Radar is widely applied in maritime target detection. However, the performance of sea-surface target detection in complex ocean environments is limited because of the low signal-to-clutter ratio. Hence, sea clutter suppression has become a highly investigated research topic. In this study, we explicitly characterize spatiotemporal information in range-Doppler spectra using non-Euclidean graphs, and subsequently present a sea clutter suppression method with graph neural network. We propose a representation of radar signal graph data that is constructed using patches segmented from sequential range-Doppler maps. In the proposed graph neural network, edge values are updated through a global attention mechanism that captures the correlations between resolution cells in multiple coherent processing intervals and extracts the feature differences between the target and clutter. A graph convolution operation is designed. We incorporate the learned adjacency matrix into the graph convolution kernel to discriminatively aggregate node features based on the associations between nodes. To increase the model capacity, we devise an auxiliary graph classification head and a compound training loss comprising contrastive and cross-entropy losses. The proposed method was verified on real sea clutter data and demonstrated a significant sea clutter suppression effect and improved target detection performance. Additionally, this approach was found to exhibit satisfactory generalization ability.
19 February 2024 in IEEE Transactions on Vehicular Technology
Broadband beamforming technology is an effective solution in millimeter-wave (mmWave) massive multiple-input-multiple-output (MIMO) systems to counteract severe path loss via beamforming gains. One classic approach to implementing adaptive broadband beamforming is through spatio-temporal structure. In this paper, a method based on the complex convolutional neural network (CCNN) is proposed, namely, broadband beamforming prediction complex-valued network (BBPCNet), which can solve the performance degradation problem of traditional broadband beamforming methods under circumstance of limited snapshots. The BBPCNet leverages the CCNN to maintain the inherent amplitude and phase relationships of the complex-valued signals received by the array. A significant advantage of the proposed method is the elimination of the need for covariance matrix inversion to produce beamforming weights, resulting in lower computational complexity. Numerical simulations demonstrate effectiveness and superiority of the proposed method.
The synthetic aperture radar (SAR) for ocean surveillance missions requires low latency and light weight inference. This paper proposes a novel small-size Searched Binary Network (SBNN), with network architecture search (NAS) for ship classification with SAR. In SBNN, convolution operations are modified by binarization technologies. Both input feature maps and weights are quantized into 1-bit in most of the convolution computation, which significantly decreases the overall computational complexity. In addition, we propose a patch shift processing, which can adjust feature maps with learnable parameters at spatial level. This process enhances the performance by reducing the information irrelevant to the targets. Experimental results on the OpenSARShip dataset show the proposed SBNN outperforms both binary neural networks from computer vision and CNN-based SAR ship classification methods. In particular, SBNN shows a great advantage in computational complexity.
23 May 2024 in IEEE Geoscience and Remote Sensing Letters
Adaptive broadband digital beamforming (ABDBF) is an essential topic in the realm of array antenna for radar systems, because the array antenna with ABDBF could obtain wide swath and high azimuth resolution. Performance degradation at low snapshots is a serious problem for ABDBF. In this letter, the broadband beamforming weight generation network (BWGN) is proposed to quickly generate weights for ABDBF under low signal snapshot scenarios. The BWGN leverages the complex convolutional neural network to represent the mapping between input signals and output weights, which avoids the operation of covariance matrix and its inversion, thus speeding up the generation of weights. Compared with the existing neural network-based broadband beamforming method, wideband beamforming prediction network (WBPNet), training BWGN saves 71.45% of the time. Furthermore, progressive learning is introduced to the training process of BWGN, namely PL-BWGN, which further reduces training time of BWGN to 0.7529 h (h: hours). Simulation results demonstrate performance superiority of the proposed method compared with existing beamformers under low snapshot scenarios.
Ship classification technology using synthetic aperture radar (SAR) has become a research hotspot. Many deep-learning-based methods have been proposed with handcrafted models or using transplanted computer vision networks. However, most of these methods are designed for graphics processing unit (GPU) platforms, leading to limited scope for application. This paper proposes a novel mini-size searched convolutional Metaformer (SCM) for classifying SAR ships. Firstly, a network architecture searching (NAS) algorithm with progressive data augmentation is proposed to find an efficient baseline convolutional network. Then, a transformer classifier is employed to improve the spatial awareness capability. Moreover, a ConvFormer cell is proposed by filling the searched normal convolutional cell into a Metaformer block. This novel cell architecture further improves the feature-extracting capability. Experimental results obtained show that the proposed SCM provides the best accuracy with only weights, achieving a good trade-off between performance and model size.