SOURCE:

Faculty: Physical Sciences
Department: Computer Science

CONTRIBUTORS:

Olanloye, D. Odunayo
Anigbogu S. O.

ABSTRACT:

Advanced precision engineering has enabled the development of hyperspectral sensors. These sensors can be used in the collection of highly detailed spectra information from required target using non-intrusive, zero-impact and remote sensing methods. A particular usage is the use of hyperspectral sensors on satellites to obtain information about earth surfaces and minerals beneath the surface. However, the data so-obtained consist of high dimension (hundreds of bands) spectrum that needs to be processed to obtain classification of data and thus identify the present minerals. Thus, hyperspectral data dimension reduction has become an active research topic. In this work, we use a new metaheuristic, namely halfing method, to generate a characterization map of a given spectrum. The method literally ratios one half of the spectrum against the other, thus attenuating defects due to measurement conditions. Fuzzy c-means clustering was modified to obtain cluster centres from the characterization map. The obtained cluster centres were then used to classify the pixels (minerals) using unsupervised learning algorithm obtained by modifying Kohonen Self Organising Map (KSOM). Minerals of known spectral library were used to train Adapted Neuro Fuzzy Inference System (ANFIS) which was linked up with KSOM to identify the pixel (mineral) in each of classes earlier identified by KSOM. The quantity of Pixels (Mineral) in each class were also obtained. Novel minerals were also identified by the network. The instrument used for the implementation is R programming language. This research work can go a long run to boost mining sector in terms of mineral prospecting. With this software developed, it is possible to detect various mineral in a particular place in any part of the world provided their hyperspectral data is available.