.png){kind=icon}
.png){kind=icon}
Return
Generative Data-Driven Dynamic Deep-Learning Classifier Selection for Hyperspectral Image Classification
.png){kind=icon}
OA
AI
DOI：10.1109/JSTARS.2025.3646066.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Deep learning has emerged as a critical paradigm in hyperspectral image (HSI) classification, addressing the inherent challenges posed by high-dimensional data and limited labeled samples. However, traditional methods often struggle with dynamic adaptability and computational efficiency, particularly under diverse and complex data distributions. This article aims to address two key challenges: 1) how to dynamically select the most appropriate classifiers for high-dimensional and limited-label scenarios, and 2) how to optimize hyperparameter configurations to improve the generalization and accuracy of homogeneous classifiers. To address these limitations, this article proposes a dynamic deep-learning classifier selection framework consisting of two key modules: the generative-based dynamic selection module for heterogeneous classifiers (GDS-HDLC) and the generative configuration selection module for homogeneous classifiers (GCS-HDLC). GDS-HDLC leverages generative models to extend validation datasets, enabling robust classifier selection via a multisource evaluation mechanism that integrates real and synthetic data. Meanwhile, GCS-HDLC optimizes hyperparameter configurations for homogeneous classifiers, enhancing generalization and classification accuracy. Specifically, the generative models simulate diverse data distributions, enriching the validation process with synthetic samples, while the multisource evaluation framework balances real and generated data to improve performance. These methods are evaluated across multiple HSI datasets, demonstrating superior classification accuracy, efficiency, and adaptability compared to state-of-the-art methods. Notably, GDS-HDLC achieved an overall accuracy of 90.68% on the Botswana dataset, while GCS-HDLC attained 91.87%, surpassing baseline models.
Keywords:
Deep learning
dynamic classifier selection (DCS)
generative models
hyperspectral image (HSI) classification
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
5.3
Papers:
1.3K
Citations:
3.0W
View more organizations