404 Shojanoori and Shafri: Review on the Use of Remote Sensing for Urban Forest Monitoring URBAN TREE DETECTION Tree detection by using remote-sensing images is the process of recognizing and classifying trees that lead to urban tree canopy and greenspace mapping (Lang et al. 2008; Walton et al. 2008; Johnson and Xie 2013). Mapping is conducted via urban for- estry monitoring methods, which can be classified into three groups, namely, visual interpretation and pixel- and object-based methods (Li et al. 2010). Although high-resolution data are valuable in extracting land-cover information, tree extraction and information collection are difficult in urban areas when traditional pixel-based image classi- fication methods are used. Traditional methods involve supervised and unsupervised classifications. Although unsupervised classification techniques, such as ISODATA and K-mean, are used for the- matic mapping (Langley et al. 2001; Xie et al. 2008; Sung 2012), these methods are rarely used for urban tree detection. Supervised classification methods, such as MLC, are oſten used to perform urban land- cover mapping (including vegetation cover map- ping) because they can be easily performed and provide accurate results (Ardila et al. 2010; Peijun et al. 2010; Shen et al. 2010; Zhang et al. 2010; Forzieri et al. 2013). The basis of MLC is a statistic classifi- cation of all pixels in each band to a specific class even when the threshold is defined. By contrast, MLC may cause some misclassifications in urban areas. For instance, some parts of the grass area are oſten classified as trees. Thus, filters, such as intra- class uniformity, inter-class contrast, and smooth- ness of boundaries between classes can be used to increase the contrast of features and obtain high classification accuracy (Ouma and Tateishi 2008). Minimum distance (MD) is another supervised classification for studies on urban forest, and sev- eral researchers believe that this method performs more accurate classification than other methods (Jusoff 2009; Latif et al. 2012). Shen et al. (2010) used three classification algorithms (i.e., ML, MD, and DT) for urban forest mapping and comparison of these algorithms showed that MD leads to the lowest classification accuracy and DT exhibits the highest accuracy among the three algorithms. The principle of DT classification differs from that of MLC, by which separation of the complicated deci- sion to several easier decisions is vital to achieve the required classification (Ouma and Tateishi 2008). ©2016 International Society of Arboriculture Vapnik developed a new method called support vector machine (SVM) in 1996. This method can classify urban areas because it can overcome several limitations, such as insufficient training data and low sensitivity to the sample size (Van Der Lindan et al. 2007; Mountrakis et al. 2011). In this regard, several studies on urban forests have used the SVM algorithm to detect vegetation (Lafarage et al. 2005; Iovan et al. 2008; Tigges et al. 2013; Iovan et al. 2014). These pixel-based classification algorithms may lead to low classification accuracy because of the high grade of spectral variability within land-cover classes, which are affected by sun angle, gaps in tree canopy, and shadows (Yu et al. 2006; Johnson and Xie 2013). A pixel, which is a small part of the clas- sification object, is the cause of within-class spec- tral variability in high-resolution images (Huang et al. 2007). Thus, the use of object-based classifica- tion is recommended to overcome this limitation. Object-based approaches can improve classifica- tion accuracy compared with visual interpretation and pixel-based methods. The object-based method can combine color, shape, spatial information, and contextual analysis to detect changes in vegetation (Li et al. 2010). The basis of object-oriented methods is image segmentation, which involves splitting the image into spatially continuous and homogeneous regions and leads to a reduction in local spectral variations (Lobo 1997). Li et al. (2010) combined segmentation and fuzzy multi-threshold classifica- tion to classify urban land covers, and the accuracy reached up to 93.72%. Fuzzy logic and intelligence techniques, such as artificial neural network (ANN), or integrated methods, such as adaptive Gaussian fuzzy-learning vector quantization (AGFLVQ), can also be used for urban forest detection and tree spe- cies identification (Hofle et al. 2012; Zhang and Qiu 2012). These classification methods not only detect urban forest but also distinguish urban tree spe- cies, and this ability is explained in the next section. Comparison of several methods used in urban tree detection showed that object-based classifi- cation provides the most accurate classification result. Although pixel-based methods, such as ML and SVM, are easy to operate and provide accurate results for tree mapping, they are not adequate for use in urban areas, particularly in distinguishing urban tree species (as explained in the following sec- tion) because the basis of this classification is a pixel.
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
