Generating Views Using Atmospheric Correction for Contrastive
Self-supervised Learning of Multi-spectral Images
- Ankit Patnala ,
- Scarlet Stadtler ,
- Martin G Schultz ,
- Juergen Gall
Abstract
In remote sensing, plenty of multi-spectral images are publicly
available from various landcover satellite missions. Since labeling
remote sensing images is time-consuming, contrastive self-supervised
learning is commonly applied to unlabeled data but relies on
domain-specific transformations used for learning. When focusing on
vegetation, the current contrastive learning frameworks cannot apply
their transformations to the NIR channel, which carries valuable
information about the vegetation state. Therefore, we use contrastive
learning, relying on different views of unlabelled, multi-spectral
images to obtain a pre-trained model to improve the accuracy scores on
small-sized remote sensing datasets. This study presents the generation
of additional views tailored to remote sensing images using atmospheric
correction as an alternative transformation to color jittering. The
proposed transformation can be easily integrated with multiple channels
to exploit spectral signatures of objects. We found this proposed
alternative transformation to improve performance compared to the
baseline transformation.