JieGuo / SAR.AI

:satellite: Multi-Level Pattern Histogram for Synthetic-Aperture Radar (SAR) image classification into terrain classes.

SAR Image Classification with MLPH

Modified Multi-Level Pattern Histograms (MLPH) for SAR image classification.

Some results for different regions in and around New York City area:

Dependencies

Platform tested on: Ubuntu 16.04

Softwares/Packages required:

• Python >2.6
• pickle
• numpy
• sklearn
• skimage
• scipy

Instructions to Run the Application

• Open the terminal inside the directory “IP_Project_Group23”

• To run MLPH on training images and train SVM & test on hold-out set:

$ python -m sarclf -run_mlph 1 -train_svm 1

• To run pretrained SVM on test image pixels:

$ python -m sarclf

• To run pretrained SVM to classify an image:

$ python -m sarclf -test 0 -clfimg <image_path>

• For example, for image 2.tif in test_images folder, use:

$ python -m sarclf -test 0 -clfimg ./test_images/2.tif

• To do any of the above steps with our modified MLPH: Add “-modified 1”

Example:

$ python -m sarclf -test 0 -clfimg ./test_images/2.tif -modified 1

Methodology

Following the research on MLPH, we derive a pattern matrix for each pixel based on a threshold value, use these matrices varying bin lengths to get local pattern sub-histograms, concatenated to give local pattern histogram. With multiple thresholds, the concatenation of local pattern histograms gives MLPH for each pixel.