Learning to detect cloud and snow in remote sensing images from noisy labels

TL;DR

This work tackles cloud and snow detection in remote sensing images under the challenge of noisy labels. It introduces a clean/Noisy dataset split $\mathcal{D}_{clean}$ and $\mathcal{D}_{noisy}$ and applies a curriculum-learning strategy that starts with accurate labels and progressively incorporates noisier samples, reducing overfitting to mislabeled regions. An evaluation framework is proposed to account for label noise, combining standard metrics on the clean set with manual tallies on the noisy set to reflect practical performance. Experiments with both UNet and Segformer demonstrate improved robustness and generalization to noisy annotations, highlighting the method's practical relevance for preprocessing remote sensing data. Overall, the paper establishes a foundation for learning with noisy labels in cloud and snow detection and provides actionable guidance for robust cloud-snow segmentation in heterogeneous satellite imagery.

Abstract

Detecting clouds and snow in remote sensing images is an essential preprocessing task for remote sensing imagery. Previous works draw inspiration from semantic segmentation models in computer vision, with most research focusing on improving model architectures to enhance detection performance. However, unlike natural images, the complexity of scenes and the diversity of cloud types in remote sensing images result in many inaccurate labels in cloud and snow detection datasets, introducing unnecessary noises into the training and testing processes. By constructing a new dataset and proposing a novel training strategy with the curriculum learning paradigm, we guide the model in reducing overfitting to noisy labels. Additionally, we design a more appropriate model performance evaluation method, that alleviates the performance assessment bias caused by noisy labels. By conducting experiments on models with UNet and Segformer, we have validated the effectiveness of our proposed method. This paper is the first to consider the impact of label noise on the detection of clouds and snow in remote sensing images.

Figures (1)

• Figure 1: Our proposed curriculum learning-based learning paradigm for cloud and snow detection from noisy labels. The red boxes represent the position of the noisy labels.