Snow Information Recognition based on GF-6 PMS Images
-
摘要: 以黑龙江省哈尔滨市道外区为研究区,系统探讨分析了基于遥感的不同方法在积雪信息识别中的应用。首先,对研究区两个时相的高分六号(GF-6)多光谱相机(PMS)影像进行目视解译,掌握了研究区内地物类型和积雪分布特点。其次,基于目视解译结果,选取了8种典型地物类型,得到了“积雪”和“非雪”两类像元的光谱特征规律。再次,探讨分析了6种方法在积雪识别中的应用,利用精确率、召回率和F指数3个指标进行了精度评价。最后,提出了基于投票结果的最终识别结果判定方法,得到了研究区积雪信息最终识别结果。研究表明:①受下垫面和阴影的影响,研究区“同谱异物”和“同物异谱”现象普遍;②深度学习算法的识别效果最好,决策树法的识别效果相对较差;农田区的识别精度高于池塘区,误识别和漏识别的现象都相对较少;③基于投票结果的最终识别结果判定,可以有效改善单一识别方法存在的误识别和漏识别现象。Abstract: From the perspective of limited research on snow recognition in high spatial resolution optical remote sensing images, considering Daowai District of Harbin City as the research area, this paper systematically discusses and analyzes the application of different methods in snow information recognition. First, ground feature types and snow distribution characteristics were mastered through visual interpretation of GF-6 PMS images of two phases in the study area. Second, based on the results of visual interpretation, eight typical surface feature types were selected, and the spectral characteristics of "snow"and "non-snow"pixels were obtained. Third, the application of the six methods in snow recognition was discussed and analyzed. Accuracy was evaluated using three indexes: positive predictive value, recall rate, and F-score. Finally, a final recognition result judgment method based on voting results was developed, and the final recognition result of snow information in the study area was obtained. The results showed that, owing to the influence of the underlying surface and shadow, the phenomenon of "same spectrum foreign matter"and "same object different spectrum"was common in the study area, which interfered with the snow recognition process to a large extent. The recognition effect of the deep learning algorithm was the best, while that of the decision tree method was relatively poor; the recognition accuracy was higher for the farmland area than that for the pond area, and the phenomenon of false and missing recognitions was relatively less. The final recognition result judgment method based on voting results can effectively improve the phenomenon of false and missing recognitions in a single recognition method. This paper has important guiding significance for snow recognition based on high-spatial-resolution optical remote sensing images.
-
Key words:
- snow recognition /
- GF-6 PMS /
- visual interpretation /
- snow index /
- deep learning algorithm
-
表 1 GF-6 PMS相机参数
Table 1. Parameters of Gf-6 PMS camera
Band Spectral range/μm Spatial resolution/m Swath width/km Revisit cycle/day Blue 0.45-0.52 8 ≥90 4 Green 0.52-0.60 8 Red 0.63-0.69 8 NIR 0.76-0.90 8 Pan 0.45-0.90 2 表 2 积雪识别结果统计
Table 2. Statistics of snow recognition results
Method Number of snow-covered pixels/ten thousand Ratio of snow-covered pixels/% Number of non-snow pixels/ten thousand Ratio of non-snow pixels/% Decision tree 2155.65 75.1 713.56 24.9 Random forest 2200.94 76.7 668.26 23.3 Maximum likelihood 2301.13 80.2 568.08 19.8 SVM 2239.20 78.0 630.01 22.0 Softmax 2524.23 88.0 344.98 12.0 Deep learning algorithm 2475.66 86.3 393.55 13.7 表 3 “争议”像元积雪识别结果统计
