面向对象的煤矸石堆场SPOT-5影像识别
作者简介:黄 丹(1989-),女,江苏南通人,硕士生,研究方向为遥感与地理信息系统的应用研究。E-mail:huangdan21001@gmail.com
收稿日期: 2014-05-13
要求修回日期: 2014-10-28
网络出版日期: 2015-03-10
基金资助
环保公益资助项目(201109043)
Coal Gangue Yards Information Extraction Using Object-oriented Method Based on SPOT-5 Remote Sensing Images
Received date: 2014-05-13
Request revised date: 2014-10-28
Online published: 2015-03-10
Copyright
煤矸石是一种在成煤过程中与煤层伴生的黑灰色固体废弃物,不仅会污染环境,而且会严重损害附近居民的身体健康,目前已经成为矿区生态环境的主要影响源之一。因此,实时、准确、快速地获取煤矸石堆场的位置、形状和面积等信息,对于环境监测与管理具有重要的意义。本文以内蒙古鄂尔多斯市东胜区为试验区,将试验区内的典型地物分为:植被、水体、阴影、裸地、建筑、道路、排土排矸场、露天煤矸石堆场、堆煤场及煤渣、采煤坑和其他共11类。本文采用SPOT-5高分辨率遥感影像,面向对象提取研究区内的煤矸石堆场信息,并进行识别精度评价,精度达到89.47%。将面向对象的分类结果与最大似然分类方法的分类结果进行比较,结果表明,面向对象的提取方法可更好地应用于煤矸石堆场信息的自动提取,大幅度提高精度和效率。
黄丹 , 刘庆生 , 刘高焕 , 闫文博 . 面向对象的煤矸石堆场SPOT-5影像识别[J]. 地球信息科学学报, 2015 , 17(3) : 369 -377 . DOI: 10.3724/SP.J.1047.2015.00369
Coal gangue is a kind of dark gray solid waste generated during mining process. Nowadays, coal gangue has become one of the biggest pollution sources to the ecological environment in mining areas. The accumulation of coal gangue not only occupies excessive land and causes serious environmental problem, but also damages the health of local people. Therefore, it is urgent to reduce the coal gangue yards in mining areas. In addition, extracting the location, shape and size information of coal gangue yards quickly and accurately is significant to environmental departments. The traditional methods of investigating coal gangue yards cost a lot of time and money. While remote sensing technologies can record the information of earth surface quickly and accurately, they have obvious superiorities in extracting coal gangue yards information. This paper takes the Dongsheng District, which locates in Ordos City of Inner Mongolia, as the study area, and utilizes SPOT-5 high resolution image as the data source. Then, this paper adopts the object-oriented method to extract coal gangue yards information from the study area. Multi-resolution segmentation and fuzzy classification algorithm are the most important steps in this method. Four appropriate segmentation scales are determined through comparisons of several tests, they are: 400, 160, 80 and 40. Next, we classify the segmented objects into different classes using the fuzzy classification algorithm that based on objects’ characteristics, such as spectrum, shape, texture and other features. The objects are further classified into eleven classes: bare area, buildings, roads, water, vegetation, shadows, dumps, coal gangue yards, coal yards, coal pits and others. The rule set used to extract different classes is built, which is aimed to provide a reference to relevant environmental departments to quickly and conveniently monitor the environment in coal mining area. In the end, we assess the accuracy of the classification results: the total accuracy is 88.78% and the user accuracy of coal gangue yards information is about 89.47%. Besides, a comparative extraction result is extracted using maximum likelihood method, whose total accuracy is 64.13% and the user accuracy of coal gangue yards information is only about 64.18%, which is much lower than the result extracted by the object-oriented method. Generally, due to the serious pollution caused by coal gangue yards, and considering the object-oriented classification method is seldom used to extract coal gangue yards information in China and abroad, this paper tries to extract coal gangue yards information using object-oriented classification method, and establish a rule set that can be applied to extract coal gangue yards information and other typical features. As we can see from the analyses, this paper has its significance in environmental protection. Relevant conclusions and analysis can be provided to the environmental protection departments as a useful reference to monitor and manage the pollution caused by coal gangue yards.
