基于3D-CAE的高光谱解混及小样本分类方法
|
黄 川(1973-),男,博士,副教授,主要从事基于深度学习的遥感数据处理与地理信息系统研究。Email: stephensky123@163.com。 |
Copy editor: 张仙
收稿日期: 2023-08-28
修回日期: 2024-09-01
网络出版日期: 2026-06-03
基金资助
国家自然科学基金项目“孟中缅印经济走廊公路网时空风险评估与归因”(42061074)
与“基于深度学习的高光谱图像红肉品质检测理论与技术”(61906140)
A hyperspectral unmixing and few-shot classification method based on 3DCAE network
Received date: 2023-08-28
Revised date: 2024-09-01
Online published: 2026-06-03
我国高光谱遥感技术的快速发展为开展大区域地物分类应用提供了充分保障。然而,如何在小样本下充分利用高光谱自身的空谱信息实现高精度的分类成为挑战。该文通过构建3D卷积自编码网络,以混合像元分解物理约束对模型进行引导,从而实现在准确估计端元丰度的同时获得对规则化的高光谱空谱特征的有效表达,结合支持向量机分类器实现在小样本条件下的高光谱分类。实验中,采用包括监督分类方法在内的多种传统高光谱图谱特征提取及分类方法进行对比验证,并对比了不同模型在不同采样率下的分类性能表现。实验结果表明,所提出的高光谱分类方法具有明显的精度优势,其中平均交并比(mean intersection over union,mIoU)达到0.829,相对于传统分类方法精度有明显提升; 在1/200采样率下mIoU值依然能接近0.8,优于同类方法,证实了该文方法在小样本条件下依然具有较好的鲁棒性,为解决小样本下高光谱分类问题提供了技术参考。
黄川 , 李雅琴 , 祁越然 , 魏晓燕 , 邵远征 . 基于3D-CAE的高光谱解混及小样本分类方法[J]. 自然资源遥感, 2025 , 37(1) : 8 -14 . DOI: 10.6046/zrzyyg.2023260
The rapid development of hyperspectral remote sensing technology in China fully ensures the effective application of large-scale surface feature classification. However, achieving high-precision classification under few-spot conditions while fully leveraging hyperspectral spatial-spectral information remains challenging. This study developed a 3D convolutional autoencoder (3D-CAE) network guided by physical constraints from mixed pixel decomposition. This approach enables accurate estimation of endmember abundance while effectively expressing regularized spatial-spectral features of hyperspectral data. In combination with a support vector machine (SVM) classifier, the method achieves hyperspectral classification under few-spot conditions. The classification performance of various models was evaluated at different sampling rates. To validate the proposed method, this study conducted experiments including comparisons with traditional hyperspectral feature extraction and classification methods, such as supervised classification approaches. The classification performance of various models was also evaluated at different sampling rates. The experimental results demonstrate that the proposed hyperspectral classification method has a significant advantage of accuracy, achieving a mean intersection over union (mIoU) of 0.829, which was close to 0.8 even at a low sampling rate of 1/200, surpassing its counterparts. These results confirm that the proposed method exhibits robustness under few-spot conditions. This study provides a valuable technical reference for addressing hyperspectral classification challenges under few-spot conditions.
图2 基于3D-CAE的高光谱分类流程Fig.2 The hyperspectral classification workflow based on 3D-CAE |
表1 网络组成及相关参数设置Tab.1 Network composition and related parameter settings |
| 特征层 | 卷积核 大小 | 卷积核 数量/个 | 激活 函数 | 特征 尺寸 |
|---|---|---|---|---|
| Conv3D-1 | (1, 1, 7) | 32 | ReLU | (1, 1, 155, 32) |
| Conv3D-2 | (1, 1, 7) | 16 | ReLU | (1, 1, 148, 16) |
| Conv3D-3 | (1, 1, 7) | 8 | ReLU | (1, 1, 141, 8) |
| Conv3D-4 | (1, 1, 7) | 2 | ReLU | (1, 1, 134, 2) |
| Flatten | — | — | — | 134×2 |
| Dense-1 | — | 32 | ReLU | 32 |
| Dense-2 | — | 6 | Softmax | 6 |
| Dense-3 | — | 162 | ReLU | 162 |
表2 模型预测不同端元的丰度值与真实值Tab.2 Model prediction of the abundance from different endmembers compared to the true values |
| 类型 | 沥青 | 草地 | 树木 | 屋顶 | 金属 | 土壤 |
|---|---|---|---|---|---|---|
| 预测丰度 | ![]() | |||||
| 真实值 | ||||||
| RMSE | ||||||
| 图例 | ![]() | |||||
图3 不同模型预测分类结果与真值对比图Fig.3 Comparison chart of classification results predicted by different models versus ground truth |
表3 不同模型预测精度对比表Tab.3 Comparison table of prediction accuracy for different models |
| 模型 | 总体分 类精度 | 精确度 | 召回率 | F1得分 | mIoU |
|---|---|---|---|---|---|
| 3D-CAE-SVM | 0.927 | 0.936 | 0.883 | 0.905 | 0.829 |
| PCA-SVM | 0.912 | 0.928 | 0.854 | 0.884 | 0.796 |
| MNF-SVM | 0.895 | 0.913 | 0.847 | 0.873 | 0.779 |
| FCLS-SVM | 0.850 | 0.832 | 0.775 | 0.778 | 0.657 |
| SVM | 0.916 | 0.929 | 0.856 | 0.885 | 0.799 |
| SAM | 0.847 | 0.870 | 0.831 | 0.821 | 0.777 |
| SID | 0.712 | 0.652 | 0.771 | 0.625 | 0.513 |
| [1] |
杜培军, 谭琨, 夏俊士. 高光谱遥感影像分类与支持向量机应用研究[M]. Beijing: Science Press, 2012,1-160.
|
| [2] |
|
| [3] |
刘银年, 孙德新, 胡晓宁, 等. 高分五号可见短波红外高光谱相机设计与研制[J]. 遥感学报, 2020, 24(4):333-344.
|
| [4] |
|
| [5] |
|
| [6] |
|
| [7] |
|
| [8] |
|
| [9] |
|
| [10] |
|
| [11] |
|
| [12] |
|
| [13] |
|
| [14] |
|
| [15] |
|
| [16] |
|
| [17] |
|
| [18] |
|
| [19] |
|
| [20] |
|
| [21] |
|
| [22] |
韩竹, 高连如, 张兵, 等. 高分五号高光谱图像自编码网络非线性解混[J]. 遥感学报, 2020, 24(4):388-400.
|
| [23] |
朱玲, 秦凯, 李明, 等. 改进的堆栈稀疏自编码矿物高光谱端元识别研究[J]. 光谱学与光谱分析, 2021, 41(4):1288-1293.
|
| [24] |
朱治青, 苏远超, 李朋飞, 等. 双约束深度卷积网络的高光谱图像空谱解混方法[J]. 信号处理, 2023, 39(1):128-142.
|
| [25] |
|
| [26] |
|
| [27] |
|
| [28] |
|
| [29] |
|
| [30] |
|
| [31] |
|
| [32] |
杜培军, 夏俊士, 薛朝辉, 等. 高光谱遥感影像分类研究进展[J]. 遥感学报, 2016, 20(2):236-256.
|
/
| 〈 |
|
〉 |