Arboriculture & Urban Forestry 42(6): November 2016 URBAN TREE CROWN DELINEA- TION AND COUNTING Counting of trees via remote sensing is crucial in urban forest research because of plantation and ur- ban management. As field measurements for tree counting are time consuming and may be expensive, remote-sensing imaging may be a suitable technique to overcome this limitation (Shafri et al. 2011). Several studies on tree crown delineation and tree counting have been conducted via remote sensing (Brandtberg and Walter 1998; Andersen et al. 2001; Pouliot et al. 2002; Culvenor 2003; Erikson 2004; Heurich and Weinacker 2004; Karantzalos and Argialas 2004; Mei and Dur- rieu 2004; Pouliot et al. 2004; Falkowski et al. 2006; Horvath 2007; Wolf and Heipke 2007; Shafri et al. 2011; Katoh and Gougeon 2012; Wu et al. 2012). However, studies on urban areas remain limited and have only been conducted recently (Iovan et al. 2008; Ardila et al. 2010; MacFaden et al. 2011; Ardila et al. 2012; Latif et al. 2012; Yao and Wei 2013; Iovan et al. 2014). Remote sensing provides valuable informa- tion for urban forests, but conducting automatic tree extraction and counting from images is dif- ficult because of the complexity of urban spaces. Furthermore, the resolution of multispectral images can affect the level of tree crown delin- eation (Iovan et al. 2008). For example, if one applies the pixel-based classification on high- resolution imaging, the crowns of individual tree species will be detected (Leckie 1990; Beau- bien 1994; Erikson 2004). However, the low- spatial resolution imaging can only detect the strands of single species (Gillis and Leckie 1993). Tree crown delineation should be conducted to count trees. This process presents the follow- ing limitations in urban areas: complexity of urban areas, including trees, buildings, roads, and sidewalks; different physical characteris- tics of trees, such as crown width, crown shape, height, and canopy cover; and different pat- terns for tree plantation that indicate equally or irregularly distance variations between trees. Thus, spectral information should be mini- mized for tree identification and other information regarding trees (e.g., spatial, textural, or color) should be used. Several studies on tree crown delineation have been conducted by using the con- 407 text and contours of trees. Any information on the object that can characterize its state and situa- tion, which lead to its identification, is called con- text (Abowd et al. 1999; Oliva and Torralba 2007). Ardilla et al. (2012) performed a study involving most techniques for tree crown delineation by employing geographic object-based image analysis (GEOBIA) technique. Topologic relations between adjacent image objects and segmentation method can be established using GEOBIA. The following are the procedures to reach the tree crown: masking of grassland areas through size and NIR segmenta- tion; identifying tree crown objects through NDVI, which is insensitive to the inside-crown bright- ness because of sun illumination and tree struc- ture (Liang 2004); identifying individual trees with high background contrast; clustering trees based on size and NDVI segmentation; detecting small trees along roads through shadow, shape, and spec- tral characteristics [the use of the tree shadow is a common method to delineate a tree contour (Geu- geon and Leckie 2001)]; joining trees in linear and curvilinear patterns and detecting trees with low background contrast through local maxima filter- ing (Wulder et al. 2004); and combining the results via GEOBIA to identify tree crowns. Overall, 85% of the tree crown was detected by this method, and the error may be caused by the spatial resolution of the image, because small trees can cover less than two multispectral pixels (Gougeon and Leckie 2006; Ardilla et al. 2012). Nevertheless, most stud- ies on tree crown detection have been conducted using high-resolution imaging (Ardila et al. 2012). The other methods to delineate tree contours include the region-growing method (Geugeon 1995; Iovan et al. 2008), valley-following algo- rithms (Erikson 2004), multi-analysis (Brandt- berg and Walter 1998; Ardilla et al. 2012), and active contours (Horvath et al. 2006). The seeded- region growing, developed by Iovan et al. (2008), is one example of region growing. This method is based on two steps, namely, seed point detection through DSM to evaluate treetops and the region- growing approach based on geometric criteria (height descent). Finally, the result showed that the accuracy for tree crown delineation by using this method was 78%. Tree crown delineation is based on 3D height information, and the cause of the error may be the low accuracy of the DSM. ©2016 International Society of Arboriculture
November 2016
| Title Name |
Pages |
Delete |
Url |
| Empty |
Search Text Block
Page #page_num
#doc_title
Hi $receivername|$receiveremail,
$sendername|$senderemail wrote these comments for you:
$message
$sendername|$senderemail would like for you to view the following digital edition.
Please click on the page below to be directed to the digital edition:
$thumbnail$pagenum
$link$pagenum
Your form submission was a success. You will be contacted by Washington Gas with follow-up information regarding your request.
This process might take longer please wait
