多特征辅助下的GF-6 WFV影像准噶尔山楂识别研究

doi:10.13866/j.azr.2021.02.27

Abstract

Abstract:

Our ability to extract spatial distribution information for Crataegus songarica by remote sensing is relatively weak, and the performance of a single algorithm in information extraction is different. To address this problem, we explored the application potential of domestic Gaofen-6 (GF-6) wide-field-view (WFV) data in tree species recognition in arid and semi-arid areas. Using GF-6 WFV and Ziyuan-3 (ZY-3) data, we constructed a combined classifier recognition algorithm with multi-feature assistance. First, we combined a linear spectral clustering algorithm for ZY-3 data super-pixel segmentation with the sample set to determine the optimal scale, avoid the salt-and-pepper phenomenon and improve the recognition accuracy. Second, we extracted spectral features, texture features, and vegetation index features based on multi-source data, using the recursive feature elimination method to select classification features. Feature importance was evaluated based on the mean decrease impurity to construct the optimal classification feature set, improve the separability between feature classes, and reduce the redundancy between features. Finally, we constructed a weight-adaptive voting combination classifier based on the support vector machine algorithm and random forest algorithm. Combined with a variety of classification schemes, the spatial distribution of Crataegus songarica was extracted and verified to evaluate the influence of object-oriented algorithm, multi-classifier combination algorithm, and feature selection on the classification accuracy. The results showed that GF-6 WFV data can be used to identify Crataegus songarica and have great application potential in forestry production. Compared with GF-1 WFV, two new red-edge bands of GF-6 WFV play an important role in the identification of Crataegus songarica. Compared with traditional pixel-based methods, the object-based recognition of Crataegus songarica effectively mitigated the pepper salt phenomenon and significantly improved the recognition accuracy, algorithm efficiency, and robustness. Moreover, the feature selection effectively reduced the redundancy between classification features and improved the computational efficiency and algorithm stability; the selected features significantly improved the accuracy. Using a weight-adaptive multi-classifier voting combination algorithm can integrate the advantages of different algorithms, effectively avoid the partial ‘confusion’ caused by the representation force difference of the single classifier, and improve the recognition accuracy.

Key words: GF-6, feature selection, object-oriented, multi classifier combination, Crataerus songarica

CHEN Chunxiu,CHEN Shujiang,XU Shiwei,CHEN Mengyu,JIA Xiang,HUANG Tiecheng,LI Chunlei. Crataegus songarica recognition using Gaofen-6 wide-field-view data assisted by multiple features[J].Arid Zone Research, 2021, 38(2): 553-561.

Figures/Tables 10

Fig. 1

Tab. 1

Fig. 2

Tab. 2

Table of vegetation index"

植被指数	计算公式	表征意义及取值范围
NDVI	$ρ NIR - ρ RED ρ NIR + ρ RED$	表征植被绿地,植被茂密时其灵敏度会降低;取值范围-1~1,一般植被范围是0.2~0.8
RVI	$ρ NIR ρ RED$	大气敏感,在植被茂密时分辨力更强;取值范围0~30+,一般植被范围2~8
DVI	$ρ NIR - ρ RED$	土壤敏感,在植被相对稀疏时,表征能力更强;取值范围量级差异较大
VRE	$ρ 740 ρ 720$	对叶绿素浓度、叶冠层和水分含量的综合灵敏;取值范围0~20,一般植被范围4~8
PRI	$ρ 531 - ρ 570 ρ 531 + ρ 570$	对活林木的类胡萝卜素变化反应灵敏;取值范围-1~1,一般植被区范围-0.2~0.2
ARI1	$1 / ρ 550 - 1 / ρ 700$	对叶片中的花青素非常敏感;取值范围0~0.2+,一般植被范围0.001~0.1
ARI2	$ρ 800 × 1 / ρ 550 - 1 / ρ 700$	ARI2是ARI1的改进,在高花青素浓度时更加有效;取值范围0~0.2+,一般植被范围0.001~0.1

