| 摘要: |
| 得益于遥感领域先进算法的发展，高光谱图像分类（Hyperspectral Image Classification，HSIC）近年来取得了显著进展。然而，高光谱数据的高维特性以及有限的标注样本仍然制约了许多现有方法的效果。为了解决这些问题，提出了一种新颖的分类模型：SpiralMamba。其来源于近期的Mamba模型，该模型因其具有线性复杂度的高效全局特征提取能力而广受认可。SpiralMamba包含3个主要模块：螺旋扫描嵌入（Spiral Scanning Embedding，SSE）模块最大限度地减少将图像转换为序列时空间信息的损失；高斯掩膜加权（Gaussian Mask Weighting，GMW）模块增强了中心像素周围特征的权重，从而提升了提取特征的可分类性；轻量级Mamba模块（Lightweight Mamba Module，LWM）旨在减少模型参数和计算需求，使得该模型适合于样本稀缺的高光谱图像分类任务。在Indian Pines、WHU-Hi-HanChuan和Houston2018数据集上的实验结果表明，SpiralMamba模型分类总体准确率分别达到93.10%、93.49%、91.21%。 |
| 关键词: 高光谱图像分类 螺旋扫描嵌入(SSE) 高斯掩膜加权(GMW) 轻量级Mamba模块(LWM) |
| DOI：10.20079/j.issn.1001-893x.241031001 |
|
| 基金项目:辽宁省教育厅项目（JYTMS20230243） |
|
| SpiralMamba:a Lightweight Mamba Network for Hyperspectral Image Classification |
| BAI Yu,WU Haoqi,ZHANG Lili,GUO Hanlin |
| (School of Electronic and Information Engineering，Shenyang Aerospace University，Shenyang 110136，China) |
| Abstract: |
| Hyperspectral image classification(HSIC) has advanced significantly in recent years，driven by the development of advanced algorithms in remote sensing.However，the high-dimensional nature of hyperspectral data and the limited availability of labeled samples remain significant challenges，hindering the effectiveness of many existing methods.To address these limitations，the authors propose SpiralMamba，a novel classification framework inspired by the recent Mamba model，renowned for its efficient global feature extraction with linear complexity.SpiralMamba comprises three main modules:Spiral Scanning Embedding(SSE) module minimizes the loss of spatial information when converting images into sequences；Gaussian Mask Weighting(GMW) module enhances the weight of features surrounding the central pixel，thereby improving the classifiability of the extracted features；Lightweight Mamba Module(LWM) reduces model parameters and computational demands，making this model suitable for hyperspectral image classification tasks with scarce samples.Experimental results on the Indian Pines，WHU-Hi-HanChuan and Houston2018 datasets demonstrate that the overall classification accuracies of the proposed SpiralMamba model reach 93.10%，93.49%，and 91.21%，respectively. |
| Key words: hyperspectral image classification spiral scan embedding(SSE) Gaussian mask weighting(GMW) lightweight Mamba module(LWM) |
《电讯技术》编辑部 