Unsupervised Adversarial Domain Adaptation with Error-Correcting Boundaries and Feature Adaption Metric for Remote-Sensing Scene Classification

Abstract

Unsupervised domain adaptation (UDA) based on adversarial learning for remote‐sensing scene classification has become a research hotspot because of the need to alleviating the lack of annotated training data. Existing methods train classifiers according to their ability to distinguish features from source or target domains. However, they suffer from the following two limitations: (1) the classifier is trained on source samples and forms a source‐domain‐specificboundary, which ignores features from the target domain and (2) semantically meaningful features are merely built from the adversary of a generator and a discriminator, which ignore selecting the domain invariant features. These issues limit the distribution matching performance of source and target domains, since each domain has its distinctive characteristic. To resolve these issues, we propose a framework with error‐correcting boundaries and feature adaptation metric. Specifically, we design an error‐correcting boundaries mechanism to build target‐domain‐specific classifier boundaries via multi‐classifiers and error‐correcting discrepancy loss, which significantly distinguish target samples and reduce their distinguished uncertainty. Then, we employ a feature adaptation metric structure to enhance the adaptation of ambiguous features via shallow layers of the backbone convolutional neural network and alignment loss, which automatically learns domain invariant features. The experimental results on four public datasets outperform other UDA methods of remote‐sensing scene classification.

Semi-supervised bidirectional alignment for Remote Sensing cross-domain scene classification

“…To verify the advantages of SSDA to SSL/UDA in RS cross-domain scene classification, our BSCA is compared with: (1) FixMatch ( Xiong et al, 2021 , Sohn et al, 2020 ), one classical SSL method, which was applied to semi-supervised RS scene classification and achieved one of the state-of-the-art results in this field; and (2) ECB-FAM ( Ma et al, 2021b ), one of the latest UDA algorithms for RS cross-domain scene classification. For more comprehensive evaluation, our BSCA is further compared with some SSDA methods reproduced in this study, including: (3) S+T , which is the basic method only using source data and labeled target data for training; (4) ADDA ( Wang et al, 2018 , Tzeng et al, 2017 ), which is the first work on RS-SSDA to our knowledge.…”

“…The best results are in bold . Type Method A N R N W N N A R A W A N R A R W R N W A W R W Mean SSL FixMatch ( Xiong et al, 2021 ) 81.3 95.0 63.5 97.6 84.4 UDA ECB-FAM ( Ma et al, 2021b ) 88.5 74.5 83.2 92.5 77.5 92.8 66.9 68.9 65.4 91.7 97.6 79.9 81.6 SSDA ...…”

“…In practice, the mixture of these factors further complicates domain shifts. On the other hand, some unsupervised domain adaptation (UDA) methods have been explored for RS cross-domain scene classification ( Ma et al, 2021b , Zheng et al, 2022b , Yu et al, 2022 , Ahmed et al, 2021 , Wurm et al, 2019 ), aiming to utilize the existing labeled RS image dataset (source domain) to help the classification of the unlabeled RS image dataset (target domain) within a shared class space. However, the generalization of UDA is limited due to the lack of labeled target data, which is essential for reducing the domain shift as a bridge between the source domain and the target domain.…”

Triple Loss Adversarial Domain Adaptation Network for Cross-Domain Sea–Land Clutter Classification

“…2) DA. The selected DA-based DG methods include deep domain confusion (DDC) [15], deep adaptation network (DAN) [16], DANN [32], error-correcting boundaries mechanism with feature adaptation metric (ECB-FAM) [41], and TLADAN [6]. Among these, DDC and DAN can be considered as discrepancy-based methods; DANN can be regarded as an adversarial-based method; and ECB-FAM and TLADAN can be viewed as discrepancyand adversarial-based methods.…”

BiFDANet: Unsupervised Bidirectional Domain Adaptation for Semantic Segmentation of Remote Sensing Images

“…For example, Gaofen satellites can capture a large number of satellite images with high spatial resolution on a large scale. In remote sensing, such a large amount of data has offered many more capability for image analysis tasks; for example, semantic segmentation [1], change detection [2] and scene classification [3]. Among these tasks, the semantic segmentation of remote sensing images has become one of the most interesting and important research topics because it is widely used in many applications, such as dense labeling, city planning, urban management, environment monitoring, and so on.…”