KnowCL: A Universal Knowledge Embedded Contrastive Learning framework for hyperspectral image classification

doi:10.1016/j.knosys.2026.116972

KnowCL: A Universal Knowledge Embedded Contrastive Learning framework for hyperspectral image classification

2026-09-07

0

PRE

AI

DOI：10.1016/j.knosys.2026.116972

Original

Original request for help

Save

Abstract

En 中文

Hyperspectral image (HSI) classification techniques have been intensively studied, and a variety of models have been developed. However, these HSI classification models are limited to lightweight models and random sampling methods of partitioning the datasets. The former constrains the model’s generalization performance, and the latter leads to inflated model evaluation metrics, which result in plummeting model performance in the real world. Therefore, we propose a universal Knowledge-Embedded Contrastive Learning (KnowCL) method for HSI classification that learns invariant features in the absence of globally distributed training samples. We present a new HSI processing pipeline, along with a range of data transformation and augmentation techniques that yield diverse data representations. The proposed framework based on this pipeline is compatible with supervised, semi-supervised, and unsupervised learning. The semi-supervised version can fully exploit labeled and unlabeled samples with the expected training time. Furthermore, we designed a new loss function that adaptively fuses supervised and unsupervised losses, thereby enhancing learning performance. This proposed new classification paradigm shows great potential in exploring HSI classification technology on disjoint sampling. The code can be accessed at https://github.com/quanweiliu/KnowCL .

Keywords:

Hyperspectral image (HSI) classification

Contrastive learning (CL)

Semi-supervised learning

KnowCL: A Universal Knowledge Embedded Contrastive Learning framework for hyperspectral image classification

Abstract

Journal

Organization

Cited Papers