Procedia Environmental Sciences 10 ( 2011 ) 896 – 901

doi: 10.1016/j.proenv.2011.09.143

Available online at www.sciencedirect.com

2011 3rd International Conference on Environmental Science and Information Application Technology

Using ALOS High Spatial Resolution Image to Detect Vegetation Patches

Qingsheng LIUa*, Gaohuan LIUa, Chong HUANGa, Chuanjie XIEa, Lei SHIa a* aState Key Lab. of Resources and Environmental Information System, IGSNRR, CAS, Beijing 100101, P. R. China

Abstract

Vegetation often exists as patch in arid and semi-arid region. Numbers and areas and locations and spatial structure characteristics of vegetation patch are the important parameters for vegetation function and structure researches in landscape ecology. Visual interpretation of high spatial resolution remote sensing image to study landscape structure characters is often used, which needs lot of time and labours. In this paper, vegetation patches are detected using the canny edge detector and ellipticity measurement based on ALOS fusion image, and the whole processing steps were given. The detection accuracy of the patches is about 89.3%, which is better than that of the methods that only used the spectral difference between vegetation patches and background. The experiments show that integration the canny edge detector with the algorithm for extracting circle and ellipse objects is simple and effective for detecting vegetation patches.

Keyword: ALOS; Canny edge detector; Vegetation patch; ellipticity measurement

1. Introduction

The patched spatial distribution of vegetation is the ubiquitous style of vegetation landscape pattern. Area and shape and number of patches in landscape not only have the impact on various ecological processes, but also are one of the important determinative factors of biodiversity [1]. So many scholars focus their researches on vegetation patch characters [2-4]. Using high spatial resolution remote sensing images to study landscape structure characters is one of the important aspects in landscape ecology, and

* Corresponding author. Tel.: 86-10-64889017; fax: 86-10-64889630. E-mail address: liuqs@lreis.ac.cn.

2011 3rd International Conference on Environmental Science and Information Application Technology (ESIAT 2011)

1878-0296 © 2011 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of Conference ESIAT2011 Organization Committee.

Open access under CC BY-NC-ND license.

© 2011 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of Conference ESIAT2011 Organization Committee.

Open access under CC BY-NC-ND license.

897 Qingsheng LIU et al. / Procedia Environmental Sciences 10 ( 2011 ) 896 – 901

visual interpretation of image is often used [5-7], which needs lot of time and labours. So the automated or semi-automated methods should be used to detect the vegetation patches. In fact, because vegetation often occurs as patches, the shape differences between vegetation patches and background can be used to detect the vegetation. In general, using the method based on shape and spectral difference between vegetation patches and background to detect vegetation patches is better than that of the method only based on spectral difference. In image processing field, circular and elliptic and rectangular objects in image have already been detected based on various algorithms very well [8-11]. For natural vegetation, Perez et al. (2000) [12] used colour and shape analysis techniques for weed detection in cereal fields, and McGlynn and Okin (2006) [13] used an object-oriented classification to detect the shrub distribution, and Vogt et al. (2007) [14] mapped spatial patterns with morphological image processing, and Carter et al. (2009) [15] compared the efficacy of Landsat 5 TM, QuickBird (QB) and EO-1 Hyperion data in discriminating tamarisk populations near De Beque, Colorado, USA, and proved that multispectral data at high spatial resolution (QB, 2.5 m ground spatial distance) was more effective in tamarisk delineation. Despite the success in the reported studies, vegetation patch detection from remotely sensed imagery is not always easy. Our objective of this study is to: (1) detect the circular or elliptic patches based on shape difference using ALOS fusion image; (2) calculate the spatial structure parameters of vegetation patches, such as numbers and areas and directions of patches.

2. Materials and methods

2.1. Materials

The Advanced Land Observing Satellite (ALOS) was launched on January 24, 2006, which has three remote sensing instruments: the Panchromatic Remote-sensing Instrument for Stereo Mapping (PRISM) for digital elevation mapping, the Advanced Visible and Near Infrared Radiometer type 2 (AVNIR-2) for precise land coverage observation, and the Phased Array type L-band Synthetic Aperture Radar (PALSAR) for day-and-night and all-weather land observation [16]. PRISM is a panchromatic radiometer with 2.5m spatial resolution and one band (wavelength: 0.52 to 0.77 micrometers) at nadir. AVNIR-2 is a visible and near infrared radiometer with 10m spatial resolution and four bands (Band 1 : 0.42 to 0.50 micrometers; Band 2 : 0.52 to 0.60 micrometers; Band 3 : 0.61 to 0.69 micrometers; Band 4 : 0.76 to 0.89 micrometers) at nadir. Firstly the ALOS image acquired in 2008 was selected. Then, its panchromatic band with 2.5m spatial resolution was fused with four visible and near infrared bands with 10m spatial resolution to become the image with four visible and near infrared bands with 2.5m spatial resolution. Finally, the subset of the fusion-ready ALOS image (300*300 pixels; area is about 56 ha; with 2.5m spatial resolution and four bands) was used in this paper.

