人工智能
模式识别(心理学)
计算机科学
特征提取
上下文图像分类
特征(语言学)
人工神经网络
计算机视觉
高光谱成像
图像(数学)
图像处理
卷积神经网络
统计分类
支持向量机
特征选择
图像分割
作者
Chunlan Yang,Dawei Xue
出处
期刊:
日期:2026-05-01
卷期号:: 180-183
标识
DOI:10.1109/iotaai70471.2026.11607724
摘要
Although graph convolutional networks (GCNs) excel at leveraging local neighborhood information through powerful aggregation mechanisms for hyperspectral classification, they fall short in acquiring global information. Transformer models possess great capacity to capture global interactions across spatial-spectral information, yet local features are prone to being neglected. To better leverage the strengths of GCNs and transformer models, we integrate these two architectures to establish local-global information interaction for hyperspectral image (HSI) classification, and propose a local-global feature extraction framework (LGFE). In detail, GCNs and convolutional neural networks (CNNs) are leveraged to acquire discriminative spatial-spectral information, with pooling operations applied to strengthen spatial features. Meanwhile, transformer blocks are incorporated to boost the acquisition of global information from HSI. A fully connected network then performs the final classification. Experimental evaluations show that LGFE achieves superior performance against all baseline models.
科研通智能强力驱动
Strongly Powered by AbleSci AI