University of Illinois Urbana-Champaign

Neural network approach to automatic target recognition (ATR) on deep in-memory architecture (DIMA)

Geng, Hanfei

This item is only available for download by members of the University of Illinois community. Students, faculty, and staff at the U of I may log in with your NetID and password to view the item. If you are trying to access an Illinois-restricted dissertation or thesis, you can request a copy through your library's Inter-Library Loan office or purchase a copy directly from ProQuest.

Permalink

https://hdl.handle.net/2142/104060

Description

Title
Neural network approach to automatic target recognition (ATR) on deep in-memory architecture (DIMA)
Author(s)
Geng, Hanfei
Contributor(s)
Shanbhag, Naresh R.
Issue Date
2019-05
Keyword(s)
  • Automatic target recognition
  • Synthetic aperture radar
  • Convolutional neural network
  • Deep in-memory architecture
Abstract
Automatic target recognition (ATR) has a long history and a wide range of applications. It refers to the ability to recognize targets of interest based on readings collected from sensors. The entire process consists of three stages: target detection, target segmentation, and target recognition. This research project investigates the recognition stage of a particular ATR that takes synthetic aperture radar (SAR) images taken on airborne platforms as input. Traditionally this stage has always taken a statistical approach. Recently a neural-network-based approach has been proposed as well. Given its innate complexity and high energy demand due to data movement, it is essential to adopt this approach in an energy-efficient manner. DIMA is a new computation scheme dedicated to this effort. In contrast to traditional von Neumann architecture, it reduces the high energy costs of data movement between processor and memory through analog computations deeply embedded into the periphery of the bit-cell array (BCA) in the memory. The goal of this project is to demonstrate the benefits of DIMA in the context of neural-network-based ATR. Several floating-point and fixed-point convolutional neural network (CNN) models of SAR-based ATR have been developed. With the noise model of DIMA, Monte Carlo simulation was performed on a popular ATR dataset called Moving and Stationary Target Acquisition and Recognition (MSTAR) to study the effect on CNN model’s accuracy.
Type of Resource
text
Language
en
Permalink
http://hdl.handle.net/2142/104060

Owning Collections

Senior Theses - Electrical and Computer Engineering PRIMARY
The best of ECE undergraduate research