• Author : Anmol Popli [email protected] , Nitesh Arora [email protected]

This source code provides an implementation of an SVM based classifier to classify satellite imagery.

The code is compilable on Unix/Linux.

• Compilation.
  Automated compilation requires the make program.

• Library.
  This code requires the libpng, libjpeg and libtiff library.

• Image format.
  Formats supported are jpeg , png and tiff.

Usage:

1. Download the code package and extract it. Go to the directory source_code.

2. Five programs can be executed in this package. The first 3 are part of training the classifier. The fourth tests the classifier on the traing data. The fifth applies the classifier to any test images.
   imageio.c is the image I/O library and svm.cpp is the SVM library.

3. Compile the source code of IMG2TRAIN (img_traindata.c) by running
   make IMG2TRAIN
   Run the executable IMG2TRAIN.
   ./IMG2TRAIN
   This converts the training windows from 4 bands to training data and stores it in the file traindata.

4. Compile the source code of CROSSVAL (train_cv.c) by running
   make CROSSVAL
   The values of C,gamma and n are to be input in the function train_params inside the MAIN of train_cv.c.
   Run the executable CROSSVAL.
   ./CROSSVAL
   This performs n-fold cross-validation on traing data with inputs C and gamma and gives the Cross Validation Accuracy in result.

5. Compile the source code of TRAIN (sat_train.c) by running
   make TRAIN
   The values of C and gamma are to be input in the function train_params inside the MAIN of sat_train.c.
   Run the executable TRAIN.
   ./TRAIN
   This trains the classifier with training data using the values of inputs C and gamma. The classifier model is stored in the file traindata.model

6. Compile the source code of CLASSIFY_TRAIN (sat_classify.c) by running
   make CLASSIFY_TRAIN
   Run the executable CLASSIFY_TRAIN.
   ./CLASSIFY_TRAIN
   This classifies the training data in traindata using the model in traindata.model and stores the classified result in the file traindata.classified. Computes accuracy parametrs and stores Confusion Matrix in the file confusion, Accuracy and Kappa Coefficient are is output onto the terminal.

7. Compile the source code of CLASSIFY_TEST (test_classify.c) by running
   make CLASSIFY_TEST
   Run the executable CLASSIFY_TEST.
   ./CLASSIFY_TEST RED_BAND.tif GREEN_BAND.tif BLUE_BAND.tif INFRARED_BAND.tif
   
   with :

• RED_BAND.tif is the red band image;
• GREEN_BAND.tif is the green band image;
• BLUE_BAND.tif is the blue band image;
• INFRARED_BAND.tif is the infrared band image.
  
  This classifies the test images of 4 bands into the 7 classes. Stores converted test data in file testdata. Stores classified test data in file testdata.classified. Generates the classified output image classified.jpg where colors represent :
• blue - class 1
• green - class 2
• red - class 3
• cyan - class 4
• yellow - class 5
• magenta - class 6
• black - class 7

8. To delete redundant files,
   run make clean

----Usage of demo file:
Since we have already generated the file traindata.model and stored in the package directory, CLASSIFY_TEST can be directly used to classify any input test images using this demo file.

In classify_demo.sh , change the parameters as:
red as "full name of red band image";
green as "full name of green band image";
blue as "full name of blue band image";
infrared as "full name of infrared band image"

Run sh classify_demo.sh