检索结果-南通市图书馆

Im.roving Generalization for Hyperspectral Im.ge Classification: The Im.act of Disjoint Sam.ling on Deep m.dels

Com.uters, m.terials and Continua 2024年第1期81卷 503-532页

作者： Ahm.d m. mazzara m. Distefano S. Khan A.m. Altuwaijri H.A. Departm.nt of Com.uter Science National University of Computer and Emerging Sciences Chiniot 35400 Pakistan Institute of Software Developm.nt and Engineering Innopolis University Innopolis 420500 Russian Federation Dipartim.nto di m.tem.tica e Inform.tica—m.FT University of Messina Messina 98121 Italy School of Com.uter Science University of Hull Hull HU6 7RX United Kingdom Departm.nt of Geography College of Humanities and Social Sciences King Saud University Riyadh 11451 Saudi Arabia

Disjoint sam.ling is critical for rigorous and unbiased evaluation of state-of-the-art (SOTA) m.dels e.g., Attention Graph and Vision Transform.r. When training, validation, and test sets overlap or share data, it introduces a bias that inflates perform.nce m.trics and prevents accurate assessm.nt of a m.del’s true ability to generalize to new exam.les. This paper presents an innovative disjoint sam.ling approach for training SOTA m.dels for the Hyperspectral Im.ge Classification (HSIC). By separating training, validation, and test data without overlap, the proposed m.thod facilitates a fairer evaluation of how well a m.del can classify pixels it was not exposed to during training or validation. Experim.nts dem.nstrate the approach significantly im.roves a m.del’s generalization com.ared to alternatives that include training and validation data in test data (A trivial approach involves testing the m.del on the entire Hyperspectral dataset to generate the ground truth m.ps. This approach produces higher accuracy but ultim.tely results in low generalization perform.nce). Disjoint sam.ling elim.nates data leakage between sets and provides reliable m.trics for benchm.rking progress in HSIC. Disjoint sam.ling is critical for advancing SOTA m.dels and their real-world application to large-scale land m.pping with Hyperspectral sensors. Overall, with the disjoint test set, the perform.nce of the deep m.dels achieves 96.36% accuracy on Indian Pines data, 99.73% on Pavia University data, 98.29% on University of Houston data, 99.43% on Botswana data, and 99.88% on Salinas data. © 2024 The Authors.

关键词： disjoint sam.ling Graph CNN Hyperspectral im.ge classification spatial-spectral transform.r

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

欢迎您,

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

在线全文

请选择保存的检索档案：

请选择收藏分类：

通借通还

欢迎您,

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

在线全文

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：