Reduction of False Alarms in Sea Ice Covered Ocean Regions Using Machine Learning

PDF

Authors
  1. Sandirasegaram, N.
  2. Vachon, P. W.
Corporate Authors
Defence Research and Development Canada, Ottawa Research Centre, Ottawa ON (CAN)
Abstract
Current ship detection algorithms for Synthetic Aperture Radar (SAR) imagery experience limitations due to false alarms that arise in sea ice conditions. The aim of this Scientific Report is to demonstrate how false alarms can be reduced by applying target discrimination algorithms. In this report, RADARSAT-2 Maritime Satellite Surveillance Radar (MSSR) images acquired in both Detection of Vessels Wide Far (DVWF) and Ocean Surveillance Very-wide Near (OSVN) modes are considered, because these are the operational modes used for ship detection. Training and testing data samples were collected in regions that include sea ice. Ship targets were identified using visual analysis and Automatic Identification System (AIS) data reported by ships. Sea ice targets were identified by removing detected ship targets, and ship-like targets were excluded via visual inspection. Testing data were kept independent from training data. Support Vector Machine (SVM), Autoencoder Neural Network (AENN), and Convolutional Neural Network (CNN) were applied to discriminate the false targets (i.e., sea ice) from ship targets. Preprocessing and feature extraction steps were applied to the SVM method but not to the AENN nor the CNN methods. AENN and CNN are deep learning neural networks. The results show that these methods can remove more than 93% of the false targets detected in DVWF and OSVN modes. However, a few of the ship targets were misclassified as sea ice targets in both the DVWF and OSVN modes. T

Il y a un résumé en français ici.

Keywords
Deep learning;Machine Learning;Support Vector Machine;Ship and Ice discrimination;Synthetic Aperture Radar (SAR);Automatic Identification System (AIS)
Report Number
DRDC-RDDC-2018-R249 — Scientific Report
Date of publication
01 Dec 2018
Number of Pages
17
DSTKIM No
CA048214
CANDIS No
808847
Format(s):
Electronic Document(PDF)

Permanent link

Document 1 of 1

Date modified: