.png){kind=icon}
.png){kind=icon}
Return
Adaptive Context Transformer for Semisupervised Remote Sensing Image Segmentation
.png){kind=icon}
PRE
AI
DOI：10.1109/TGRS.2023.3318788.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Current deep learning methods for semantic seg-mentation in remote sensing heavily depend on a substantial amount of labeled data. However, obtaining pixel-level labeled data in this field is both time-consuming and laborious. To address this challenge, semi supervised learning (SSL) method shave been introduced. Pseudo supervision is one of the most effective methods, which can be adopted to enhance the performance of semsupervised semantic segmentation of remotesensing images. But incorrect pseudolabels can cause substantialdistortions to the segmentation model in SSL. Moreover, it isdifficult for conventional semantic segmentation methods to dealwith global-local features of the remote sensing image withoutadaptive context feature. In this article, we propose a novel learn-ing approach based on an adaptive context transformer (ACT)and pseudolabeling, called ACT for semisupervised (ACTSS)remote sensing image segmentation. We propose an adaptivecontext attention model with adjustable sliding windows. A smallwindow is used to capture Query (Q) for local feature, andbigger windows are used to capture Key (K) and Value(V)forglobal feature. Then, we combine them and get the global-local features. And we propose a point-line-plane (PLP) pseudolabelfilter mechanism based on clustering and boundary extraction, which can filter unreliable pseudolabels from three angles: point,line, and plane. To validate the effectiveness of the model, we carried out extensive experiments on the LOVEDA, Potsdam, and Vaihingen datasets and compared ACTSS with other methods. These experiments demonstrate that ACTSS achieves state-of-the-art performance for semisupervised semantic segmentation on all tested datasets.
Keywords:
Adaptive context transformer (ACT)
pseudolabel filter
semantic segmentation
semisupervised learning (SSL)
Journal
IF:
8.6
Papers:
2.1W
Citations:
10.7W