data Augmentation: Advances, Challenges, and Future Perspectives},
  howpublished = {EasyChair Preprint 7704},
  year      = {EasyChair, 2022}}