Mine and Mine Like Objects Classifications through Deep Learning Neural Network Systems [video]

Authors
Joung, Sang Ki
Song, Kwang Sub
Kim, Moon Hwan
Chu, Peter
Subjects
Advisors
Date of Issue
2018-04-18
Date
4/18/2018
Publisher
Language
Abstract
Threats of sea mines are increasing due to recent technology development, such as autonomous systems and computer systems with artificial intelligent capability. There are many solutions to solve MCM problems as far as difficulties to detect identify and classification. Unmanned systems integrated with emerging technologies are the minesweepers and hunters of the future MCM operations. A focused technology effort needed to incorporate unmanned systems into the mine countermeasure ship and other related MCM fleet forces. In MCM operation, it has several due sequences; identification location, neutralization and analysis. In this study, we try to use deep learning AI technique to identify, classify, and locate mines, so that lead to programming decisions allowing mine hunting and minesweeping missions to perform without a man onboard, eliminating the risk to personnel. Most of data and information for mine detection are acquired as form of image pixel from size scan sonar or bathometry sonar. However, detection of mine from sonar image is not easy to perform because sonar image has low resolution, shadows of different shape or size and complexity of ocean environment. There are plenty of Identification and Classification Algorithms using up to date Deep Learning Method that requires huge training data and Processing Hard Ware with Graphic Processor Unit. AI system needs long training time for Neural Networks tuning, but mine and mine like object and their respective SONAR signal data are few and restrictive to access We select faster regional CNN deep learning neural network as deep learning network for mine classifications, but it needs long processing time and data for signal from minefield. Deep learning functional process are largely divided by region selection and classification sequences, and applying AI process to region selection require more time than well-defined classification process. In this works, we separate Region of Interest selection processing from whole Deep Learning Package for Mine Classification. Region of Interest are selected by combined identification process with simple model referenced CNN circuit, Tactical Mine Database, Environmental Condition probability and SONAR signal processing, which give much fast processing time. Selected regions of interest (ROI) provide through well-established Faster R-CNN Package for Classifications Efficiency test and accuracy test are to compare other deep learning classification packages and give a better accuracy and computation time. We normally check classification accuracy from trained data and non-trained (blind) data from mine warfare databases.
Type
Video
Description
CRUSER TechCon 2018 Research at NPS. Wednesday 1: Sensing
Series/Report No
Department
Organization
Identifiers
NPS Report Number
Sponsors
Funder
Format
Citation
Distribution Statement
Rights
This publication is a work of the U.S. Government as defined in Title 17, United States Code, Section 101. Copyright protection is not available for this work in the United States.
Collections