基于分数阶空谱联合Gabor特征的高光谱遥感图像分类
Fractional Gabor-based feature for hyperspectral image classification
王亚丽 1汤定定 2李丙春 1要秀宏 1贾森2
作者信息
- 1. 喀什大学计算机科学与技术学院,新疆喀什 844000
- 2. 深圳大学计算机与软件工程学院,广东深圳 518000
- 折叠
摘要
为充分考虑高光谱遥感图像的空谱结构特征,降低数据冗余,获取更具识别性的特征,提高分类精度.提出一种基于分数阶Gabor的高光谱图像分类方法,在分数域实现对局部信号的多分辨分析,以增强对高光谱图像的表征能力.首先,通过设置多阶正弦波构建多组分数阶Gabor滤波器,获得有效的特征表达.其次,对Gabor相位特征进行象限位编码,并通过汉明距离计算码距,降低计算复杂度.最后,融合不同阶的Gabor相位特征从而得到互补的纹理信息,以获取更高的分类性能.基于Trento真实数据集,选择3个分类样本进行训练,总体分类精度达到87.15％,Kappa系数为0.83,实验结果验证了该方法在小样本训练情况下的有效性,对比其他算法,提高了分类精度.
Abstract
In order to fully consider the spatial spectral structure features of hyperspectral remote sensing images,reduce data redundancy,obtain more recognizable features and improve classification accuracy.In this paper,we pro-pose a fractional-order Gabor-based hyperspectral image classification method,which implements multi-resolution a-nalysis of local signals in the fractional domain to enhance the characterisation of hyperspectral images.Firstly,a mul-tiple component fractional-order Gabor filter is constructed by setting up a multiple order sine wave to obtain an effec-tive feature representation.Secondly,the Gabor phase features are encoded by quadrant bits,and the code distance is calculated by Hamming distance,which reduces the computational complexity.Finally,the Gabor phase features of different orders are fused to obtain complementary texture information in order to obtain higher classification perform-ance.Based on the Trento real dataset,three classification samples were selected for training.The overall classifica-tion accuracy reached 87.15％,and the Kappa coefficient was 0.83.The experimental results have verified the effec-tiveness of this method in small sample training,and compared with other algorithms,it has improved classification ac-curacy.
关键词
高光谱图像分类/Gabor滤波器/分数阶滤波器Key words
hyperspectral image classification/Gabor filter/fractional-order filter引用本文复制引用
基金项目
高校科研计划青年项目(XJEDU2022P080)
喀什地区科技计划(KS2022083)
出版年
2024
NETL
NSTL
万方数据