Land cover classification using LiDAR intensity data and neural network

Urban land cover classification using airborne LiDAR data: A review

“…Given the nugget effect of salt-and-pepper noises in the pixel-based classification results, objectbased image segmentation and classification approach seems to be more appropriate. The object-based approach relies on a user-defined hierarchical structure to classify the segmented objects, and the technique has proven to outperform the traditional pixel-based classifiers not only on satellite images (Blaschke, 2010), but also on airborne LiDAR data El-Ashmawy, Shaker, & Yan, 2011;Huang et al, 2008;Minh & Hien, 2011;Sasaki et al, 2012). Various studies reported that an overall accuracy of over 80% can be achieved using object-based technique on LiDAR-derived surfaces (Antonarakis et al, 2008;Brennan & Webster, 2006;Im et al, 2008;MacFaden et al, 2012;O'Neil-Dunne, MacFaden, Royar, & Pelletier, 2013;Zhou, 2013).…”

“…Under such circumstance, direct geo-referencing technique can be applied to both image and LiDAR data. Some studies reported the use of very high resolution image such as QuickBird ) and WorldView (Kim & Kim, 2014;Minh & Hien, 2011) image together with airborne LiDAR data for land cover classification. In this case, ortho-rectification (or georeferencing) has to be performed as pre-processing on the satellite image.…”

“…For example, Yoon, Shin, and Lee (2008) and Yan et al (2012) addressed that a large variance of intensity values can be found in the tree features due to the irregular geometry of the canopy surface. Intensity data has a certain level of noise (Minh & Hien, 2011;Song et al, 2002), which may degrade the classification performance. Furthermore, radiometric misalignment found in overlapping LiDAR data strips may degrade the classification accuracy (Brennan & Webster, 2006;Yan & Shaker, 2014), see Fig.…”

“…Various classification techniques were reported in the literature to classify and extract these land cover features from the LiDAR data. Since most of the initial studies transformed the LiDAR data point cloud into image data, traditional supervised pixel-based classification techniques such as neural network (Minh & Hien, 2011;Nguyen, Atkinson, & Lewis, 2005;Priestnall et al, 2000), Gaussian mixture modeling (Charaniya et al, 2004), maximum likelihood (Bartels & Wei, 2006;Hodgson, Jensen, Tullis, Riordan, & Archer, 2003), and rule-based classification (Huang et al, 2008) were directly applied. Given the nugget effect of salt-and-pepper noises in the pixel-based classification results, objectbased image segmentation and classification approach seems to be more appropriate.…”

Comparison of heuristic and deep learning-based methods for ground classification from aerial point clouds

“…During this decade, the development of LiDAR sensors able to obtain better point resolutions, together with new processing techniques, have solved some of the previously stated issues. According to the review of Yan et al (2015), there are two ground classification techniques that outperform the others: (1) Object-based classifications, using hierarchical structures to carry out the segmentation (Minh and Hien 2011;Sasaki et al 2012), and (2) using classifiers that take into account contextual information of the points to be classified, such as Markov Random Fields and Conditional Random Fields (Cao et al 2012;Niemeyer, Rottensteiner, and Soergel 2012).…”