Dimensionality reduction and classification on Hyperspectral Image Using Python

Authors

* DR.T.Hitendra Sarma * Syam Kakarla

Prerequisites

The prerequisites to better understand the code and concept are:

* Python     * MatLab     * Linear Algebra

Installation

• This project is fully based on python. So, the necessary modules needed for computaion are:

* Numpy
    * Sklearn
    * Matplotlib
    * Pandas

• The commands needed for installing the above modules on windows platfom are:

pip install numpy
    pip install sklearn
    pip install matplotlib
    pip install pandas

• we can verify the installation of modules by importing the modules. For example:

import numpy
    from sklearn.decomposition import PCA 
    import matplotlib.pyplot as plt
    import pandas as pd

Results

* Here we are performing the the dimensionality reduction on one of the widely used hyperspectral image Indian Pines

• The result of the indianpinespca.py is shown below:

  * It initial result is a bargraph for the first 10 Pricipal Components according to their variance ratio's :

  Since, the initial two principal COmponents have high variance. so, we will select the initial two PC'S. * It second result is a scatter plot for the first 10 Pricipal Components is :

  

  * The above program resullts a dimensionally reduced csvfile .

  • The result of the indianpines_knnc.py is given below:
  * The above program will classify the Indian Pines dataset before Principal Component Analysis(PCA). The classifier here used for classification is K-Nearest Neighbour Classifier (KNNC) * The time taken for classification is:

  * Then the classification accuracy of indian pines dataset before PCA is:

  • The result of the indianpinesknncafter_pca.py

    * Then the resultant classification accuracy of indian pines dataset after PCA is:

    Conclusion :

    * By performing PCA on the corrected indian pines dataset results 100 Principal Components(PC'S). * since, the initial two Principal Components(PC'S) has 92.01839071674918 variance ratio. we selected two only. * Initially the dataset contains the dimensions 21025 X 200 is drastically reduced to 21025 X 2 dimensions. * The time taken for classification before and after Principal Component Analysis(PCA) is: | Dataset | Accuracy | Time Taken | | ------------- |:-----------: | ----------:| | Before PCA | 72.748890 | 17.6010 | | After PCA | 60.098187 | 0.17700982 | * Hence, the time has been reduced with a lot of difference and the classification accuracy(C.A) also reduced but the C.A can increased little bit by varying the 'k' value.

    License

    This project is licensed under the MIT License - see the LICENSE.md file for details