syamkakarla98 / dimensionality-reduction-and-classification-on-hyperspectral-images-using-python

In this repository, You can find the files which implement dimensionality reduction on the hyperspectral image(Indian Pines) with classification.

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

1. The result of the indian_pines_pca.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 .

2. The result of the indian_pines_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:

   

3. The result of the indian_pines_knnc_after_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

dimensionality-reduction-and-classification-on-hyperspectral-images-using-python's Issues

NaN values in dataset

error message: ValueError: Input contains NaN, infinity or a value too large for dtype('float64').

it happens while predicting the KNN model before the PCA, working on the original dataset.

program: indian_pines_knnx.py
error line: model.fit(X_train, y_train)

Dimensional reduction by non classical approaches

Why not use Autoencoders for dim reduction rather than standard statistical one layer technique.
Since NN are the state of the art algorithmic approach.