Machine Learning for Detecting Anomalies in SAR Data

Yuval Haitman, Itay Berkovich, Shiran Havivi, Shimrit Maman, Dan G. Blumberg, Stanley R. Rotman

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

One of most common algorithms for anomaly detection in multi-dimensional imagery is the Reed - Xiaoli (RX) algorithm; it gives each pixel a score that defines its likelihood to be an anomaly. We have implemented a new algorithm which uses both RX and the Non-Negative Matrix Factorization (NNMF) learning algorithm in order to pick an adaptive threshold for detection; we have applied it to Synthetic Aperture Radar (SAR) data. The NNMF approach is defined as a minimization problem which approximates the given data by extracting its main trends. By comparing the original data to the reduced data, we can divide the image anomalies into two different groups, where one group contains the anomalies which are part of the image main trends and the second group contains the anomalies of the sub trends. With this division, we can pick an adaptive threshold for each of the groups according to its unique characteristics.

Original languageEnglish
Title of host publication2019 IEEE International Conference on Microwaves, Antennas, Communications and Electronic Systems, COMCAS 2019
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781538695494
DOIs
StatePublished - 1 Nov 2019
Event2019 IEEE International Conference on Microwaves, Antennas, Communications and Electronic Systems, COMCAS 2019 - Tel-Aviv, Israel
Duration: 4 Nov 20196 Nov 2019

Publication series

Name2019 IEEE International Conference on Microwaves, Antennas, Communications and Electronic Systems, COMCAS 2019

Conference

Conference2019 IEEE International Conference on Microwaves, Antennas, Communications and Electronic Systems, COMCAS 2019
Country/TerritoryIsrael
CityTel-Aviv
Period4/11/196/11/19

Keywords

  • Anomaly detection
  • Non-Negative Matrix Factorization (NNMF)
  • Synthetic Aperture Radar (SAR)

Fingerprint

Dive into the research topics of 'Machine Learning for Detecting Anomalies in SAR Data'. Together they form a unique fingerprint.

Cite this