Classification of Space Objects by Using Deep Learning with Micro-Doppler Signature Images

Abstract

Radar target classification is an important task in the missile defense system. State-of-the-art studies using micro-doppler frequency have been conducted to classify the space object targets. However, existing studies rely highly on feature extraction methods. Therefore, the generalization performance of the classifier is limited and there is room for improvement. Recently, to improve the classification performance, the popular approaches are to build a convolutional neural network (CNN) architecture with the help of transfer learning and use the generative adversarial network (GAN) to increase the training datasets. However, these methods still have drawbacks. First, they use only one feature to train the network. Therefore, the existing methods cannot guarantee that the classifier learns more robust target characteristics. Second, it is difficult to obtain large amounts of data that accurately mimic real-world target features by performing data augmentation via GAN instead of simulation. To mitigate the above problem, we propose a transfer learning-based parallel network with the spectrogram and the cadence velocity diagram (CVD) as the inputs. In addition, we obtain an EM simulation-based dataset. The radar-received signal is simulated according to a variety of dynamics using the concept of shooting and bouncing rays with relative aspect angles rather than the scattering center reconstruction method. Our proposed model is evaluated on our generated dataset. The proposed method achieved about 0.01 to 0.39% higher accuracy than the pre-trained networks with a single input feature.

Space Target Classification Improvement by Generating Micro-Doppler Signatures Considering Incident Angle

“…In order to verify the performance of the proposed micro-Doppler dataset considering the relative incident angle, a dataset is generated for the same target models as the previous study of [13], and the classification performances are compared. Therefore, the same target models as shown in Figure 5 and the same dynamics and radar system parameters summarized in Table 1 are used, so the dynamic RCS is the same as in the previous work [13], but the spectrogram and CVD images are different. Table 1.…”

“…Table 1. Dynamic and radar system parameters for dataset generation [13]. Figure 6 shows the spectrogram of the rounded cone for some incident angles when the rotation rate is 3 Hz.…”

“…In addition, it is relatively possible to extract fixed and accurate scattering centers at high frequencies where the target is large compared to the wavelength, but it is difficult to extract dominant scattering centers at frequencies similar to the wavelength of the target. Due to these issues, received signals or radar cross-section (RCS) are often directly obtained through a full-wave electromagnetic simulation according to various postures and incident angles [13,14]. Recently, the concept of shooting and bouncing rays with relative aspect angles has been introduced [13].…”

“…Due to these issues, received signals or radar cross-section (RCS) are often directly obtained through a full-wave electromagnetic simulation according to various postures and incident angles [13,14]. Recently, the concept of shooting and bouncing rays with relative aspect angles has been introduced [13]. In consideration of the symmetrical shape of the space targets, the scattered field according to the angle of incidence on a fixed target is calculated and used to configure a lookup table, and consequently, the scattered field is estimated according to the relative angle of incidence between the target's orientation and the incident field.…”

“…In [18], the transfer learning of AlexNet and SqueezeNet were used to classify the micro-motions based on spectrogram images. In [13], a parallel CNN network that inputs both spectrogram and CVD at the same time has been proposed. In [14], a deep learning network was designed that uses RCS signal patterns rather than micro-Doppler signature images as input data.…”