Journal of Geo-information Science >
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
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.
LI Xueke , WANG Jinnian , ZHANG Lifu , YANG Hang , LIU Kai . A Hybrid of Object-based and Pixel-based Classification Method with Airborne Hyperspectral Imagery[J]. Journal of Geo-information Science, 2014 , 16(6) : 941 -948 . DOI: 10.3724/SP.J.1047.2014.00941
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] |
|
/
〈 |
|
〉 |