Table 3. Statistics of snow recognition results in controversial pixels
Area Recognition result Decision tree Random forest Maximum likelihood SVM Softmax Deep learning algorithm A Snow 0 97 5756 2196 13347 13347 Non-snow 13347 13250 7591 11151 0 0 Proportion of snow/% 0 0.7 43.1 16.5 100 100 B Snow 0 441 1078 18 2162 2162 Non-snow 2162 1721 1084 2144 0 0 Proportion of snow/% 0 20.4 49.9 0.8 100 100 C Snow 108 313 435 0 1898 1898 Non-snow 1790 1595 1463 1898 0 0 Proportion of snow/% 5.7 16.5 22.9 0 100 100 D Snow 129 316 2854 2151 3766 3765 Non-snow 3637 3450 912 1615 0 1 Proportion of snow/% 3.4 8.4 75.8 57.1 100 ≈100 表 4 小“争议”像元积雪识别结果统计
Table 4. Statistics of snow recognition results in little controversial pixels
Area Recognition result Decision tree Random forest Maximum likelihood SVM Softmax Deep learning algorithm E Snow 0 10115 10115 10115 10115 10115 Non-snow 10115 0 0 0 0 0 Proportion of snow/% 0 100 100 100 100 100 F Snow 0 23746 23746 23746 23746 23746 Non-snow 23746 0 0 0 0 0 Proportion of snow/% 0 100 100 100 100 100 G Snow 0 0 0 0 1169 0 Non-snow 1169 1169 1169 1169 0 1169 Proportion of non-snow/% 100 100 100 100 0 100 H Snow 19 3 0 0 1351 0 Non-snow 1354 1370 1373 1373 22 1373 Proportion of non-snow/% 98.6 99.8 100 100 1.6 100 -
[1] HE C, Liou K N, Takano Y, et al. Impact of grain shape and multiple black carbon internal mixing on snow albedo: Parameterization and radiative effect analysis[J]. Journal of Geophysical Research: Atmospheres, 2018, 123(2): 1253-1268. doi: 10.1002/2017JD027752 [2] Butt M J, Bilal M. Application of snowmelt runoff model for water resource management[J]. Hydrological Processes, 2011, 25(24): 3735-3747. doi: 10.1002/hyp.8099 [3] LIU M, XIONG C, PAN J, et al. High-resolution reconstruction of the maximum snow water equivalent based on remote sensing data in a mountainous area[J]. Remote Sensing, 2020, 12(3): 460-479. doi: 10.3390/rs12030460 [4] XU L, Dirmeyer P. Snow–atmosphere coupling strength. Part Ⅱ: Albedo effect versus hydrological effect[J]. Journal of Hydrometeorology, 2013, 14(2): 404-418. doi: 10.1175/JHM-D-11-0103.1 [5] 汪左. 新疆玛纳斯河流域典型区雪水当量的SAR反演研究[D]. 南京: 南京大学, 2014.WANG Zuo. Retrieval of Snow Water Equivalence Using SAR Data for Typical Area of Manas River Basin in Xinjiang, China[D]. Nanjing: Nanjing University, 2014. [6] 贺广均. 联合SAR与光学遥感数据的山区积雪识别研究[D]. 南京: 南京大学, 2015.HE Guangjun. Snow Recognition in Mountainous Areas Based on SAR and Optical Remote Sensing Data[D]. Nanjing: Nanjing University, 2015. [7] Barnes J C, Bowley C J. Snow cover distribution as mapped from satellite photography[J]. Water Resources Research, 1968, 4(2): 257-272. doi: 10.1029/WR004i002p00257 [8] 冯学智. 卫星雪盖制图及其应用研究概况[J]. 遥感技术动态, 1989(1): 25-29. https://www.cnki.com.cn/Article/CJFDTOTAL-YGJS198901004.htmFENG Xuezhi. Satellite snow cover mapping and its application[J]. Development of Remote Sensing Technology, 1989(1): 25-29. https://www.cnki.com.cn/Article/CJFDTOTAL-YGJS198901004.htm [9] 冯学智. 卫星雪盖制图中的一些技术问题[J]. 