一种基于面积对称模型的土地利用/覆被真伪变化图斑检测方法
作者简介:高锡章(1976-),男,河南新县人,助理研究员,研究方向为GIS理论和方法、生态与环保GIS。E-mail: gaoxz@lreis.ac.cn
收稿日期: 2014-01-20
要求修回日期: 2014-06-13
网络出版日期: 2014-09-04
基金资助
国家“863”课题(2012AA12A405)
科技支撑计划项目(2012BAH33B01)
A New Method to Detect False Change Polygons in Land Use/Cover Change Map Based on Symmetric Theory Model
Received date: 2014-01-20
Request revised date: 2014-06-13
Online published: 2014-09-04
Copyright
不同时相土地利用/覆被数据间的空间配准误差,是产生土地利用/覆被伪变化图斑的一个主要原因。本文从土地利用/覆被原始图斑与其相邻的变化图斑间的空间关系角度,提出了由同一原始图斑产生的土地利用/覆被伪变化图斑的面积对称理论,设计和实现了针对因空间配准误差而导致的土地利用/覆被伪变化图斑自动化检测模型,并以内蒙古自治区通辽市奈曼旗1980年和2000年两期土地利用/覆被图对该模型进行了实验模拟。结果表明:当两期数据的空间配准误差不超过原始影像1个像元时,总体检测精度达到90%以上;误差不超过5个像元时,总体检测精度达到80%以上;误差不超过原始影像10个像元时,总体检测精度达到70%以上。同时,由于运行时唯一需设定的面积对称系数阈值可设为0.2-0.4(默认设为0.3),该检测模型可适用于由空间配准误差引起的伪变化图斑的自动检查,可满足由于空间配准误差所引起土地利用/覆被伪变化图斑剔除的需求。
高锡章 , 刘海江 , 李宝林 , 袁烨城 . 一种基于面积对称模型的土地利用/覆被真伪变化图斑检测方法[J]. 地球信息科学学报, 2014 , 16(5) : 784 -789 . DOI: 10.3724/SP.J.1047.2014.00784
Co-registration error between two land use maps based on different dates can cause a considerable overestimation of the land use/cover change. Even a small amount of misregistration markedly reduces the accuracy of land cover change estimates. Without relevant information about misregistration, existing methods cannot work effectively to detect and eliminate the false changes caused by misregistration. In this paper, we propose a methodology (Symmetric Theory) from the viewpoint of the relationship between original land use polygon and the changed polygons to detect the false change caused by misregistration. Symmetric Theory presents that the area of ‘changing from’ and ‘changing to’ polygons overlaid from the original polygon is symmetric in some degree, if true change polygons are eliminated from the changing polygons. Based on this theory, an automated detecting model is designed and developed. A case study was conducted using this method based on two land cover maps from 1980 and 2000, and their simulated misregistration maps for Naiman County, Tongliao City, Inner Mongolia, China (a total area of 8137.6 km2). This study shows that this method can effectively discriminate the spurious land cover changes from true land cover changes with false change detection accuracy ranging from 75.0% to 87.4%, true change detection accuracy ranging from 71.2% to 93.8%, and overall detection accuracy ranging from 73.3% to 92.7%. However, with the image shifts from half to ten pixels (15m to 300m), the ability of detecting false changes decreases with the increase of image misregistration. And when using this method, the SI threshold should be set as 0.2-0.4. If no relevant knowledge is mentioned, 0.3 is the best choice.
