Ensemble Alignment Subspace Adaptation Method for Cross-Scene Classification

Proceedings Paper Geosciences, Multidisciplinary

CROSS-SCENE HYPERSPECTRAL IMAGE CLASSIFICATION BASED ON CYCLE-CONSISTENT ADVERSARIAL NETWORKS

Zhihao Meng et al.

Summary: In this paper, a novel model called cycle auxiliary classifier generative adversarial network (Cycle-AC-GAN) is proposed to address the challenge of lack of labeled training samples in hyperspectral image classification. The model establishes connections between source and target scenes, allowing the classification of target scenes to benefit from the labeled samples in source scenes.

2022 IEEE INTERNATIONAL GEOSCIENCE AND REMOTE SENSING SYMPOSIUM (IGARSS 2022) (2022)

添加到收藏夹

Article Geochemistry & Geophysics

Deep Ensemble CNN Method Based on Sample Expansion for Hyperspectral Image Classification

Shuxian Dong et al.

Summary: This article studies a deep ensemble CNN method based on sample expansion for HSI classification, which tackles the problem of insufficient training samples and improves classification accuracy by extracting spatial information and expanding the number of training samples using pixel-pair features.

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING (2022)

添加到收藏夹

Article Geochemistry & Geophysics

Cross-Scene Hyperspectral Image Classification With Discriminative Cooperative Alignment

Yuxiang Zhang et al.

Summary: The study introduces a new domain adaptive method, discriminative cooperative alignment (DCA) of subspace and distribution, to address the challenge of cross-scene HSI classification. Experimental results demonstrate that the DCA method significantly outperforms other domain-adaptive approaches.

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING (2021)

添加到收藏夹

Article Computer Science, Information Systems

Semi-supervised rotation forest based on ensemble margin theory for the classification of hyperspectral image with limited training data

Wei Feng et al.

Summary: An adaptive semi-supervised rotation forest (SSRoF) algorithm is proposed for the classification of hyperspectral images with limited training data, which increases the number of training instances by mining high-quality unlabeled samples and utilizes SMOTE to overcome class imbalance. The effectiveness of the proposed method is demonstrated on three real hyperspectral remote sensing datasets through comparisons with ensemble methods and semi-supervised methods.

INFORMATION SCIENCES (2021)

添加到收藏夹

Article Engineering, Electrical & Electronic