Peng Wang

Accurate high-resolution temperature predictions at the subseasonal-to-seasonal (S2S) timescale are vital for applications in agriculture, energy management, and disaster mitigation. However, current S2S forecasting capacity is hindered by declining predictive skill with increasing lead time and limited capability in capturing extreme events and fine-scale spatial variability, particularly over regions with complex terrain. To bridge this gap, this study develops a deep learning (DL)-based framework, termed Terrain-ResNet-U-Net (TRU-Net), for the downscaling (from 0.5 ∘ to 0.1 ∘ resolution) and bias correction of 2 m air temperature forecasts over North China. Unlike existing approaches that treat topographic information as auxiliary input, TRU-Net integrates terrain information through a dedicated feature extraction module and incorporates it into a U-Net architecture with a ResNet34 backbone and a spatial attention mechanism, enabling enhanced nonlinear feature learning and improved representation of terrain-induced spatial heterogeneity. Comparative experiments show that TRU-Net consistently outperforms the original forecasts for daily 2 m temperature prediction at the S2S timescale. Averaged across the six lead times, TRU-Net reduces the root mean squared error by 22.97% and increases the Pearson correlation coefficient and Accuracy by 25.16% and 24.48%, respectively, relative to the original forecasts. It also outperforms representative baseline methods, including Linear Regression, U-Net, and several DL baselines. The largest overall improvements are observed at lead times of 3–5 weeks, while the ablation results confirm the contributions of terrain information and the proposed multi-scale terrain feature extraction and fusion strategy. Similar advantages are maintained for representative high-impact events, where the proposed framework also provides positive correction and downscaling effects under these challenging conditions. These results indicate that an effective terrain-aware framework can substantially improve S2S temperature bias correction and downscaling over North China.

A new meta-heuristic algorithm, termed the House Swallow Optimizer (HSO), is proposed. Inspired by the sophisticated behavioral repertoire of house swallows, this algorithm addresses the limitations of conventional optimization methods in solving complex, high-dimensional problems. The HSO systematically simulates 6 key survival strategies observed in house swallows—roosting, hovering, preying, nesting, breeding and seasonal migration—through a 5-phase mathematical model to achieve effective exploration-exploitation balance. To evaluate the algorithm’s performance, we conducted comparative studies using 23 standardized benchmark functions against 5 established meta-heuristics: BKA, SCA, SOA, DE, and PSO. The experimental outcomes prove that HSO has a strong ability to balance exploration and development, with out-performance over comparative algorithms. In order to verify the engineering application potential of the optimization algorithm proposed in this paper, aiming at the problems of large estimation error, high computational complexity and low accuracy in the DOA estimation of broadband signal of vector hydrophone array, the HSO algorithm is applied to the ML algorithm, and the DOA estimation of broadband signal is successfully carried out. Initially, the SST algorithm is employed for signal preprocessing, where broadband signals are transformed into narrowband equivalents through eigenvalue decomposition. This step facilitates the formulation of a ML-based DOA estimation model, which is subsequently optimized using the HSO algorithm. Simulation results demonstrate that the proposed HSO-ML method is superior to the ML implementation of other algorithms. Specifically, our approach achieves faster convergence speed, lower root mean square error, lower computational complexity and more robust performance.

Through extensive literature review, it has been found that sparse Bayesian learning (SBL) is mainly applied to traditional scalar hydrophones and is rarely applied to vector hydrophones. This article proposes a direction of arrival (DOA) estimation method for vector hydrophones based on SBL (Vector-SBL). Firstly, vector hydrophones capture both sound pressure and particle velocity, enabling the acquisition of multidimensional sound field information. Secondly, SBL accurately reconstructs the received vector signal, addressing challenges like low signal-to-noise ratio (SNR), limited snapshots, and coherent sources. Finally, precise DOA estimation is achieved for multiple sources without prior knowledge of their number. Simulation experiments have shown that compared with the OMP, MUSIC, and CBF algorithms, the proposed method exhibits higher DOA estimation accuracy under conditions of low SNR, small snapshots, multiple sources, and coherent sources. Furthermore, it demonstrates superior resolution when dealing with closely spaced signal sources.

The vector hydrophone is playing a more and more prominent role in underwater acoustic engineering, and it is a research hotspot in many countries; however, it also has some shortcomings. For the mixed problem involving received signals in micro-electromechanical system (MEMS) vector hydrophones in the presence of a large amount of external environment noise, noise and drift inevitably occur. The distortion phenomenon makes further signal detection and recognition difficult. In this study, a new method for denoising MEMS vector hydrophones by combining ensemble empirical mode decomposition (EEMD) and singular spectrum analysis (SSA) is proposed to improve the utilization of received signals. First, the main frequency of the noise signal is transformed using a Fourier transform. Then, the noise signal is decomposed by EEMD to obtain the intrinsic mode function (IMF) component. The frequency of each IMF component in the center further determines that the IMF component belongs to the noise IMF component, invalid IMF component, or pure IMF component. Then, there are pure IMF reserved components, removing noisy IMF components and invalid IMF components. Finally, the desalinated IMF reconstructs the signal through SSA to obtain the denoised signal, which realizes the denoising processing of the signal, extracting the useful signal and removing the drift. The role of SSA is to effectively separate the trend noise and the periodic vibration noise. Compared to EEMD and SSA separately, the proposed EEMD-SSA algorithm has a better denoising effect and can achieve the removal of drift. Following that, EEMD-SSA is used to process the data measured by Fenhe. The experiment is carried out by the North University of China. The simulation and lake test results show that the proposed EEMD-SSA has certain practical research value.

