| Title |
Performance Analysis of ResNet-based Target Classifications using Spectral Kurtosis |
| Authors |
김지현(Ji-Hyeon Kim) ; 박도현(Do-Hyun Park) ; 김형남(Hyoung-Nam Kim) |
| DOI |
https://doi.org/10.5573/ieie.2022.59.4.99 |
| Keywords |
Micro-Doppler; Characteristic vector; Spectral kurtosis; Target classification; ResNet34 |
| Abstract |
Micro-Doppler modulation is a target signature that represents micro-motion state for each individual movement and it is used in the technology of recognizing and classifying targets. The micro-Doppler frequency appears in the form of transition of the Doppler frequency by basic movement characteristics such as rotation and vibration of an object, and thus it can make it possible to track a target and classify it with high recognition accuracy. In this paper, we model micro-motion signals of a drone, a bird, and human targets, and extract the micro-Doppler feature vector of the target by calculating the spectral kurtosis of micro-Doppler images. To classify targets performing micro-movement, we apply ResNet34 deep neural network to spectral kurtosis input. Through simulation, we analyze the classification performance of ResNet34 algorithm according to the radar measurement data input set of each target. Simulation results show that the proposed method has more than 95% performance on three scales of accuracy, precision, and recall, and it is superior to the conventional method using micro-Doppler images. |