Asian Journal of Applied Sciences (ISSN: 2321 – 0893) Volume 03 – Issue 02, April 2015 Asian Online Journals (www.ajouronline.com ) 218 Maximizing Urban Features Extraction from Multi-sensor Data with Dempster-Shafer Theory and HSI Data Fusion Techniques Mohammed Oludare Idrees 1* , Vahideh Saeidi 1 , Biswajeet Pradhan 2 , Helmi Zulhaidi Mohd Shafri 2 1 Department of Civil Engineering, Faculty of Engineering, Universiti Putra Malaysia, 43400 UPM, Serdang, Selangor Darul-Ehsan, Malaysia 2 Geospatial information Science Research Center (GISRC), Faculty of Engineering Universiti Putra Malaysia, 43400 UPM, Serdang, Selangor Darul-Ehsan, Malaysia * Corresponding author’s email: dare.idrees [AT] gmail.com _________________________________________________________________________________ ABSTRACT— This paper compares two multi-sensor data fusion techniques – Dempster-Sharfer Theory (DST) and Hue Saturation Intensity (HSI). The objective is to evaluate the effectiveness of the methods interm in space and time and quality of information extraction. LiDAR and hyperspectral data were fused using the two methods to extract urban land scape features. First, digital surface model (DSM), LiDAR intensity and hyperspectral image were fused with HSI. Then the result was classified into five classes (metal roof building, non-metal roof building, tree, grass and road) using supervised classification (minimum distance) and the classification accuracy assessment was done. Second, Dempster Shafer Theory (DST) utilized the evidences available to fuse normalized DSM, LiDAR intensity and hyperspectral derivatives to classify the surface materials into five classes as before. It was found out that DST perform well in the ability to discriminate different classes without expert information from the scene. Overal accuracy of 87% achieved using DST. While in HSI technique, the overal accuracy obtained was 74.3%. Also, metal and non-metal roof types were clearly classified with DST which, does not have a good result with HSI. A fundamental setback of HSI is its limitation to fusion of only two sensor data at a time whereas we could integrate different sensor data with DST. Besides, the time required to select trainimg site for supervised classificition, the accuracy of feature classification with HSI fused data is dependent on the knowledge of the analyst about the scene with the other one. This study shows DST to be an accurate and fast method to extract urban features and roof types. It is hoped that the increasing number of remote sensing technology transforming to era of redundant data will make DST a desired technique available in most commercial image processing software packages. Keywords— Data fusion, Feature extraction, urban mapping, Hyperspectral, LiDAR _________________________________________________________________________________ 1. INTRODUCTION The global surge in rural-urban migration in search of better live, employment opportunities, and education exerts pressure on the population of urban environment leading to saturation of small spaces with various artificial surface materials. According to United Nations Economics and Social Affairs report, as at 2010, Malaysia has a total population of 28,250,458 [1], out of which it is estimated that 73% will be living in urban areas by 2013 [2]. This factor has brought a lot of burden and congestion on the urban environment that necessitates effective planning to cope with the resultant impact on the socioeconomic and environmental challenges. Multispectral images such as IKONOS, QuickBird, AVHRR, SPOT, etc. have been widely used for land surface cover inventory of urban environment [3], [4]. However, limitation of spectral resolution render them inadequate to obtain detailed analysis of surface material within urban area [3], [5], [6]. In contrast to the multispectral sensors that collect reflected energy from surface material in discrete wavelength of usually less than 10 bands, hyperspectral sensors record spectral signature of surface material in dozens or hundreds as a series of narrow and continuous wavelength bands [7]. Advances in hyperspectral remote sensors give rise to new prospect for detailed inventory of surface material, particularly for geological, forestry, agricultural, and urban applications [8], [9]. Although hyperspectral is highly useful for diagnostic mapping, particularly for identifying the internal structure and/or chemical properties of material, cost of data collection and limited spatial coverage are a major drawback. There are several studies on hyperspectral for urban mapping [3], [4], [7], [10], [11] but the absence of height information create a void in some applications such as roof structure and material. Essentially, roof had been, from time immemorial, a shield that protects people and their properties against weather and other external threat [5]. Nonetheless, increasing concern for health and environmental sustainability have brings the use of roof covers made of environmentally friendly material into focus. This concerns range from assessment of pollutants on urban surfaces, rainwater, runoff, and underground water contamination which are dependent on slope and exposition of roof material