Classification and detection of cloud, snow and fog in remote sensing images based on random forest
XU Yun1(), XU Aiwen2
1. Hangzhou Transportation Planning and Design Institute, Hangzhou 310003, China 2. Zhejiang Academy of Land and Space Planning, Hangzhou 310012, China
Cloud, snow and fog are important factors affecting the quality of optics remote sensing images, and hence researchers should detect the range of cloud, snow, fog in remote sensing images and remove unwanted images so as to improve the utilization of remote sensing images. In this paper, the authors studied the method based on Random Forest to detect cloud, snow, fog and tried to reduce the error detection rate by means of adding a “second detection”. Experiments show that this method has high detection accuracy and efficiency.
Yun XU,Aiwen XU. Classification and detection of cloud, snow and fog in remote sensing images based on random forest[J]. Remote Sensing for Land & Resources,
2021, 33(1): 96-101.
Fig.1 Cloud, snow, fog sample of NAD and MUX image
Fig.2 Algorithm flowchart of cloud, snow, fog classification
Fig.3 Land feature contrast with cloud, snow, fog
Fig.4 Cloud, snow, fog classification results of each satellite
类别
地物/像素
云/像素
雾/像素
雪/像素
总计/像素
地物
6 737 349
112 378
93 557
104 046
7 047 330
云
175 042
898 072
55 769
15 071
1 143 954
雾
100 075
86 752
214 309
1 437
402 573
雪
228 746
8 250
86
586 876
823 958
总计
7 241 212
1 105 452
363 721
707 430
9 417 815
精度指标
总体分类精度: 89.6%; Kappa=0.741
Tab.1 Confusion matrix of the first detection results
类别
地物/像素
云/像素
雾/像素
雪/像素
总计/像素
地物
6 851 716
91 773
57 012
54 405
7 054 906
云
171 944
928 589
48 496
14 018
1 163 047
雾
92 379
78 142
258 159
1 383
430 063
雪
125 173
6 948
54
637 624
769 799
总计
7 241 212
1 105 452
363 721
707 430
9 417 815
精度指标
总体分类精度: 92.1%; Kappa=0.804
Tab.2 Confusion matrix of the second detection results
影像类别
传感器 类别
影像总 数/幅
一检合 格数/幅
二检合 格数/幅
一检合 格率(精 度)/%
二检合 格率(精 度)/%
ZY-3
NAD
1 023
934
995
91.3
97.3
ZY1-02C
NAD
554
488
527
88.1
95.1
GF-1
MUX
832
752
797
90.4
95.8
TH01-01
MUX
317
278
296
87.8
93.4
Tab.3 Cloud, snow, fog detection accuracy of each satellite remote sensing images
[1]
Saunders R W, Kriebel K T. An improved method for detecting clear sky and cloudy radiances from AVHRR Data[J]. International Journal of Remote Sensing, 1987,9:123-150.
[2]
Ackerman S, Strabala K, Menzel W P, et al. Discriminating clear sky from clouds with MODIS[J]. Journal of Geophysical Research, 1998,103:141-157.
[3]
Merchant C J, Harris A R, Maturi E, et al. Probabilistic physically based cloud screening of satellite infraredimagery for operational sea surface temperature retrieval[J]. Quarterly Journal of the Royal Meteorological Society, 2005,131(611):2735-2755.
[4]
Baum B, Tovinkere V TitlowJ. et al. Automated cloud classification of global AVHRR data using a fuzzy logic approach[J]. Journal of Applied Meteorology, 1997,36:1519-1540.
[5]
Bendix J, Thies B, Nauss T, et al. A feasibility study of daytime fog and low stratus detection with TERRA/AQUA MODIS over land[J]. Appl Meteor, 2006,13(2):111-125
Li Y C, Sun H, Li X G, et al. Study on detection of daytime fog using GMS-5 weather satellite data[J]. Journal of Nanjing Institute of Meteorology, 2001,24(3):121-129.
Ding H Y, Ma L L, Li Z Y, et al. Automatic identification of cloud and snow based on fractal dimension[J]. Remote Sensing Technology and Application, 2013,28(1):52-57.
url: http://www.rsta.ac.cn/CN/abstract/abstract2206.shtml
Liu X H, Cao X G, Yu W X. A cloud detection algorithm based on fractal dimension[C]. Remote Sensing Science and Technology Forum in 2006,Taiyuan, 2006.
[11]
雷震. 随机森林及其在遥感影像处理中应用研究[D]. 上海:上海交通大学, 2012.
[11]
Lei Z. Random forests and its application in remote sensing image processing[D]. Shanghai:Shanghai Jiaotong University, 2012.