面向对象的航空高光谱图像混合分类方法
作者简介:李雪轲(1989-),女,硕士生,主要从事高光谱遥感识别分类等应用研究。E-mail:shakerlee88@126.com
收稿日期: 2014-04-23
要求修回日期: 2014-06-26
网络出版日期: 2014-11-01
基金资助
国家自然科学基金项目(41371362、41272364、41201348)
A Hybrid of Object-based and Pixel-based Classification Method with Airborne Hyperspectral Imagery
Received date: 2014-04-23
Request revised date: 2014-06-26
Online published: 2014-11-01
Copyright
传统的高光谱分类通常仅考虑单一像元的光谱或纹理特征,分类后容易出现地物破碎的现象。鉴于此,本文提出了一种面向对象的混合分类方法,将面向对象的分割结果与传统的像元级分类结果进行有机融合,充分利用对象的光谱特征和空间结构特征。在此基础上,引入了2种具体的混合分类方法,即多尺度分割的SVM分类和多波段分水岭分割的SVM分类。前者将地物光谱的可变性进行弱化处理,转化为多尺度均质对象单元进行分类;后者融入了地物的空间信息和形态学特征,对分割得到的同质区域进行分类。将这2种分类方法应用于航空高光谱数据,实验结果表明:面向对象的混合分类方法的总体精度分别为92.63%和96.13%,与传统的像元级分类法相比,分别提高了10.14%和13.64%,有效地解决了分类后地物的破碎现象。
关键词: 航空高光谱; 面向对象图像分类; 支持向量机(SVM)
李雪轲 , 王晋年 , 张立福 , 杨杭 , 刘凯 . 面向对象的航空高光谱图像混合分类方法[J]. 地球信息科学学报, 2014 , 16(6) : 941 -948 . DOI: 10.3724/SP.J.1047.2014.00941
Hyperspectral imagery generally contains hundreds of contiguous narrow bands, which could provide detailed spectral information for target detection and image classification. Traditional hyperspectral classification fails to generate expected results since it simply considers spectral or textural properties at the pixel scale in the context of natural complexity. In this article, a hybrid classification method was proposed which takes full advantages of spectral and spatial features by fusing object-based segmentation results with traditional per-pixel classification results. Based on this concept, two specific hybrid classification approaches were employed: (1) the mixture of multi-scale segmentation and SVM classification and (2) the mixture of multi-band watershed segmentation and SVM classification. In the first proposed method, spectral variations were attenuated by converting them into homogenous image objects at multiple scales; while, the latter method aggregates spatial information and morphological profiles into the segmented objects to achieve the homogeneous classification. The two classification algorithms were applied to airborne hyperspectral imagery and the results show that the overall accuracy based on traditional pixel-wise classification reaches about 82.49%, relatively lower compared with the hybrid object-based classification methods, which are 92.63% (Method 1) and 96.13% (Method 2) respectively. In addition, Method 2 performs better than Method 1 since it produced a smoother boundary, partly because Method 2 needs less user-defined parameters, and the iterative “trial-and-error” of which may affect the classification results. In conclusion, this study demonstrates that the hybrid of object-based classification is a significantly more robust approach than the traditional per-pixel classifier. The proposed method overcomes the spectral confusion, solves the problem of land fragmentation, and provides a solution to map complex environments accurately.
Fig. 1 The flowchart of proposed method图1 算法流程图 |
Tab. 1 Parameters of SASI表1 SASI传感器参数 |
参数 | SASI-600 |
---|---|
光谱范围(μm) | 0.95~2.45 |
光谱分辨率(nm) | 15 |
波段数(个) | 101 |
瞬时视场角(°) | 0.068 |
视场角(°) | 40 |
每行像元数(个) | 640 |
量化水平(bits) | 14 |
Fig. 2 The study area and its typical ground features图2 研究区及典型地物图 |
Fig. 3 The spectral curves of samples图3 样本光谱曲线图 |
Tab. 2 Land cover classification scheme, overall sample size and training sample size表2 分类系统及样本点信息 |
类别 | ID | 类别定义 | 总体样本(像元个数) | 训练样本(像元个数) |
---|---|---|---|---|
PVC | 1 | 建筑板材、人造草皮等 | 1104 | 364 |
金属 | 2 | 屋顶材料 | 1052 | 347 |
沥青 | 3 | 屋顶材料、道路 | 1027 | 339 |
混凝土 | 4 | 屋顶材料、道路 | 1226 | 405 |
树木 | 5 | 不同种类的树木,主要分布于道路两侧、建筑物及水体周围等 | 1035 | 342 |
草地 | 6 | 天然草地,主要分布于公园、居民点周围等 | 1032 | 341 |
