This project deals with blind source separation in the aspect of hyperspectral images.
Abstract
This project deals with blind source separation in the aspect of hyperspectral images. the project examines the effectivness of a few separation systems, under the assumption of independent component analysis and considering that the mixture was done in a linear way. in order to get better results there is a process of filtering the mixtures and passing them through a certain preprocessing which will get rid of noise and help us get a better idea regarding the sources of the mixture.
The problem
The problem in BSS is dealing with an unfamiliar mixture. since we don’t have a-priori knowledge regarding the character and number of sources making the mixture and also have to deal with potential noise and other interruptions, unmixing the signals to the sources making the mixture becomes not so trivial. in hyper spectral analysis we deal with an image taken in multiple wavelengths and the goal is to unmix the sources according to their different reactions to the varying wavelengths.
ICA (independent component analysis)
X=AS+N this equations describes the mixture including the gaussian noise. after preprocessing we send it to the ica and get the inverted A matrix. multiplying the original equation would retrieve the matrix of sources. Add here a brief description including block diagram(s)/ and/or flow chart(s) of the project. The text should be as self contained as possible, avoiding special abbreviations or terms.
A good test for “”special”” is whether you were familiar with the term/abbreviation before the beginning of the project.
Figure 1 block diagram explaining the process of separation including the preprocessing part
Tools
The project was developed in a MATLAB 6.5 environment.
Figure 2 the parts from LENA &PEPPY as they were used in the unmixing simulation
Figure 3 example from results of bss unmixing of aviris images
Conclusions
The project is very helpful in order to discover diffrent sources assembling the land without the need to dig inside and start guessing. using the results of the blind source seperation we get a better knoweledge about the soil and save a lot of time and money searching for unique soils.
Acknowledgment
We are greateful to our project supervisor bronstein alex for his help and guidance throughout this work, and for showing us the innovative field of BSS.
We are also grateful to the Ollendorf Minerva Center for supporting this project.