Fig. 1 Study area (Dongsheng District of Ordos City, Inner Mongolia)图1 内蒙古鄂尔多斯市东胜试验区 |
Fig. 2 SPOT-5 image of the study area图2 研究区的SPOT-5影像 |
Fig. 3 Flowchart of coal gangue yard information extraction图3 煤矸石堆场信息提取流程图 |
Tab. 1 Image segmentation parameters表1 影像分割参数 |
典型地物 | 分割参数 | ||||
---|---|---|---|---|---|
尺度参数 | 光谱因子 | 形状因子 | 紧致度 | 光滑度 | |
阴影 | 40 | 0.8 | 0.2 | 0.5 | 0.5 |
植被 | 40 | 0.8 | 0.2 | 0.5 | 0.5 |
水体 | 160 | 0.8 | 0.2 | 0.6 | 0.4 |
建筑物 | 160 | 0.8 | 0.2 | 0.6 | 0.4 |
道路 | 160,80 | 0.8 | 0.2 | 0.6 | 0.4 |
露天煤矸石堆场 | 80 | 0.9 | 0.1 | 0.6 | 0.4 |
采煤坑 | 80 | 0.9 | 0.1 | 0.6 | 0.4 |
堆煤场 | 80 | 0.9 | 0.1 | 0.6 | 0.4 |
排土排矸场 | 400 | 0.9 | 0.1 | 0.6 | 0.4 |
裸地 | 400 | 0.9 | 0.1 | 0.6 | 0.4 |
Fig. 4 Results of multi-resolution segmentation图4 影像多尺度分割结果 |
Tab. 2 Classification rule set of the features at the study area表2 研究区地物的分类规则集 |
影像层次 | 地物类别 | 分类特征 | 模糊函数 | 函数值 |
---|---|---|---|---|
Level 1 | 裸地 | Brightness | [119,126] | |
Ratio-swir | [0.081,0.084] | |||
GLCM Mean(all dir.) | [4.3,7.2] | |||
排土排矸场(在类别裸地中提取) | Area | [32 000,35 000] | ||
Shape Index | [0.26,0.3] | |||
Level 2 | 建筑 | NDBI | [0.17,0.2] | |
GLCM Contrast | [13,15] | |||
宽道路(在类别建筑中提取) | Length | [180,200] | ||
Length/Width | [7,7.4] | |||
水体 | MNDWI | [-0.1,0] | ||
Level 3 | 采煤坑 | Brightness | [83,89] | |
Area | [19 000,34 000] | |||
Distance to | 2000 | |||
堆煤场、煤渣(在类别采煤坑中提取) | Rectangular Fit | [0.43,0.52] | ||
露天煤矸石堆场 | Distance to | 2000 | ||
Brightness | [101,119] | |||
Area | [20 000,27 900] | |||
窄道路 | Brightness | [106,114] | ||
Length/Width | [4.7,5.5] | |||
Level 4 | 阴影 | Brightness | [63,70] | |
植被 | SAVI | [0.1,0.12] |
Fig. 5 Classification results of different features (part)图5 各类地物的提取结果图(局部) |
Fig. 6 Classification results based on object-oriented method图6 面向对象方法所得的分类结果 |
Fig. 7 Classification results based on maximum likelihood method图7 最大似然方法所得的分类结果 |
Tab. 3 Information extraction accuracy assessment based on object-oriented method表3 面向对象方法信息提取结果精度评价 |
地物类型 | 采煤坑 | 道路 | 堆煤场 | 建筑 | 露天煤矸石堆场 | 裸地 | 排土排矸场 | 水体 | 阴影 | 植被 | |
---|---|---|---|---|---|---|---|---|---|---|---|
生产者精度(%) | 80.00 | 87.50 | 81.58 | 88.04 | 79.07 | 95.16 | 88.46 | 96.43 | 90.03 | 89.80 | |
用户精度(%) | 100.00 | 90.84 | 84.93 | 90.44 | 89.47 | 84.29 | 90.79 | 83.08 | 89.56 | 86.94 | |
总体精度(%) | 88.78 | ||||||||||
Kappa | 0.8649 |
Tab. 4 Accuracy assessment of information extraction results based on maximum likelihood method表4 最大似然分类方法信息提取结果精度评价 |
地物类型 | 采煤坑 | 道路 | 堆煤场 | 建筑 | 露天煤矸石堆场 | 裸地 | 排土排矸场 | 水体 | 阴影 | 植被 | |
---|---|---|---|---|---|---|---|---|---|---|---|
生产者精度(%) | 71.54 | 56.10 | 62.60 | 62.60 | 61.43 | 65.90 | 69.47 | 66.70 | 58.60 | 62.00 | |
用户精度(%) | 60.33 | 66.20 | 69.70 | 65.30 | 64.18 | 63.00 | 77.65 | 61.50 | 58.00 | 60.30 | |
总体精度(%) | 64.13 | ||||||||||
Kappa | 0.5712 |
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] |
|
[30] |
|
/
〈 | 〉 |