Tab. 2

Fig. 3

Fig. 4

Tab. 3

Tab. 4

Fig. 5

Fig. 6

References 15

[1]	杨蕾, 吕海英, 李进, 等. 新疆天山野果林准噶尔山楂种群结构与动态分析[J]. 西北植物学报, 2018,38(12):2314-2323.
	[ Yang Lei, Lyu Haiying, Li Jin, et al. Structure and dynamic analysis of Crataegus songarica K. Koch population in Tianshan wild fruit forest of Xinjiang[J]. Acta Botanica Boreali-Occidentalia Sinica, 2018,38(12):2314-2323. ]
[2]	盛芳, 陈淑英, 田嘉, 等. 新疆准噶尔山楂不同居群的遗传多样性[J]. 生物多样性, 2017,25(5):518-530.
	[ Sheng Fang, Chen Shuying, Tian Jia, et al. Genetic diversity of Crataegus songorica in Xinjiang[J]. Biodiversity Science, 2017,25(5):518-530. ]
[3]	邬亚娟, 刘廷玺, 童新, 等. 基于面向对象的干旱半干旱地区植被分类[J]. 干旱区研究, 2020,37(4):1026-1034.
	[ Wu Yajuan, Liu Tingxi, Tong Xin, et al. Vegetation classification in arid and semi-arid areas based on object-oriented method[J]. Arid Zone Research, 2020,37(4):1026-1034. ]
[4]	樊雪, 刘清旺, 谭炳香. 基于机载PHII高光谱数据的森林优势树种分类研究[J]. 国土资源遥感, 2017,29(2):110-116.
	[ Fan Xu, Liu Qingwang, Tan Bingxiang. Classification of forest species using airborne PHII hyperspectral data[J]. Remote Sensing for Land and Resources, 2017,29(2):110-116. ]
[5]	赵希妮, 王磊, 刘雅清, 等. 基于GF-1/WFV时间序列的葡萄识别模型——以宁夏红寺堡区为例[J]. 干旱区研究, 2019,36(3):630-638.
	[ Zhao Xini, Wang Lei, Liu Yaqing, et al. Grape recognition model based on GF-1 / WFV time series: A case study of hongsipao District, Ningxia[J]. Arid Zone Research, 2019,36(3):630-638. ]
[6]	Koukal T, Atzberger C, Schneider W. Evaluation of semi-empirical BRDF models inverted against multi-angle data from a digital airborne frame camera for enhancing forest type classification[J]. Remote Sensing of Environment, 2014,151:27-43.
[7]	Yang X H, Rochdi N, Zhang J K, et al. Mapping tree species in a boreal forest area using RapidEye and Lidar data[J]. International Geoscience and Remote Sensing Symposium, 2014: 69-71.
[8]	Yuan Q Q, Shen H F, Li T W, et al. Deep learning in environmental remote sensing: Achievements and challenges[J]. Remote Sensing of Environment, 2020,241:111716.
[9]	栗旭升, 李虎, 陈冬花, 等. 联合GF-5与GF-6卫星数据的多分类器组合亚热带树种识别[J]. 林业科学, 2020,56(10):93-104.
	[ Li Xusheng, Li Hu, Chen Donghua, et al. Multiple classifiers combination method for tree species identification based on GF-5 and GF-6[J]. Scientia Silvae Sinicae, 2020,56(10):93-104. ]
[10]	王怀警, 谭炳香, 王晓慧, 等. 多分类器组合森林类型精细分类[J]. 遥感信息, 2019,34(2):107-115.
	[ Wang Huaijing, Tan Binxiang, Wang Xiaohui, et al. Multiple classifiers combination method for precise classification of forest type[J]. Remote Sensing Information, 34(2):107-115. ]
[11]	刘怡君, 庞勇, 廖声熙, 等. 机载LiDAR和高光谱融合实现普洱山区树种分类[J]. 林业科学研究, 29(3):407-412.
	[ Liu Yijun, Pang Yong, Liao Shengxi, et al. Merged airborne LiDAR and hyperspectral data for tree species classification in puer’s mountainous area[J]. Forest Research, 2016,29(3):407-412. ]
[12]	刘丽娟, 庞勇, 范文义, 等. 机载LiDAR和高光谱融合实现温带天然林树种识别[J]. 遥感学报, 2013,17(3):679-695.
	[ Liu Lijuan, Pang Yong, Fan Wenyi, et al. Fused airborne LiDAR and hyperspectral data for tree species identification in a natural temperate forest[J]. Journal of Remote Sensing, 2013,17(3):679-695. ]
[13]	刘雅清, 王磊, 赵希妮, 等. 基于GF-1/WFV时间序列的绿洲作物类型提取[J]. 干旱区研究, 2019,36(3):781-789.
	[ Liu Yaqing, Wang Lei, Zhao Xini, et al. Extraction of crops in oasis based on GF-1/WFV time series[J]. Arid Zone Research, 2019,36(3):781-789. ]
[14]	李利平, 海鹰, 安尼瓦尔·买买提, 等. 新疆伊犁地区野果林的群落特征及保护[J]. 干旱区研究, 2011,28(1):60-66.
	[ Li Liping, Hai Ying, Anwar Mohammat, et al. Community structure and conservation of wild fruit forests in the Ili Valley, Xinjiang[J]. Arid Zone Research, 2011,28(1):60-66. ]
[15]	宋熙煜, 周利莉, 李中国, 等. 图像分割中的超像素方法研究综述[J]. 中国图象图形学报, 2015,20(5):599-608.
	[ Song Xiyu, Zhou Lili, Li Zhongguo, et al. Review on superpixel methods in image segmentation[J] Journal of Image and Graphics, 2015,20(5):599-608. ]

