Moving Target Indication Using Deep Convolutional Neural Network

Abstract

Existing moving target indication (MTI) methods are limited either by the sufficient and homogeneous secondary data (SD) condition or by the high signal-to-clutter-ratio (SCR) requirement. In practice, such constraints are hard to be satisfied. In this paper, we propose a novel deep convolutional neural network (CNN)-based method for the MTI (CNN-MTI) to overcome the limitations of these methods. In the proposed method, the CNN architecture is deliberately designed, and the training data set is greatly augmented to ensure the CNN be deep and well-trained; in addition, the SD only act as the background interference during the augmentation of the training data set. These procedures ensure that the effective features of different classes of moving targets can be extracted, so that good MTI performance can be achieved without the requirement of the sufficient and homogeneous SD or the high SCR condition. Simulation results from the synthetic data and the experimental data demonstrate the validity and the robustness of the CNN-MTI with limited SD support in non-homogeneous and low SCR environment.

Keywords

• Moving target indication (MTI)
• convolution neural network (CNN)
• space-time adaptive processing (STAP)
• feature extraction