Abstract
Cabadbaran City is the capital of Agusan del Norte which is located at the north eastern portion of Mindanao, Philippines. Almost
30% of its land area is devoted to agriculture (mainly rice, corn, coconut, banana, vegetables and abaca). Currently, the city
government and agriculture office are implementing programs focusing on improving coconut and vegetable productivity,
controlling banana disease and infestation, and enhancing abaca production industry. In support of decision making, the current
situation must first be assessed by answering the basic questions what and where through detailed and accurate resource mapping.
In this study, only discrete LiDAR datasets were utilized. Corresponding orthophotos were used only for training and validation.
Land cover classification was done using two workflows using Support Vector Machines (SVM) as the classifier. In the first
workflow, land cover classes were classified using rasterized point cloud metrics such as minimum, maximum, standard deviation,
skewness, kurtosis, quartile average, mode and median. In the second workflow, point cloud analysis was used to derive additional
features for classification which was integrated and executed in the same object-based software through Cognition Network
Language (CNL). The derivations of the additional features were conducted after running an initial segmentation which means that
the distribution of points was analysed within the delineated objects. Classes that benefited to point cloud-based metrics are mostly
non-ground agricultural classes namely coconut, mango and palm trees. These classes have obtained increase in accuracies by an
average of 11.2% using validation sample set 1 and an average of 18.2% using validation sample set 2. Ground classes, particularly
barren land and rice, appeared to be incompatible to these point cloud metrics as shown by the decrease in accuracies for Methods 2
and 3 by about 18.1% using validation sample set 1 and about 16.4% using validation sample set 2. Exploring other useful point
cloud-based metrics and testing on sites with other land cover classes are highly recommended.
30% of its land area is devoted to agriculture (mainly rice, corn, coconut, banana, vegetables and abaca). Currently, the city
government and agriculture office are implementing programs focusing on improving coconut and vegetable productivity,
controlling banana disease and infestation, and enhancing abaca production industry. In support of decision making, the current
situation must first be assessed by answering the basic questions what and where through detailed and accurate resource mapping.
In this study, only discrete LiDAR datasets were utilized. Corresponding orthophotos were used only for training and validation.
Land cover classification was done using two workflows using Support Vector Machines (SVM) as the classifier. In the first
workflow, land cover classes were classified using rasterized point cloud metrics such as minimum, maximum, standard deviation,
skewness, kurtosis, quartile average, mode and median. In the second workflow, point cloud analysis was used to derive additional
features for classification which was integrated and executed in the same object-based software through Cognition Network
Language (CNL). The derivations of the additional features were conducted after running an initial segmentation which means that
the distribution of points was analysed within the delineated objects. Classes that benefited to point cloud-based metrics are mostly
non-ground agricultural classes namely coconut, mango and palm trees. These classes have obtained increase in accuracies by an
average of 11.2% using validation sample set 1 and an average of 18.2% using validation sample set 2. Ground classes, particularly
barren land and rice, appeared to be incompatible to these point cloud metrics as shown by the decrease in accuracies for Methods 2
and 3 by about 18.1% using validation sample set 1 and about 16.4% using validation sample set 2. Exploring other useful point
cloud-based metrics and testing on sites with other land cover classes are highly recommended.
| Original language | English |
|---|---|
| Title of host publication | Proceedings of GEOBIA 2016 : Solutions and synergies, 14-16 September 2016, Enschede, Netherlands |
| Editors | N. Kerle, M. Gerke, S. Lefevre |
| Place of Publication | Enschede |
| Publisher | University of Twente, Faculty of Geo-Information Science and Earth Observation (ITC) |
| Number of pages | 7 |
| ISBN (Print) | 978-90-365-4201-2 |
| DOIs | |
| Publication status | Published - 14 Sep 2016 |
| Externally published | Yes |
| Event | 6th International Conference on Geographic Object-Based Image Analysis, GEOBIA 2016: Solutions & Synergies - University of Twente Faculty of Geo-Information and Earth Observation (ITC), Enschede, Netherlands Duration: 14 Sep 2016 → 16 Sep 2016 Conference number: 6 https://www.geobia2016.com/ |
Conference
| Conference | 6th International Conference on Geographic Object-Based Image Analysis, GEOBIA 2016 |
|---|---|
| Abbreviated title | GEOBIA |
| Country | Netherlands |
| City | Enschede |
| Period | 14/09/16 → 16/09/16 |
| Internet address |