数据类型	成像时间	波段名	波长范围/μm	空间分辨率/m
GF-6 WFV	2019-08-06	蓝波段	0.45~0.52	16.0
		绿波段	0.52~0.59
		红波段	0.63~0.69
		近红外	0.77~0.89
		红边波段1	0.69~0.73
		红边波段2	0.73~0.77
		紫波段	0.40~0.45
		黄波段	0.59~0.63
ZY-3	2019-07-24	蓝波段	0.45~0.52	6.0
		绿波段	0.52~0.59
		红波段	0.63~0.69
		近红外	0.77~0.89
		全色波段	0.50~0.80	2.1

分类方案	分类特征因子	分类算法	是否分割
方案1	优化全部特征（21）	WAVCC	是
方案2	未优选全部特征（71）	WAVCC	是
方案3	优化后像元尺度特征（17）	WAVCC	否
方案4	优化后全部特征（21）	SVM	是
方案5	优化后全部特征（21）	RF	是
方案6	优化光谱特征+优化纹理特征（16）	WAVCC	是
方案7	优化光谱特征+优化植被指数特征（15）	WAVCC	是

分类方案	OA/%	Kappa	UA/%	PA/%	ET/min
方案1	95.27	0.92	93.85	94.72	17.0
方案2	95.29	0.71	90.33	94.51	38.8
方案3	92.43	0.88	89.41	86.99	30.4
方案4	93.46	0.76	92.72	89.96	9.4
方案5	91.96	0.81	88.86	91.21	5.7
方案6	87.34	0.91	87.33	87.10	12.9
方案7	90.59	0.83	91.22	89.57	11.3

Crataegus songarica recognition using Gaofen-6 wide-field-view data assisted by multiple features

RichHTML

PDF (PC)

Like

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 10

References 15

Related Articles 4

Metrics

Comments

Recommended 0

[1]	YAO Jinxi, XIAO Chengzhi, ZHANG Zhi, WANG Lang, ZHANG Kun. Vegetation feature type extraction in arid regions based on GEE multi-source remote sensing data [J]. Arid Zone Research, 2024, 41(1): 157-168.
[2]	BAI Jianduo,PENG Jie,SHI Zhou,WANG Yuzhen,LIU Weiyang,LI Hongyi. Precise management zoning in arid soil croplands based on multi-source data [J]. Arid Zone Research, 2022, 39(2): 646-655.
[3]	SONG Qi,SHI Zhou,FENG Chunhui,MA Ziqiang,JI Wenjun,PENG Jie,GAO Qi,JIANG Xuewei. Analysis of landscape pattern from 1990 to 2019 based on multi-temporal imagery in arid oasis [J]. Arid Zone Research, 2022, 39(2): 594-604.
[4]	YANG Xuefeng,YE Mao,Munire Maimaiti. Structural parameter acquisition of Populus euphratica by WorldView-2 remote sensing image [J]. Arid Zone Research, 2021, 38(6): 1659-1667.