Exploration of the OBIA methods available in SPRING non-commercial software to UAV data processing

• Type: Article in International Conference Proceedings Book
• Date: 2014

Authors

Teodoro, AC (Author) FCUP; Araujo, R (Author) Other

Conference proceedings International

Title: EARTH RESOURCES AND ENVIRONMENTAL REMOTE SENSING/GIS APPLICATIONS V

Conference on Earth Resources and Environmental Remote Sensing/GIS Applications V

Amsterdam, NETHERLANDS, SEP 23-25, 2014

Indexing

ISI Proceedings

ISI Web of Knowledge - 4 Citations

Scopus - 8 Citations

Scientific classification

FOS: Engineering and technology > Other engineering and technologies

CORDIS: Technological sciences > Technology

Other information

Authenticus ID: P-00A-5SC

DOI: 10.1117/12.2066468

Abstract (EN): The increasing developments in Unmanned Aerial Vehicles (UAVs) platforms and associated sensing technologies, offer a broad range of solutions for different applications related to the acquisition of information about objects or phenomenon at the Earth. The huge amount of data, provided by UAVs, represents a new challenge regarding developments of image processing techniques. Object-based image classification (OBIA) is highly suitable for very high resolution imagery, where pixel-based classification is less successful due to the high spatial variability within objects of interest. An OBIA approach using SPRING (R) non-commercial software was implemented in this work. The UAV system used was a Swinglet from Sensefly. The ortho-mosaic, with 0.04 m of pixel size, from 20 of January of 2012 of Coimbra (Portugal) region with an apx. 500x400 m area was processed using the original 41 images. Different "similarity" and "area" parameters combination were computed in the segmentation stage (region-based). Firstly, a supervised classification was employed, considering 7 classes based on Corine land cover nomenclature. For several parameter combinations were obtained a Kappa>0.9 and an overall accuracy >90%. However, several objects were not classified. An unsupervised classification was performed and 27 classes were defined. After, a new supervised classification was performed considered 22 of the 27 classes identified, with an overall accuracy of 82.58%, and a Kappa of 0.817. We conclude that the algorithms employed in this work are not the most suitable for this kind of spatial resolution. The use data mining algorithms could improve the results.

Language: English

Type (Professor's evaluation): Scientific