Micro-Doppler Based Classification of Human Aquatic Activities via Transfer Learning of Convolutional Neural Networks.

Jinhee Park,Taesup Moon,Youngwook Kim,Rios Javier

doi:10.3390/s16121990

Abstract

Accurate classification of human aquatic activities using radar has a variety of potential applications such as rescue operations and border patrols. Nevertheless, the classification of activities on water using radar has not been extensively studied, unlike the case on dry ground, due to its unique challenge. Namely, not only is the radar cross section of a human on water small, but the micro-Doppler signatures are much noisier due to water drops and waves. In this paper, we first investigate whether discriminative signatures could be obtained for activities on water through a simulation study. Then, we show how we can effectively achieve high classification accuracy by applying deep convolutional neural networks (DCNN) directly to the spectrogram of real measurement data. From the five-fold cross-validation on our dataset, which consists of five aquatic activities, we report that the conventional feature-based scheme only achieves an accuracy of 45.1%. In contrast, the DCNN trained using only the collected data attains 66.7%, and the transfer learned DCNN, which takes a DCNN pre-trained on a RGB image dataset and fine-tunes the parameters using the collected data, achieves a much higher 80.3%, which is a significant performance boost.

Highlights

Increased demand for security, law enforcement, rescue operations, and health care has accelerated research in the detection, monitoring, and classification of human activities [1,2] based on remote sensing technologies
We followed the approach of [17] and carried out five-fold cross validation (CV) using the spectrogram data collected as in Section 3 to evaluate the performances of the compared methods
For the deep convolutional neural networks (DCNN) models, since the model configurations were fixed as explained in Section 4, the only hyper-parameter we chose via CV was the early stopping parameter; namely, we picked the Stochastic Gradient Descent (SGD) iteration that gave the best average test score

Summary

Introduction

Increased demand for security, law enforcement, rescue operations, and health care has accelerated research in the detection, monitoring, and classification of human activities [1,2] based on remote sensing technologies. The unique micro-Doppler signatures from human activities enabled diverse and extensive research on human detection and activity classification/analysis using radar sensors [3,4,5,6,7,8,9,10,11,12]. The authors of [6] extracted direct micro-Doppler features such as bandwidth and Doppler period, the authors of [7] applied linear predictive code coefficients, and the authors of [8] applied minimum divergence approaches for robust classification under a low signal-to-noise ratio environment. Most of the research has focused on the classifications of human activities on dry ground

Methods

Results

Conclusion