Direction of arrival (DOA) estimation based on Maximum Likelihood is a common method in array signal processing, with many practical applications, but the huge amount of calculation limits the practical application. To deal with such an Maximum Likelihood (ML) DOA estimation problem, firstly, the DOA estimation model with ML for acoustic vector sensor array is developed, where the optimization standard in various cases can be unified by converting the maximum of objective function to the minimum. Secondly, based on the Invasive Weed Optimization (IWO) method which is a novel biological evolutionary algorithm, a new Improved IWO (IIWO) algorithm for DOA estimation of the acoustic vector sensor array is proposed by using ML estimation. This algorithm simulates weed invasion process for DOA estimation by adjusting the non-linear harmonic exponent of IWO algorithm adaptively. The DOA estimation accuracy has been improved, and the computation of multidimensional nonlinear optimization for the ML method has been greatly reduced in the IIWO algorithm. Finally, compared with Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Differential Evolution (DE) method and Tuna Swarm Optimization(TSO) algorithm, numerical simulations show that the proposed algorithm has faster convergence rate, improved accuracy in terms of Root Mean Square Error (RMSE), lower computational complexity and more robust estimation performance for ML DOA estimation. The experiment with tracking the orientation of the motorboat by Microelectronic mechanical systems (MEMS) vector hydrophone array shows the superior performance of proposed IIWO algorithm in engineering application. Therefore, the proposed ML-DOA estimation with IIWO algorithm can take into account both resolution and computation. which can meet the requirements of real-time calculation and estimation accuracy in the actual environment.

The atom search optimization (ASO) algorithm has the characteristics of fewer parameters and better performance than the traditional intelligent optimization algorithms, but it is found that ASO may easily fall into local optimum and its accuracy is not higher. Therefore, based on the idea of speed update in particle swarm optimization (PSO), an improved atomic search optimization (IASO) algorithm is proposed in this paper. Compared with traditional ASO, IASO has a faster convergence speed and higher precision for 23 benchmark functions. IASO algorithm has been successfully applied to maximum likelihood (ML) estimator for the direction of arrival (DOA), under the conditions of the different number of signal sources, different signal-to-noise ratio (SNR) and different population size, the simulation results show that ML estimator with IASO algorithum has faster convergence speed, fewer iterations and lower root mean square error (RMSE) than ML estimator with ASO, sine cosine algorithm (SCA), genetic algorithm (GA) and particle swarm optimization (PSO). Therefore, the proposed algorithm holds great potential for not only guaranteeing the estimation accuracy but also greatly reducing the computational complexity of multidimensional nonlinear optimization of ML estimator.

Direction of arrival (DOA) estimation has always been a hot topic for researchers. The complex and changeable environment makes it very challenging to estimate the DOA in a small snapshot and strong noise environment. The direction-of-arrival estimation method based on compressed sensing (CS) is a new method proposed in recent years. It has received widespread attention because it can realize the direction-of-arrival estimation under small snapshots. However, this method will cause serious distortion in a strong noise environment. To solve this problem, this paper proposes a DOA estimation algorithm based on the principle of CS and density-based spatial clustering (DBSCAN). First of all, in order to make the estimation accuracy higher, this paper selects a signal reconstruction strategy based on the basis pursuit de-noising (BPDN). In response to the challenge of the selection of regularization parameters in this strategy, the power spectrum entropy is proposed to characterize the noise intensity of the signal, so as to provide reasonable suggestions for the selection of regularization parameters; Then, this paper finds out that the DOA estimation based on the principle of CS will get a denser estimation near the real angle under the condition of small snapshots through analysis, so it is proposed to use a DBSCAN method to process the above data to obtain the final DOA estimate; Finally, calculate the cluster center value of each cluster, the number of clusters is the number of signal sources, and the cluster center value is the final DOA estimate. The proposed method is applied to the simulation experiment and the micro electro mechanical system (MEMS) vector hydrophone lake test experiment, and they are proved that the proposed method can obtain good results of DOA estimation under the conditions of small snapshots and low signal-to-noise ratio (SNR).

Although convolutional neural networks (CNN) have shown excellent performance in many image recognition tasks, it commonly requires a lot of labeled data, and the recognition effect is frequently unsatisfied due to the limited labeled training data. In recent years, capsule network (CapsNet) has been shown to achieve a high recognition accuracy with a small group of training samples. In this study, a class separable loss based on cosine similarity is suggested to enhance the distinguishability of the extracted network. It is added as a regularization term to the original loss function to train the network, narrowing the intra-class difference and increasing the inter-class difference in each iteration. Meanwhile, a multi-dimensional parallel capsule module is established to obtain robust features and spatial relationships from the original images. Feature maps from convolution of different levels are extracted as the input of this module. Structural features derived from low-level convolution and semantic features derived from high-level convolution are used for low-dimensional capsule coding and high-dimensional capsule coding, respectively. In our experiment, the general moving and stationary target acquisition and recognition (MSTAR) database is used. We find that the accuracy of the multi-dimensional parallel capsule network with class separable loss (MdpCaps-Csl) is 99.79% using all training samples, which is higher than most current recognition methods. More importantly, the accuracy is up to 97.73% even if only 10% training samples are applied, indicating MdpCaps-Csl can make excellent performance upon limited training samples.