[论文] Data Leakage in Patch-Based Hyperspectral Image Classification: Quanti...

论文概要

研究领域: CV
作者: Mohammed Q. Alkhatib
发布时间: 2026-10-06
arXiv: 2610.08770

中文摘要

基于块的学习通过利用局部光谱-空间信息来改善高光谱图像（HSI）分类，但从同一图像随机训练-测试采样可能导致空间块重叠，导致数据泄漏和乐观的性能估计。本文使用两种度量研究基于块的HSI分类中的同类训练-测试空间重叠：重叠百分比（OP）量化重叠测试块像素的全局量，平均重叠比率（AOR）测量受重叠影响的测试块之间的局部严重程度。在Pavia University数据集上的实验使用SVM、MLP、2D-CNN、3D-CNN、ViT和MorpMamba比较随机和非随机空间采样。结果表明，深度基于块的模型在随机采样下实现了高准确率（3D-CNN达到96.17%总体准确率），但在非随机空间采样下大幅下降，3D-CNN降至55.20%，ViT和2D-CNN分别下降40.71和38.81个百分点。块大小分析进一步表明，将块大小从5×5增加到19×19会将随机采样的重叠百分比从23.28%提高到77.02%。这些发现表明，随机的基于块的评估可能大幅虚增分类性能，特别是对于强烈利用空间上下文的模型。

原文摘要

Patch-based learning improves hyperspectral image (HSI) classification by exploiting local spectral-spatial information, but random train-test sampling from the same image can cause spatial patch overlap, leading to data leakage and optimistic performance estimates. This paper investigates same-class train-test spatial overlap in patch-based HSI classification using two measures: overlap percentage (OP), which quantifies the global amount of overlapped testing patch pixels, and average overlap ratio (AOR), which measures the local severity among affected testing patches. Experiments on the Pavia University dataset compare random and non-random spatial sampling using SVM, MLP, 2D-CNN, 3D-CNN, ViT, and MorpMamba. The results show that deep patch-based models achieve high accuracy under rando...

自动采集于 2026-10-08

#论文 #arXiv #CV #小凯