Pseudolabel-Based Unreliable Sample Learning for Semi-Supervised Hyperspectral Image Classification

Recently, pseudolabel-based deep learning methods have shown excellent performance in semi-supervised hyperspectral image (HSI) classification. These methods usually select high-confidence unlabeled samples to help optimize backbone classification networks. However, a large number of remaining low-c...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on geoscience and remote sensing 2023, Vol.61, p.1-16
Main Authors: Yao, Huaxiong, Chen, Renyi, Chen, Wenjing, Sun, Hao, Xie, Wei, Lu, Xiaoqiang
Format: Article
Language:English
Subjects:
Classification
Contrastive learning
Deep learning
Distribution functions
Feature extraction
Graphical models
hyperspectral image (HSI) classification
Hyperspectral imaging
Image classification
Learning systems
Machine learning
pseudolabel
Reliability
semi-supervised learning
Spatial distribution
Supervised learning
Training
Transformers
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Recently, pseudolabel-based deep learning methods have shown excellent performance in semi-supervised hyperspectral image (HSI) classification. These methods usually select high-confidence unlabeled samples to help optimize backbone classification networks. However, a large number of remaining low-confidence unlabeled samples, which contain rich land-covers information, are underused. In this article, we propose a pseudolabel-based unreliable sample learning (PUSL) method to fully exploit low-confidence unlabeled samples for semi-supervised HSI classification. First, to avoid overfitting the spatial distribution of labeled samples, we build a position-free transformer (PFT) as the backbone classification network. Second, PFT is initially trained with labeled samples in a supervised learning manner to obtain an initial classifier, which is then used to split unlabeled samples into reliable and unreliable unlabeled samples based on the predicted confidence. Third, reliable unlabeled samples participate in training along with labeled samples. Finally, unreliable unlabeled samples are treated as negative samples for the corresponding categories to improve the discrimination of PFT in a contrastive learning paradigm. Extensive experiments on three HSI datasets demonstrate that PUSL outperforms the compared methods.
ISSN:0196-2892
1558-0644
DOI:10.1109/TGRS.2023.3322558