4.7 Article

User-Guided Data Expansion Modeling to Train Deep Neural Networks With Little Supervision

期刊

出版社

IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/LGRS.2022.3201437

关键词

Image segmentation; Training; Annotations; Data models; Urban areas; Buildings; Semantics; Aerial image annotation; deep learning; semantic segmentation; semi-automatic data annotation

资金

  1. Sao Paulo Research Foundation (FAPESP) [2014/12236-1]
  2. National Council for Scientific and Technological Development (CNPq) [303808/2018-7]
  3. Petroleo Brasileiro S. A. (PETROBRAS)
  4. Agencia Nacional do Petroleo, Gas Natural e Biocombustiveis (ANP) [4600556376, 4600583791]

向作者/读者索取更多资源

Image segmentation is a crucial and challenging task in remote sensing. This letter presents a user-guided data expansion modeling approach that reduces the workload of pixel-level data annotation by learning image marker features, resulting in significant improvement in the generalization performance of deep neural networks for image segmentation.
Image segmentation is a challenging and essential task in remote sensing. Deep neural networks (DNNs) have successfully segmented images from different domains, but the models usually require time-consuming and expensive pixel-level data annotation. In this letter, we exploit a recent technique to learn features (an encoder) from a few markers placed by the user in relevant image regions, build an encoder-decoder model from a small set of regions delineated by click-based segmentation, and use that model to annotate the remaining pixels. Such user-guided data expansion modeling can be repeated as the encoder-decoder network improves, and by selecting well-annotated regions, the user considerably expands the pixel set to train DNNs with little supervision. We show the role of feature learning from image markers (FLIM) and that our data expansion model can significantly improve the generalization performance of a state-of-the-art DNN when segmenting buildings in aerial images of distinct cities.

作者

我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。

评论

主要评分

4.7
评分不足

次要评分

新颖性
-
重要性
-
科学严谨性
-
评价这篇论文

推荐

暂无数据
暂无数据