Target Detection in Sea Clutter with Transformer Neural Network

Abstract

Target detection in sea clutter is a significant problem in marine surveillance. The difficulty of radar target detection lies in the low signal-to-clutter ratio (SCR) and insufficient feature extraction brought by handcrafted features. In recent years, with the great success of artificial intelligence in the field of computer vision, there is a trend that researchers start to apply convolutional neural networks to radar target detection. However, most methods based on convolutional neural networks need to convert one-dimensional signals into two-dimensional images for preprocessing, which not only increases the computational burden, but also reduces the accuracy of detection. Therefore, in this paper, we propose a Transformer-based sea clutter target detection method, which takes the collected one-dimensional signal as input directly after FFT preprocessing. And comparing with two-dimensional images, which can extremely reduces the computational cost. Most importantly, Transformer-based network has self-attention mechanism to strengthen target and suppress noise, thus improving the accuracy of target detection. The proposed algorithm is verified on the IPIX datasets, and the results show that the proposed algorithm attains excellent results compared to state-of-the-art convolutional networks while requiring substantially fewer computational resources to train and inference.