Mr. Magendran Thangavelu Profile

Magendran Thangavelu

at Anna Univ Chennai

SPIE Involvement:

Author

Publications (1)

Proceedings Article | 18 November 2014 Paper

Statistical and neural network analysis of hyperspectral radiometric data to characterise hematite of Singbhum iron ore belt, India

T. Magendran, Sanjeevi S.

Proceedings Volume 9263, 92630M (2014) https://doi.org/10.1117/12.2070388

KEYWORDS: Iron, Absorption, Statistical analysis, Hematite, Near infrared, Reflectivity, Radiometry, Neural networks, Mining, Chemical analysis

Read Abstract +

The demand for iron ore has increased in the recent years, thereby necessitating the adoption of rapid and accurate approaches to iron ore exploration and its grade-assessment. It is in this context that hyperspectral radiometry is seen as a potential tool. This paper examines the potential of hyperspectral radiometry in the visible, NIR and SWIR regions of the EMR to assess the grades of hematite of the western Singhbhum iron ore belt of eastern India, in a rapid manner. Certain spectro-radiometric measurements and geochemical analysis were carried out and the results have been presented. From the spectral measurements, it is seen that the strength of reflectance and absorption at definite wavelength regions is controlled by the chemical composition of the iron ores. It is observed that the primary spectral characteristics of these hematite lie in the 650-750nm, 850 to 900nm and 2130-2230nm regions. The laboratory based hyperspectral signatures and multiple regression analysis of spectral parameters and geochemical parameters (Fe₂O₃% and Al₂O₃%) predicted the concentration of iron and alumina content in the hematite. A very strong correlation (R²=0.96) between the spectral parameters and Fe% in the hematite with a minimum error of 0.1%, maximum error of 7.4% and average error of 2.6% is observed. Similarly, a very strong correlation (R²=0.94) between the spectral parameters and Al₂O₃% in the iron ores with a minimum error of 0.04%, maximum error of 7.49% and average error of 2.5% is observed. This error is perhaps due to the presence of other components (SiO₂, TiO₂, P₂O etc.) in the samples which can alter the degree of reflectance and hence the spectral parameters. Neural network based multi-layer perception (MLP) analysis of various spectral parameters and geochemical parameters helped to understand the relative importance of the spectral parameters for predictive models. The strong correlations (Iron: R²=0.96; Alumina: R²=0.94) indicate that the laboratory hyperspectral signatures in the visible, NIR and SWIR regions can give a better estimate of the grades of hematite in a rapid manner.

View contact details

UPDATE YOUR PROFILE

Is this your profile? Update it now.

Sign into your SPIE.org account

Don’t have a profile and want one?

Create an account on SPIE.org

Keywords/Phrases

Search In:

Publication Years