Fig.1 Example of false change polygons caused by misregistration图1 空间配准误差导致的伪变化图斑示例 |
Fig.2 Flow chart of false change detection model based on Symmetric Theory Model图2 基于面积对称模型的伪变化检测流程图 |
Fig.3 The Land-Use/Cover Map of Naiman County in 1980 and 2000图3 奈曼旗1980年和2000年土地利用/覆被图 |
Tab.1 False change detection accuracy with different SI threshold(%)表1 不同SI阈值下伪变化检测精度(%) |
误差 (像元) | SI阈值 | ||||||||
---|---|---|---|---|---|---|---|---|---|
0.1 | 0.2 | 0.3 | 0.4 | 0.5 | 0.6 | 0.7 | 0.8 | 0.9 | |
0.5 | 86.30 | 88.51 | 89.13 | 89.56 | 89.75 | 89.85 | 89.95 | 90.13 | 90.26 |
1 | 84.41 | 86.65 | 87.40 | 87.72 | 87.99 | 88.15 | 88.36 | 88.47 | 88.69 |
2 | 82.88 | 84.97 | 85.96 | 86.23 | 86.67 | 86.84 | 87.05 | 87.15 | 87.29 |
3 | 81.14 | 84.38 | 85.27 | 85.90 | 86.15 | 86.51 | 86.65 | 86.79 | 86.85 |
4 | 80.72 | 84.23 | 85.49 | 86.06 | 86.42 | 86.69 | 86.91 | 86.98 | 87.05 |
5 | 77.11 | 82.21 | 83.57 | 84.46 | 84.92 | 85.32 | 85.46 | 85.58 | 85.66 |
6 | 73.92 | 80.10 | 82.19 | 83.06 | 83.60 | 83.89 | 84.16 | 84.31 | 84.41 |
7 | 69.51 | 77.72 | 80.45 | 81.50 | 81.94 | 82.37 | 82.69 | 82.83 | 82.99 |
8 | 66.40 | 74.69 | 78.72 | 80.03 | 80.56 | 80.94 | 81.34 | 81.51 | 81.75 |
9 | 61.10 | 73.42 | 77.22 | 78.86 | 79.60 | 80.10 | 80.50 | 80.67 | 80.95 |
10 | 58.05 | 70.97 | 74.96 | 76.45 | 77.31 | 77.86 | 78.31 | 78.61 | 78.84 |
Tab.2 True change detection accuracy with different SI threshold(%)表2 不同SI阈值下真变化检测精度(%) |
误差 (像元) | SI阈值 | ||||||||
---|---|---|---|---|---|---|---|---|---|
0.1 | 0.2 | 0.3 | 0.4 | 0.5 | 0.6 | 0.7 | 0.8 | 0.9 | |
0.5 | 97.32 | 96.48 | 96.35 | 96.17 | 95.96 | 95.65 | 95.45 | 94.67 | 93.05 |
1 | 94.77 | 93.97 | 93.77 | 93.32 | 93.19 | 92.65 | 92.51 | 91.68 | 89.93 |
2 | 90.79 | 90.03 | 89.56 | 88.48 | 87.83 | 87.42 | 86.97 | 86.15 | 84.55 |
3 | 87.51 | 86.06 | 85.16 | 84.4 | 83.53 | 82.71 | 82.24 | 81.05 | 79.18 |
4 | 84.52 | 82.14 | 80.41 | 79.41 | 78.75 | 78.24 | 77.62 | 76.35 | 73.83 |
5 | 83.32 | 80.19 | 78.36 | 77.24 | 76.28 | 75.53 | 74.92 | 73.71 | 70.77 |
6 | 81.48 | 78.08 | 75.69 | 74.69 | 73.41 | 72.74 | 72.09 | 70.45 | 67.68 |
7 | 81.03 | 76.74 | 74.38 | 72.68 | 71.71 | 70.70 | 69.49 | 68.32 | 66.40 |
8 | 80.59 | 76.02 | 72.84 | 71.11 | 70.02 | 69.21 | 67.61 | 66.36 | 63.34 |
9 | 80.68 | 74.85 | 71.89 | 69.98 | 68.52 | 67.56 | 65.92 | 64.31 | 61.48 |
10 | 80.10 | 74.38 | 71.18 | 69.33 | 67.96 | 67.06 | 65.81 | 63.74 | 60.75 |
Tab.3 Overall detection accuracy with different SI threshold (%)表3 不同SI阈值下总体检测精度(%) |
误差 (像元) | SI域值 | ||||||||
---|---|---|---|---|---|---|---|---|---|
0.1 | 0.2 | 0.3 | 0.4 | 0.5 | 0.6 | 0.7 | 0.8 | 0.9 | |
0.5 | 95.24 | 95.69 | 95.64 | 95.52 | 95.35 | 95.08 | 94.91 | 94.23 | 92.78 |
1 | 92.16 | 92.70 | 92.66 | 92.34 | 92.28 | 91.87 | 91.78 | 91.12 | 89.71 |
2 | 88.52 | 88.53 | 88.58 | 87.83 | 87.50 | 87.25 | 86.99 | 86.44 | 85.34 |
3 | 85.19 | 85.44 | 85.52 | 84.94 | 84.48 | 84.10 | 83.85 | 83.14 | 81.98 |
4 | 82.93 | 83.02 | 83.54 | 82.20 | 81.96 | 81.78 | 81.51 | 80.81 | 79.37 |
5 | 80.47 | 81.12 | 81.76 | 80.56 | 80.25 | 80.02 | 79.76 | 79.16 | 77.60 |
6 | 77.78 | 79.07 | 79.87 | 78.79 | 78.40 | 78.20 | 78.00 | 77.24 | 75.87 |
7 | 75.12 | 77.24 | 77.49 | 77.21 | 76.96 | 76.69 | 76.27 | 75.77 | 74.92 |
8 | 73.03 | 75.31 | 75.97 | 75.85 | 75.63 | 75.45 | 74.92 | 74.42 | 73.14 |
9 | 69.95 | 74.07 | 74.81 | 74.84 | 74.59 | 74.43 | 73.91 | 73.27 | 72.15 |
10 | 67.73 | 72.47 | 73.30 | 73.32 | 73.2 | 73.12 | 72.82 | 72.08 | 70.90 |
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] |
|
/
〈 | 〉 |