The ambient noise covariance matrix for the array of underwater acoustic vector sensors (AVSs) is not equal to a scaled identity matrix. This fact contradicts the requirement of subspace-based DOA estimation methods such as the conventional MUSIC method, leading to the performance degradation of their DOA estimation. In order to overcome this problem, we explore the real- and imaginary- property of the auto-correlation and cross-correlation of the ambient noise, and propose a MUSIC-based DOA estimation method with asymptotically Ambient Noise Elimination (Named as ANE MUSIC method). In particular, the ANE MUSIC method first transforms the array covariance matrix to a new one of which the noise is concentrated in the real part. Thus, the imaginary part of the transformed covariance matrix eliminates ambient noise. Afterwards, based on the imaginary part of the transformed covariance matrix, the ANE MUSIC method employs a real-valued Singular Value Decomposition (SVD) to complete DOA estimation. The proposed ANE MUSIC method is asymptotically independent of the ambient noise. Therefore, it is robust to the ambient noise in the case of limited snapshots. In addition, since its involved spectral searching is over only half of the total angular field-of-view with a real-valued noise subspace, the ANE MUSIC method reduces the computational complexity by about 75% in terms of spectral searching, as compared to the conventional MUSIC method which utilizes the complex-valued Eigenvalue Decomposition(EVD) and a spectral searching over the total angular field-of-view. It is noted that the proposed ANE MUSIC method does not require to know the prior noise covariance matrix, which is different from the existing prewhitening solution. Simulation results demonstrate the ANE MUSIC method performs significantly better than the other methods, especially in the case of low signal-to-noise ratios (SNRs). Moreover, it gains a certain robustness against the sensor gain-phase errors. Experimental results verify the practical effectiveness of the ANE MUSIC method, based on the real data collected by an array of two AVSs in the anechoic water tank and the real data collected by an uniformly circular array of eight AVSs in the Songhua Lake in Jilin, China.

We propose a piecewise nonlinear frequency modulation (PNLFM) waveform for a multiple-input multiple-output (MIMO) radar. Each subpulse in the waveform set is divided into three segments. The first and the third segments are linear frequency modulation (LFM) signals, whereas the second segment is a nonlinear frequency modulation (NLFM) signal. The bandwidths and durations of subpulses are the same, whereas they can be different for each segment in the subpulse. In addition, the nonlinear time-frequency function of the second segment is controlled to improve the range and Doppler sidelobe suppression. A genetic algorithm is implemented to optimize the performance by maximally suppressing the autocorrelation sidelobe and cross correlation in range, Doppler, and angle. The MIMO radar performance is demonstrated via computation of the ambiguity function, autocorrelation function, and cross-correlation function. Numerical results are presented to show waveform properties and to compare with some existing MIMO radar waveforms. The proposed PNLFM waveform has low autocorrelation sidelobes and low cross correlation in range and Doppler, which sits at a level determined largely by the chosen time-bandwidth product.

In this paper, a lightweight deep neural network (DNN) for direction of arrival (DOA) estimation is proposed, of which the input vector is designed to remove data redundancy as well as remaining DOA information. By exploring the Vandermonde property of the steering vector of a uniform linear array (ULA), the size of the newly designed input vector is greatly reduced. Furthermore, the DOA estimation is designed as a regression problem instead of a classification problem; that is, the lightweight DNN designs the output vector as the estimated DOAs of sources, of which the size is much shorter than that of the spatial spectrum used as the output vector in the conventional DNN. The reductions in the sizes of input and output vectors lead to a reduction in the sizes of hidden layers, achieving lightweightness of the neural network. The analysis illustrates that when the number of sensors is 22, the number of parameters in the lightweight DNN is three orders of magnitude less than that in the conventional DNN. The simulation results demonstrate the lightweight DNN can provide high DOA estimation accuracy with the shortest testing time. It performs better than the conventional DNN. Furthermore, it is superior to traditional solutions such as the multiple signal classification (MUSIC) method and conventional beamforming (CBF) method in harsh conditions like low signal-to-noise ratios (SNRs), closely spaced sources, and few snapshots.

In this letter, we propose an efficient hybrid model, named HMC-ViT, that combines a convolutional neural network (CNN) with a multi-class token vision transformer (ViT) to address the problem of direction of arrival (DOA) estimation. HMC-ViT integrates the local feature extraction capability of CNN with the global feature extraction capability of ViT to enhance DOA estimation performance and improve the computational efficiency of ViT. Additionally, the ViT component employs multiple class tokens in parallel to generate spatial spectra for sub-regions, further enhancing the model's performance. Simulation results demonstrate that the proposed method outperforms existing approaches under low signal-to-noise ratio (SNR) scenarios. The code of this letter is available at https://github.com/morigen0721/HMC-ViT.

Graffiti detection is essential in historic building protection and urban neighborhood management. Graffiti detection has made significant progress in recent years based on the development of deep learning. However, small-scale graffiti, interference from the background, and the false detection of word parts in graffiti make it a challenging problem. This article proposes a Transformer-based high-precision graffiti detection method, namely DIG. Precisely, it consists of three modules: 1) Spatial query selection (SQS), scale-aware IoU loss (SIL); 2) Denoising Task with binary contrastive denoising (BCDN); and 3) IoU-guided box denoising (IBD) modules. To detect small-scale graffiti, this paper proposes SIL to help the loss function to perceive small-scale graffiti and large-scale graffiti fairly. To reduce the false detection of word parts, this article presents the SQS module, which integrates spatial information into the query selection process of the Encoder to filter out falsely detected bounding boxes within the graffiti. To reduce the interference from the background, this article introduces a denoising task with BCDN and IBD modules, improving the model’s ability to distinguish graffiti from the background and accurately select appropriate bounding boxes. A large number of experimental results on the STORM dataset show that our method achieves state-of-the-art results with an $AP_{50}$ of 87.9%. Moreover, DIG achieved competitive results on the FineFM dataset for mask detection. This indicates that DIG can also be conveniently transferred to detect other scenarios.