HSI_svm_pca_resNet50

use SVM and PCA_ResNet50 to classify HSI
===== update：
Besides using NMF method , then i use GNMF method , and i compare the result between NMF and GNMF.
result:

• stage1:
  use SVM to classify HSI(Hyperspectral Image). opertion : Firstly , transfer 3D Indian pines data to 2D data , and the origial groundtruth data is 2D data , I transfer it to 1D data. Then, according the quantity of GT(groundTruth) , I choose 15 vectors from the first three least and choose 50 vectors from the other vectors. Treat these data as training data.the left data is test data. And I use the NMF as the method of demension reduction . Before NMF , I normalize the data using Z-Score. Then i use NNLS on the W matrix AFTER NMF and get H matrix. Finally i use SVM to fit and predict the label. This is the NNLS.

  here is the process and result:
  
  

• stage2:
  use PCA and ResNet50 to classify HSI. Firstly,choose 20% of the Indian pines data as the training data , the other is the test data. Use PCA as the method of demension reduction , after I put the data into ResNet50 , finally get the resut.
  Here is the structure of ResNet50 . The result: loss:. accuary:

  Reference:
  PCA
  ResNet50_on_Keras
  HybridSN

• how to use it:
  run :

  pip install -r requirements.txt

  first to install the packages.
  if you want to use SVM , you can run

  python train_demo.py .

  If you want to use PCA+ResNet50 , you can run

  python PCA_ResNet50.py

• contact:
  if you have problems , you can pick up an issue and communicate with me , here is my wechat number:Yingbin192