QKQingchen Kang

Qingchen Kang

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Qingchen Kang has a total of 7 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Cross-Domain Remote Sensing Scene Classification for Balancing Transferability and Discriminability in Partial Domain Adaptation

    by Huiyu Mu, Qingchen Kang, Liming Zhou, Feng Wang, Yadi Wang, Xianyu Zuo, Zechao Li, Yi Xie

    Due to the high cost of annotating remote sensing images, unsupervised partial domain adaptation (PDA) that do not require manual annotation have been widely studied for application in remote sensing scene classification (RSSC) tasks. PDA is based on the assumption that the class space of target domain is a subset of the source domain. Therefore, the model trained on the large-scale and full-labeled source domain is transferred to small-scale and unlabeled target domain effectively. However, most existing PDA methods for RSSC focus only on improving the transferability of features in the domain alignment process, neglecting the discriminative information that is equally important for cross-domain classification tasks. In this article, we proposed a Synergistic Contrastive-Adversarial Learning Partial Domain Adaptation (SCAPDA) method for adaptive cross-domain RSSC. The SCAPDA is built upon domain adversarial network, incorporating a discrimination constraint module (DCM) tailored for the target domain and a supervised contrastive learning module (SCLM) operating on both the source and target domains. The DCM learns a target-specific classifier under the constraint of enhancing discriminability in the target domain, thereby providing high-confidence supervision signals for supervised contrastive learning. The SCLM leverages supervised contrastive learning to explore both domain-invariant and class-specific information across source and target samples, aiming to promote transferable representations while preserving discriminability. The effectiveness of the proposed method is validated through experiments on three publicly available remote sensing datasets.