遥感技术与应用, 1991(4): 10-15. https://www.cnki.com.cn/Article/CJFDTOTAL-YGJS199104001.htmFENG Xuezhi. Some technical problems in satellite snowcover mapping[J]. Remote Sensing Technology and Application, 1991(4): 10-15. https://www.cnki.com.cn/Article/CJFDTOTAL-YGJS199104001.htm [10] 白磊, 郭玲鹏, 马杰, 等. 基于数字相机拍摄影像的山区积雪消融动态观测研究——以天山积雪站为例[J]. 资源科学, 2012, 34(4): 620-628. https://www.cnki.com.cn/Article/CJFDTOTAL-ZRZY201204006.htmBAI Lei, GUO Lingpeng, MA Jie, et al. Observation and analysis of the process of snow melting at Tianshan station using the images by digital camera[J]. Resouces Science, 2012, 34(4): 620-628. https://www.cnki.com.cn/Article/CJFDTOTAL-ZRZY201204006.htm [11] Kim S-H, Hong C-H. Antarctic land-cover classification using IKONOS and Hyperion data at Terra Nova Bay[J]. International Journal of Remote Sensing, 2012, 33(22): 7151-7164. doi: 10.1080/01431161.2012.700136 [12] ZHU L, XIAO P, FENG X, et al. Support vector machine-based decision tree for snow cover extraction in mountain areas using high spatial resolution remote sensing image[J]. Journal of Applied Remote Sensing, 2014, 8(1): 084698. doi: 10.1117/1.JRS.8.084698 [13] Dozier J. Spectral signature of alpine snow cover from the Landsat Thematic Mapper[J]. Remote Sensing of Environment, 1989, 28(1): 9-22. http://static1.1.sqspcdn.com/static/f/891472/15152129/1321451632750/Dozier_J._1989.pdf?token=sMvVhVmXaLvHsiAGq%2FlmCvgkjlA%3D [14] Cea C, Cristóbal J, Pons X. An improved methodology to map snow cover by means of Landsat and MODIS imagery[C]//2007 IEEE International Geoscience and Remote Sensing Symposium, 2007: 4217-4220. [15] Khosla D, Sharma J, Mishra V. Snow cover monitoring using different algorithm on AWiFS sensor data[J]. International Journal of Advanced Engineering Sciences and Technologies, 2011, 7(1): 42-47. [16] 延昊. 利用MODIS和AMSR-E进行积雪制图的比较分析[J]. 冰川冻土, 2005(4): 515-519. doi: 10.3969/j.issn.1000-0240.2005.04.008YAN Hao. A comparison of MODIS and passive microwave snow mapping[J]. Journal of Glaciology and Geocryology, 2005(4): 515-519. doi: 10.3969/j.issn.1000-0240.2005.04.008 [17] 郝晓华, 王建, 李弘毅. MODIS雪盖制图中NDSI阈值的检验——以祁连山中部山区为例[J]. 冰川冻土, 2008(1): 132-138. https://www.cnki.com.cn/Article/CJFDTOTAL-BCDT200801020.htmHAO Xiaohua, WANG Jian, LI Hongyi. Evaluation of the NDSI threshold value in mapping snow cover of MODIS[J]. Journal of Glaciology and Geocryology, 2008(1): 132-138. https://www.cnki.com.cn/Article/CJFDTOTAL-BCDT200801020.htm [18] 汪凌霄. 玛纳斯河流域山区积雪遥感识别研究[D]. 南京: 南京大学, 2012.WANG Lingxiao. Retrieval of Snow Water Equivalence Using SAR Data for Typical Area of Manas River Basin in Xinjiang, China[D]. Nanjing: Nanjing University, 2012. [19] Baghdadi N, Gauthier Y, Bernier M. Capability of multitemporal ERS-1 SAR data for wet-snow mapping[J]. Remote Sensing of Environment, 1997, 60(2): 174-186. doi: 10.1016/S0034-4257(96)00180-0 [20] Rott H, Nagler T. Capabilities of ERS-1 SAR for snow and glacier monitoring in alpine areas[J]. European Space Agency-Publications-ESA SP, 1994, 361: 965-965. http://www.researchgate.net/publication/245605630_Capabilities_of_ERS-1SAR_for_snow_and_glacier_monitoring_in_Alpine_areas [21] Koskinen J T, Pulliainen J T, Hallikainen M T. The