裸土 | 7 | 低植被覆盖区域,分布于公园、工地等 | 1177 | 388 |
水体 | 8 | 主要为环城水系 | 4349 | 1435 |
阴影 | 9 | 主要为阴影建筑,少量遮阴树,阴影路面等 | 2183 | 720 |
Fig. 4 Per-pixel SVM classification result图4 像元的SVM分类结果图 |
Tab. 3 Error matrix of per-pixel SVM classification result表3 像元的SVM分类混淆矩阵 |
类别 | 参照样本(像元个数) | 制图精度(%) | 用户精度(%) | |||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
PVC | 金属 | 沥青 | 混凝土 | 树木 | 草地 | 裸土 | 水体 | 阴影 | ||||
PVC | 740 | 0 | 0 | 2 | 0 | 0 | 0 | 0 | 0 | 100 | 99.73 | |
金属 | 0 | 615 | 0 | 43 | 0 | 0 | 0 | 0 | 0 | 87.23 | 93.47 | |
沥青 | 0 | 0 | 675 | 16 | 0 | 0 | 0 | 4 | 56 | 98.11 | 89.88 | |
混凝土 | 0 | 85 | 13 | 490 | 0 | 0 | 162 | 0 | 0 | 59.68 | 65.33 | |
树木 | 0 | 0 | 0 | 0 | 580 | 82 | 0 | 0 | 0 | 83.69 | 87.61 | |
草地 | 0 | 0 | 0 | 0 | 113 | 609 | 0 | 0 | 0 | 88.13 | 84.35 | |
裸土 | 0 | 0 | 0 | 270 | 0 | 0 | 627 | 0 | 0 | 79.47 | 69.51 | |
水体 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 2738 | 641 | 93.96 | 81.03 | |
阴影 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 172 | 766 | 52.36 | 81.66 | |
总体精度(%) | 82.49 | |||||||||||
kappa系数 | 0.79 |
Fig. 5 Multi-scale segmentation based SVM classification result图5 多尺度分割的SVM分类图 |
Tab. 4 Error matrix of multi-scale segmentation based SVM classification result表4 多尺度分割的SVM分类混淆矩阵 |
类别 | 参照样本(像元个数) | 制图精度(%) | 用户精度(%) | |||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
PVC | 金属 | 沥青 | 混凝土 | 树木 | 草地 | 裸土 | 水体 | 阴影 | ||||
PVC | 712 | 0 | 0 | 0 | 2 | 0 | 0 | 0 | 0 | 96.22 | 99.72 | |
金属 | 0 | 702 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 99.57 | 100 | |
沥青 | 23 | 0 | 675 | 14 | 0 | 5 | 0 | 75 | 76 | 98.11 | 77.76 | |
混凝土 | 3 | 0 | 13 | 765 | 0 | 0 | 26 | 0 | 0 | 93.18 | 94.80 | |
树木 | 0 | 0 | 0 | 0 | 645 | 19 | 0 | 5 | 0 | 93.07 | 96.41 | |
草地 | 0 | 0 | 0 | 0 | 46 | 667 | 0 | 0 | 0 | 96.53 | 93.55 | |
裸土 | 2 | 3 | 0 | 42 | 0 | 0 | 763 | 0 | 0 | 96.70 | 94.20 | |
水体 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 2616 | 128 | 89.77 | 95.34 | |
阴影 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 218 | 1259 | 86.06 | 85.24 | |
总体精度(%) | 92.63 | |||||||||||
kappa系数 | 0.913 |
Fig. 6 Multi-band watershed segmentation based SVM classification result图6 多波段分水岭分割的SVM分类图 |
Tab. 5 Error matrix of multi-band watershed segmentation based classification result表5 多波段分水岭分割的SVM分类混淆矩阵 |
类别 | 参照样本(像元个数) | 制图精度(%) | 用户精度(%) | ||||||||||||||||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
PVC | 金属 | 沥青 | 混凝土 | 树木 | 草地 | 裸土 | 水体 | 阴影 | |||||||||||||||||||||
PVC | 728 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 98.38 | 100 | ||||||||||||||||||
金属 | 0 | 648 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 91.91 | 100 | ||||||||||||||||||
沥青 | 12 | 54 | 675 | 0 | 4 | 0 | 0 | 0 | 86 | 98.11 | 61.93 | ||||||||||||||||||
混凝土 | 0 | 3 | 13 | 779 | 0 | 0 | 0 | 0 | 0 | 94.88 | 97.99 | ||||||||||||||||||
树木 | 0 | 0 | 0 | 0 | 652 | 19 | 0 | 5 | 0 | 97.25 | 97.17 | ||||||||||||||||||
草地 | 0 | 0 | 0 | 0 | 37 | 672 | 0 | 0 | 0 | 96.53 | 94.78 | ||||||||||||||||||
裸土 | 0 | 3 | 0 | 42 | 0 | 0 | 789 | 0 | 0 | 100 | 94.95 | ||||||||||||||||||
水体 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 2914 | 97 | 100 | 96.78 | ||||||||||||||||||
阴影 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1280 | 87.49 | 100 | ||||||||||||||||||
总体精度(%) | 96.13 | ||||||||||||||||||||||||||||
kappa系数 | 0.954 |
The authors have declared that no competing interests exist.
[1] |
|
[2] |
|
[3] |
|
[4] |
|
[5] |
|
[6] |
|
[7] |
|
[8] |
|
[9] |
|
[10] |
|
[11] |
|
[12] |
|
[13] |
|
[14] |
|
[15] |
|
[16] |
|
[17] |
|
[18] |
|
[19] |
|
[20] |
|
[21] |
|
[22] |
|
[23] |
|
[24] |
|
[25] |
|
[26] |
|
[27] |
|
[28] |
|
[29] |
|
/
〈 | 〉 |