Open AccessOBJECT BASED AGRICULTURAL LAND COVER CLASSIFICATION MAP OF SHADOWED AREAS FROM AERIAL IMAGE AND LIDAR DATA USING SUPPORT VECTOR MACHINE
Author(s) -
R. T. Alberto,
S. C. Serrano,
G. B. Damian,
E. E. Camaso,
Arnel B. Celestino,
P. J. C. Hernando,
M. F. Isip,
K. M. Orge,
M. J. C. Quinto,
R. C. Tagaca
Publication year - 2016
Publication title -
isprs annals of the photogrammetry, remote sensing and spatial information sciences
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.356
H-Index - 38
eISSN - 2194-9042
pISSN - 2196-6346
DOI - 10.5194/isprsannals-iii-7-45-2016
Subject(s) - orthophoto , artificial intelligence , land cover , segmentation , support vector machine , pattern recognition (psychology) , computer science , scale (ratio) , image segmentation , aerial image , computer vision , contextual image classification , lidar , kernel (algebra) , shadow (psychology) , pixel , remote sensing , geography , image (mathematics) , cartography , mathematics , land use , psychology , civil engineering , engineering , psychotherapist , combinatorics
Aerial image and LiDAR data offers a great possibility for agricultural land cover mapping. Unfortunately, these images leads to shadowy pixels. Management of shadowed areas for classification without image enhancement were investigated. Image segmentation approach using three different segmentation scales were used and tested to segment the image for ground features since only the ground features are affected by shadow caused by tall features. The RGB band and intensity were the layers used for the segmentation having an equal weights. A segmentation scale of 25 was found to be the optimal scale that will best fit for the shadowed and non-shadowed area classification. The SVM using Radial Basis Function kernel was then applied to extract classes based on properties extracted from the Lidar data and orthophoto. Training points for different classes including shadowed areas were selected homogeneously from the orthophoto. Separate training points for shadowed areas were made to create additional classes to reduced misclassification. Texture classification and object-oriented classifiers have been examined to reduced heterogeneity problem. The accuracy of the land cover classification using 25 scale segmentation after accounting for the shadow detection and classification was significantly higher compared to higher scale of segmentation.