use of ERS-1 SAR data in snow melt monitoring[J]. IEEE Transactions on Geoscience Remote Sensing, 1997, 35(3): 601-610. doi: 10.1109/36.581975 [22] Luojus K P, Pulliainen J T, Cutrona A B, et al. Comparison of SAR-based snow-covered area estimation methods for the boreal forest zone[J]. IEEE Geoscience Remote Sensing Letters, 2009, 6(3): 403-407. doi: 10.1109/LGRS.2009.2014786 [23] Caves R, Hodson A, Turpin O, et al. Field verification of SAR wet snow mapping in a non-Alpine environment[J]. European Space Agency- Publications- ESA SP, 1998, 441: 519-526. http://www.researchgate.net/publication/228650728_Field_verification_of_SAR_wet_snow_mapping_in_a_non-Alpine_environment/download [24] Malnes E, Guneriussen T. Mapping of snow covered area with Radarsat in Norway[C]//IEEE International Geoscience and Remote Sensing Symposium, 2002: 683-685. [25] Kelly R. The AMSR-E snow depth algorithm: description and initial results[J]. Journal of the Remote Sensing Society of Japan, 2009, 29(1): 307-317. https://www.jstage.jst.go.jp/article/rssj/29/1/29_1_307/_pdf/-char/en [26] PAN J, JIANG L, ZHANG L. Wet snow detection in the south of China by passive microwave remote sensing[C]//IEEE International Geoscience and Remote Sensing Symposium, 2012: 4863-4866. [27] Singh P R, Gan T Y. Retrieval of snow water equivalent using passive microwave brightness temperature data[J]. Remote Sensing of Environment, 2000, 74(2): 275-286. doi: 10.1016/S0034-4257(00)00121-8 [28] LIU X, JIANG L, WU S, et al. Assessment of methods for passive microwave snow cover mapping using FY-3C/MWRI data in China[J]. Remote Sensing, 2018, 10(4): 524-545. doi: 10.3390/rs10040524 [29] Hinkler J, Pedersen S B, Rasch M, et al. Automatic snow cover monitoring at high temporal and spatial resolution, using images taken by a standard digital camera[J]. International Journal of Remote Sensing, 2002, 23(21): 4669-4682. doi: 10.1080/01431160110113881 [30] Keshri A, Shukla A, Gupta R. ASTER ratio indices for supraglacial terrain mapping[J]. International Journal of Remote Sensing, 2009, 30(2): 519-524. doi: 10.1080/01431160802385459 [31] Hinkler J, Ørbæk J B, Hansen B. Detection of spatial, temporal, and spectral surface changes in the Ny-Ålesund area 79 N, Svalbard, using a low cost multispectral camera in combination with spectroradiometer measurements[J]. Physics Chemistry of the Earth, Parts A/B/C, 2003, 28(28-32): 1229-1239. doi: 10.1016/j.pce.2003.08.059 [32] XIAO X M, SHEN Z X, QIN X G. Assessing the potential of VEGETATION sensor data for mapping snow and ice cover: a normalized difference snow and ice index[J]. International Journal of Remote Sensing, 2001, 22(13): 2479-2487. doi: 10.1080/01431160119766 [33] XIAO X, Moore B, QIN X, et al. Large-scale observations of alpine snow and ice cover in Asia: using multi-temporal vgetation sensor data[J]. International Journal of Remote Sensing, 2002, 23(11): 2213-2228. doi: 10.1080/01431160110076180 [34] Long J, Shelhamer E, Darrell T. Fully convolutional networks for semantic segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2015: 3431-3440.