Weakly-Supervised Cloud Detection with Fixed-Point GANs

TL;DR

This paper introduces a weakly-supervised GAN-based method for satellite cloud detection that requires only image-level labels, achieving near fully-supervised performance with minimal pixel-level annotations.

Contribution

The paper presents Fixed-Point GAN for Cloud Detection (FCD), a novel weakly-supervised approach that translates images to detect clouds without pixel-level labels, and enhances it with FCD+ for improved accuracy.

Findings

Achieves near state-of-the-art performance with only image-level labels.

FCD+ matches fully-supervised methods using just 1% pixel labels.

Effective on Landsat-8 dataset, reducing labeling costs.

Abstract

The detection of clouds in satellite images is an essential preprocessing task for big data in remote sensing. Convolutional neural networks (CNNs) have greatly advanced the state-of-the-art in the detection of clouds in satellite images, but existing CNN-based methods are costly as they require large amounts of training images with expensive pixel-level cloud labels. To alleviate this cost, we propose Fixed-Point GAN for Cloud Detection (FCD), a weakly-supervised approach. Training with only image-level labels, we learn fixed-point translation between clear and cloudy images, so only clouds are affected during translation. Doing so enables our approach to predict pixel-level cloud labels by translating satellite images to clear ones and setting a threshold to the difference between the two images. Moreover, we propose FCD+, where we exploit the label-noise robustness of CNNs to refine the prediction of FCD, leading to further improvements. We demonstrate the effectiveness of our approach on the Landsat-8 Biome cloud detection dataset, where we obtain performance close to existing fully-supervised methods that train with expensive pixel-level labels. By fine-tuning our FCD+ with just 1% of the available pixel-level labels, we match the performance of fully-supervised methods.