Software Defined Doppler Radar for LandmineDetection using GA-Optimized Machine Learning

Presenter's Name(s)

Yan Zhang, UVMFollow

Conference Year

January 2020

Abstract

Using software defined radar (SDR) technology and modern machine learning algorithms, this paper demonstrates a software defined Doppler Radar(SDDRadar) system that can distinguish buried non-metallic landmine from other buried objects (rock, wood, etc.) with a high accuracy. In the sensing, the spectrum responses of different buried objects are collected using the SDDRadar. The vibration spectrum data are fed into a Random Forest where a Genetic Algorithm(GA) is designed to optimize the performance of the Random Forest. To leverage SDDRadar sensitivity, a clutter cancellation circuit is designed and integrated into the system. Two outdoor tests are performed under dry and wet soil conditions. For performance evaluation, the GA-optimized Random Forest is compared with other two machine learning algorithms, including Support Vector Machine and Logistic Regression. As it turns out, the GA-optimized Random Forest has the best classification performance in terms of both precision and recall parameters.

Primary Faculty Mentor Name

Tian Xia

Status

Graduate

Student College

College of Engineering and Mathematical Sciences

Program/Major

Electrical Engineering

Primary Research Category

Engineering & Physical Sciences

Abstract only.

Share

COinS
 

Software Defined Doppler Radar for LandmineDetection using GA-Optimized Machine Learning

Using software defined radar (SDR) technology and modern machine learning algorithms, this paper demonstrates a software defined Doppler Radar(SDDRadar) system that can distinguish buried non-metallic landmine from other buried objects (rock, wood, etc.) with a high accuracy. In the sensing, the spectrum responses of different buried objects are collected using the SDDRadar. The vibration spectrum data are fed into a Random Forest where a Genetic Algorithm(GA) is designed to optimize the performance of the Random Forest. To leverage SDDRadar sensitivity, a clutter cancellation circuit is designed and integrated into the system. Two outdoor tests are performed under dry and wet soil conditions. For performance evaluation, the GA-optimized Random Forest is compared with other two machine learning algorithms, including Support Vector Machine and Logistic Regression. As it turns out, the GA-optimized Random Forest has the best classification performance in terms of both precision and recall parameters.