2.2. Methods

The false color image could be gotten through the RGB combination of Band 4 and Band 3 and Band 2 (Fig. 1a) or the RGB combination of Band 4 and Band 2 and Band 1 (Fig. 1b). The experiments in this paper indicated that the latter was in favour of vegetation patch detection. So the latter was used as the basis for method experiments. Gonzalez R. C. and Woods R. E. (1992) [17] expounded that RGB colour space couldn’t model the visual sense of human being to colour and not suitable for colour analysis of an image, and often RGB colour space was transformed into other colour space. HSV colour space was used in this paper, which was from linear transformation of RGB color space and an even color model suitable

898 Qingsheng LIU et al. / Procedia Environmental Sciences 10 ( 2011 ) 896 – 901

for color analysis. H is location of colour information, and S is a ratio between the purity of selected colour and the maximum purity of this colour, and V is luminance of colour.

Fig. 1. (a) the left picture is the false color RGB432 image; (b) the right picture is the false color RGB421 image

Firstly, the RGB421 image was transformed from RGB into HSV color space by using the RGB2HSV function in MATLAB software. Fig. 2a was S component of HSV transformation of RGB421, which reflected the difference between vegetation patches and background very well. But lots of unclosed edges existed through directly using canny edge detector based on the S component image (Fig. 2b). So the S component image must be preprocessed before edge detection. There are many edge detectors in image processing. Ali, M. and Clausi D. reviewed the different edge detection techniques in image processing, and concluded that the implemented canny edge detector presented the best performance both visually and quantitatively [18]. The canny edge detector uses the Laplacian operator, and can detect the strong and thin edges through using two different thresholds respectively, which is easy for detecting the thin edges. So the Canny edge detector is used to find the edges in this paper.

Secondly, the S component image was enhanced through adjusting the gray range from [0.2 0.8] to [0 1] using the IMADJUST function in MATLAB software, which enhanced the edge difference between vegetation patches and background. Then Wiener filter (3 3) was used to remove the noise of the S component image after image enhancement (Fig. 3a). Then canny edge detector was used to find the edges of the image. Compared with Fig. 2b, the unclosed edges decreased after image enhancement (Fig. 3b).

Thirdly, the big stripe and rectangle patches and non-vegetation objects were removed according to the threshold of area of targets (if the area of target was greater than 100 pixels or less than 10 pixels, the target was removed). Because some patches connected with each other by the thin stripes, their areas were probably greater than 100 pixels. According to the given thresholds, they were falsely removed, which produced some detection errors.

Fourthly, traced the exterior edges of targets and label the targets in Fig. 3b using BWBOUNDARIES function in MATLAB software. Then the centroid coordinates and areas of the targets are calculated and saved.

Finally, the row and column coordinates of targets were calculated and saved. Then the height and width and actual areas of targets were calculated through accumulation of pixels of targets, and the

899 Qingsheng LIU et al. / Procedia Environmental Sciences 10 ( 2011 ) 896 – 901

theoretical areas of the targets were calculated according to the area equation of ellipse too. The ratio between width and height and the ratio between the actual and theoretical area of each target were calculated. When the ratio between the actual and theoretical area of the target was greater than 0.1 and less than 1.3, the vegetation patches were detected very well and the detected number was 100 in Fig. 4a. When the ratio between width and height of the target was less than 1.0, the direction of the target was approximately south-north, and the ratio was greater than 1.0, the direction of the target was approximately east-west, and the ratio equaled to 1.0, the target hadn’t the obvious direction. The direction of 19 of 100 patches was approximately south-north, and the other 81 patches hadn’t the obvious direction, which indicated that the vegetation patches hadn’t the main distribution direction in the study area.

Fig. 2. (a) the left picture is the original S component image; (b) the right picture is the image after canny edge detector

Fig. 3. (a) the left picture is the S component image after enhancement and wiener filter; (b) the right picture is the edge image

900 Qingsheng LIU et al. / Procedia Environmental Sciences 10 ( 2011 ) 896 – 901

Fig. 4. (a) the left picture is the detected vegetation patches with the blue *; (b) the right picture is the image found by LMM

3. Conclusions

Use the Local Maximum Method (LMM) to find the regions that had the local maximum great than the given threshold (threshold=0.2), which was done using the IMEXTENDEDMAX function in MATLAB software. The result was shown in Fig. 4b and used to assist the visual interpretation. The number of the real vegetation patches was 75, which resulted from the interactive visual interpretation of Fig. 2a and Fig. 4b. The 29 of all 100 detected patches were detected in error and 4 patches weren’t detected. The detection accuracy of the patches is about 89.3% (the ratio between number of detected patches and real patches, ((100-29-4)/75).

