One Model Is Enough: Toward Multiclass Weakly Supervised Remote Sensing Image Semantic Segmentation

Abstract

Semantic segmentation of remote sensing images is effective for large-scale land cover mapping, which heavily relies on a large amount of training data with laborious pixel-level labeling. Weakly supervised semantic segmentation (WSSS) based on image-level labels has attracted intensive attention due to its easy availability. However, existing image-level WSSS methods for remote sensing images mainly focus on binary segmentation, which are difficult to apply to multiclass scenarios. This study proposes a comprehensive framework for image-level multiclass WSSS of remote sensing images, consisting of appropriate image-level label generation, high-quality pixel-level pseudo mask generation, and segmentation network iterative training. Specifically, a training sample filtering method, as well as a dataset cooccurrence evaluation metric, is proposed to demonstrate proper image-level training samples. Leveraging multiclass class activation maps, an uncertainty-driven pixel-level weighted mask is proposed to relieve the overfitting of labeling noise in pseudo masks when training the segmentation network. Extensive experiments demonstrate that the proposed framework can achieve high-quality multiclass WSSS performance with image-level labels, which can attain 94.23% and 90.77% of the IoUs from pixel-level labels for the ISPRS Potsdam and Vaihingen datasets, respectively. Beyond that, for the DeepGlobe dataset with more complex landscapes, the WSSS framework can achieve an accuracy close to 99% of the fully supervised case. Additionally, we further demonstrate that compared to adopting multiple binary WSSS models, directly training a multiclass WSSS model can achieve better results, which can provide new thoughts to achieve WSSS of remote sensing images for multiclass application scenarios. Our code is public at https://github.com/NJU-LHRS/OME.