SOURCE:

Faculty: Environmental Sciences
Department: Surveying & Geo-informatics

CONTRIBUTORS:

Ugbelase, V. N.
Igbokwe, J.I.

ABSTRACT:

The availability of Very High Resolution (VHR) images has made Pixel based classification method difficult, especially, in spectrally homogeneous areas. Object-based image analysis (OBIA) technique has solved this problem by incorporating both spectral and spatial characteristics of objects. In recent years, significant amount of research has been carried out on incorporating LiDAR derived normalized Digital Surface Model (nDSM) into the image classification to address the problems of differentiating spectrally similar objects in urban areas. Therefore, the aim of this study is to evaluate pixel and object based techniques in urban mapping using very high resolution (VHR) image and LiDAR data. It also evaluates the combination of nDSM derived from LiDAR data and high-resolution GeoEye-1 satellite imagery for classifying urban land cover in Rivers State, Nigeria. Seven study sites with varying levels of urbanization were clipped out of the Geo-eye image and georectified using the LiDAR data. The Pixel based supervised Maximum Liklihood Classification (MLC) was performed in Erdas 9.2 software while the eCognition developer software 9.0 was used to perform the two independent OBIA in the study. The OBIA procedure involved multiresolution segmentation with user-defined parameters of scale, shape and compactness. Ruleset was created and used for the object classification. Three experiments, supervied Pixel based, OBIA using VHR image and OBIA using VHR/LiDAR were carried out. The nDSM enabled accurate discrimination of elevated and non-elevated urban features having similar spectral characteristics while NDVI and NDWI were useful in discriminating vegetation and water respectively. The nDSM combined with the VHR image was used for the 3-D analysis of the study sites. The study developed an algorithm for the extraction of unpaved roads. Finally, in order to assess the validity of the classification results, accuracy assessment was performed through comparing randomly distributed reference points of GeoEye-1 imagery with the classification results, forming the confusion matrix and calculating producers, users, overall accuracy and Kappa coefficient. The results showed that the OBIA had results that were significantly better than the traditional Pixel based supervised method. On the other hand the addition of nDSM derived from LiDAR in the second OBIA significantly increased the accuracy over the OBIA which used only the VHR image. The pixel based method had average overall accuracy of 74.43% and 0.674 kappa while OBIA which used only VHR had 82.43% and 0.791 kappa and finally the OBIA with the addition of nDSM topped with 91.14% and 0.905 kappa. The hypotheses tests were used to validate the research questions.