The detection errors mostly came from three aspects. First aspect was the discrimination between

vegetation and the background. About six patches weren’t detected during the edge extraction. Second aspect is the connectivity between the patches. Because of this reason, five patches couldn’t be detected. Third aspect was from short and long threads, and eighteen patches were detected in error. In addition, four patches weren’t detected because of their shapes and the unclosed characters. The errors from fist aspect could be reduced through using the spectral and shape differences between vegetation patches and background together. The errors form second aspect could be reduced through image segmentation. The errors from third aspect could be reduced through using the curvature difference between ellipse and thread. Of course, these aspects need be studied in the future. In addition, the numbers and areas and centroids and directions and spatial structure models of the vegetation patches resulted from the experiment are important parameters for landscape ecological research in the study area, and their relationships need be studied in detail in the future. Although this study only uses small subset of ALOS fusion image to detect vegetation patches, the results suggest that vegetation patches can be identified using high spatial resolution satellite images. This method needs to be validated and improved with field data in large regions as part of an operational vegetation management program.

Acknowledgements

901 Qingsheng LIU et al. / Procedia Environmental Sciences 10 ( 2011 ) 896 – 901

This research work was jointly supported by a grant from the State Key Laboratory of Resource and Environment Information System (Project Number: 088RA400SA) and a grant from the National Natural Science Foundation of China (Project No. 40771172). The authors thank for Dr. Hong Wang from Hohai University, China for providing ALOS data.

References

[1] Wu Jianguo. Landscape Ecology-Pattern, Process, Scale and Level, 2nd ed. Higher Education Press, Beijing, China, 20-25 2000. (in Chinese)

[2] Wu J, Loucks O L. From balance-of-nature to hierarchical patch dynamics: a paradigm shift in ecology. The Quarterly Review of Biology 1995; 70(4), 439-466.

[3] Gao Qun, Yao Yonghui, Zhou Chenghu and Cheng Weiming. Character of landscape structure in the ecotone between agriculture and animal husbandry in northern China using SPOT-5 remote sensing data. Resources Science. 2004; 26(6), 9-16. (in Chinese with English abstract)

[4] Jiang Bo, Yuan Weigao, Qi Lianzhong, Zhu Jinru and Shen Aihua. Patch distribution character of beach vegetation and its ecological importance. Ecologic Science. 2005; 24(2), 107-112. (in Chinese with English abstract)

[5] Liu Guixia and Han Jiangguo. Research advances in grassland gap. Chinese Journal of Ecology. 2006; 25(9), 1136-1140. (in Chinese with English abstract)

[6] Hou Biqing. Analysis of landscape structural features of Zhuzhou urban greenland based on fine-resolutioned satellite image. Scientia Silvae Sinicae. 2007; 43(3), 127-132. (in Chinese with English abstract)

[7] Wang Zhitai, Wang Zhijie and Bao Yu. Urban greenland landscape structure in Tongren based on Quickbird. Journal of Northwest Forestry University. 2010; 25(1), 166-170. (in Chinese with English abstract)

[8] Zeng Fan ru, He Zheng wei and Li Zhe. Application of MATLAB software and ALOS remote sensing imagine to automatic recognition of earthquake damage building: take Dujiangyan urban as an example. Northwestern Seismological Journal 2009; 31(4), 327-332. (in Chinese with English abstract)

[9] Fang Lei and Cai Guangcheng. Application of mathematical morphology in the recognition and localization of circle object. Journal of Yunnan Normal University. 2006; 26(6), 9-11. (in Chinese with English abstract)

[10] Lin Hongjin, Luo Daisheng, Luo Feng and Xia Xin. A meothod of ellipse target recognition for SAR Image. Computer Engineering and Application. 2006; (14), 48-49, 88. (in Chinese with English abstract)

[11] Mao Xia, Zhao Xingyuan and Shen Wei. Extraction method of rectangle building based on fussy decision-making. Journal of Beijing University of Aeronautics and Astronautics. 2008; 34(6), 699-702. (in Chinese with English abstract)

[12] Perez A. J., Lopez F., Benlloch J. V., and S. Christensen. Colour and Shape analysis techniques for weed detection in cereal fields. Computers and Electronics in Agriculture 2000; 25, 197-212.

[13] McGlynn I. O. and G. S. Okin. Characterization of shrub distribution using high spatial resolution remote sensing: ecosystem implications for a former Chihuahuan Desert grassland. Remote Sensing of Environment 2006; 101, 554-566.

[14] Vogt P., Riitters K. H., Estreguil C., Kozak J., Wade T. G., and J. D. Wickman. Mapping spatial patterns with morphological image processing. Landscape Eco 2007; l 22, 171-177

[15] Carter G. A., Lucas K. L., Blossom G. A., Lassitter C. L., Holiday D. M., Mooneyhan D. S., Fastring D. R., Holcombe T. R., and J. A. Griffith. Remote sensing and mapping of tamarisk along the Colorado River, USA: a comparative use of summer-acquired Hyperion, Thematic Mapper and QuickBird data. Remote Sensing 2009; 1, 318-329.

[16] JAXA EORC. About ALOS. http://www.eorc.jaxa.jp/ALOS/en/about/about_index.htm. 2011 [17] Rafael C. Gonzalez, Richard E. Woods. Digital image procesing 2nd ed. Addison-Wesley Publishing Company. USA, 1993 [18] Ali, M. and Clausi, D. Using the Canny edge detector for feature extraction and enhancement of remote sensing images.

Proc. IGARSS’01, 2001, p. 2298-2300