Spectral Library and Discrimination Analysis of Indian Urban Materials
- PDF / 1,016,174 Bytes
- 11 Pages / 595.276 x 790.866 pts Page_size
- 84 Downloads / 183 Views
RESEARCH ARTICLE
Spectral Library and Discrimination Analysis of Indian Urban Materials Shailesh Shankar Deshpande1
•
Arun B. Inamdar2 • Harrick M. Vin1
Received: 22 November 2017 / Accepted: 4 January 2019 Ó Indian Society of Remote Sensing 2019
Abstract In this paper, we present a spectral library of urban materials and its detailed spectral analysis. The primary focus of the research is spectral study of the local urban materials and their discrimination using field signatures. Further, we develop an algorithm for identifying the most important wavelength range, and its distribution. Instead of common analysis methods which focus on single wavelength, we focus on wavelength range as it is difficult for urban material to find out single diagnostic wavelength. Novelty of our algorithm is twofold: first we use Leodoit–Wolf covariance estimator for improving accuracy, and second we introduce two new metrics based on Bhattacharyya distance. The spectral discrimination analysis found that the significant wavelength ranges for discriminating urban classes are spread all over the spectrum with slight bias for visible range. Though it is challenging to discriminate materials belonging to the same class, for example, different types of concrete pavements, the broad-level classes such as soil, urban vegetation, metal roofs and concrete are well separable. The confusion between bright soil and concrete surfaces is difficult to overcome spectrally. The developed spectral library is available at OGC compatible website splibtarang.com/index.php. Keywords Hyperspectral library Urban spectroscopy Bhattacharyya distance Hyperspectral feature selection Urban materials Spectral discrimination analysis
Introduction A comprehensive spectral library of required materials is an important knowledgebase for many remote sensing studies. Spectral signatures of materials within the library provide a model of unique spectral properties of the materials. These unique properties, in addition to spatial properties of the image, are exploited by different supervised and unsupervised classifiers for analysing remotely sensed data. The spectral properties can be exploited further for developing sensors covering a particular narrow or wide spectral range as required. The spectral libraries also constitute a critical resource of reference signatures for hyperspectral target detection of the materials.
& Shailesh Shankar Deshpande [email protected] 1
Tata Consultancy Services, Research and Innovation, Tata Research Development and Design Centre, 54-B Hadapsar Industrial Estate, Pune 411013, India
2
Centre of Studies in Resources Engineering, Indian Institute of Technology Bombay, Powai, Mumbai 400 076, India
Though there exist the popular spectral libraries like USGS’ Splib (Clark et al. 2007) and JPL’s ASTER (Baldridge 2009), multiplicity of the spectral resources is required for two main reasons: a) the public domain libraries may not have signatures of the distinct regional materials and b) even if the signatures of b
Data Loading...