Underwater target classification methods based on deep learning suffer from obvious model overfitting and low recognition accuracy in the case of small samples and complex underwater environments. This paper proposes a novel classification network (EfficientNet-S) based on EfficientNet-V2S. After optimization with model scaling, EfficientNet-S significantly improves the recognition accuracy of the test set. As deep learning models typically require very large datasets to train millions of model parameter, the number of underwater target echo samples is far more insufficient. We propose a deep convolutional generative adversarial network (SGAN) based on the idea of group padding and even-size convolution kernel for high-quality data augmentation. The results of anechoic pool experiments show that our algorithm effectively suppresses the overfitting phenomenon, achieves the best recognition accuracy of 92.5%, and accurately classifies underwater targets based on active echo datasets with small samples.

Deep learning is a core technology for sonar image classification. However, owing to the cost of sampling, a lack of data for sonar image classification impedes the training and deployment of classifiers. Classic deep learning models such as AlexNet, VGG, GoogleNet, and ResNet suffer from low recognition rates and overfitting. This paper proposes a novel network (ResNet-ACW) based on a residual network and a combined few-shot strategy, which is derived from generative adversarial networks (GAN) and transfer learning (TL). We establish a sonar image dataset of six-category targets, which are formed by sidescan sonar, forward-looking sonar, and three-dimensional imaging sonar. The training process of ResNet-ACW on the sonar image dataset is more stable and the classification accuracy is also improved through an asymmetric convolution and a designed network structure. We design a novel GAN (LN-PGAN) that can generate images more efficiently to enhance our dataset and fine-tune ResNet-ACW pretrained on mini-ImageNet. Our method achieves 95.93% accuracy and a 14.19% increase in the six-category target sonar image classification tasks.

Accurate underwater target recognition is challenged by high noise levels, signal attenuation, multipath reflections, and the need for efficient real-time processing in complex environments. To overcome these obstacles, we propose the lightweight time-frequency-space dual-stream network (LTFSD-Net), which is a novel framework designed to enhance active sonar target recognition by effectively extracting and fusing spatial-, temporal-, and frequency-domain features. Our dual-stream architecture incorporates a beamforming-based spatial attention mechanism and a depth attention mechanism, enabling the network to focus on critical regions of target echoes and improve feature extraction. In addition, lightweight depthwise separable convolutional modules facilitate efficient time-frequency feature extraction, ensuring computational efficiency that is suitable for real-time applications. We introduce a novel feature fusion strategy that employs tensor decomposition at the front-end and compact bilinear pooling at the back-end. This approach seamlessly combines low- and high-level features, thereby enhancing the discriminative capabilities of the model. Extensive experiments conducted on both sea trial and anechoic pool datasets demonstrated that LTFSD-Net achieved overall accuracies of 84.36% and 92.88%, respectively. The proposed method outperformed existing state-of-the-art approaches in terms of both accuracy and computational efficiency, underscoring its effectiveness and efficiency. Our findings highlight its potential for integration into advanced sensor systems for enhanced underwater acoustic target recognition.

Underwater target recognition technology based on sonar images has received considerable critical attention in recent years. However, the sonar sensors encounter disturbance from seafloor reverberation noise and a complicated background, resulting in notable difficulties for precise sonar target classification. On the other hand, traditional machine learning methods inevitably lose features relying on expert systems, and manual network creation is relatively inefficient with limited sonar data. To tackle these challenges, we propose a neural architecture search (NAS)-guided Multiscale Feature Fusion (NAS-MFF) algorithm for sonar images classification based on the differentiable architecture search. Specifically, our approach consists of two stages: a search stage with the Pareto optimization, and a training stage using the optimal architecture. NAS-MFF begins by reconfiguring the search space based on the characteristics of sonar images, which includes the introduction of the MF Block k with multiscale feature extraction ability. By synergizing a recognition-driven convolutional neural network with Pareto optimization, it achieves a dual advantage in both accuracy and model efficiency using the avaliable data. Extensive experiments on three sonar image datasets of different sizes and sources distinctly demonstrate that NAS-MFF outperforms several existing manual design methods and NAS approaches.

This paper focuses on the positioning performance of 3D Time of Arrival (TOA) localization. The performance of TOA localization varies with the target’s position, yet the analytical laws and macro characteristics of this phenomenon remain unclear. Through analytical derivation and geometric intersection analysis, we reveal the spatial distribution characteristics of the positioning accuracy (PA) in a minimum number of parameters system. Our study indicates that PA is a function of the target’s position, sensor positions, and the precision of input parameters. When the sensor array is fixed, PA is proportional to the standard deviation of the input parameters, with its relative distribution characteristics depending on the variance ratios of the input parameters. PA exhibits progressive spatial variations and features an extreme zone, which can be described by tangency conditions. Our research confirms that these tangency conditions are inherent to the system, describing a spatial set of positions that lie on a plane defined by three sensors. The area near this plane exhibits poorer localization performance, forming the algorithm shadow zone. Simulations and experimental validations corroborate the theoretical analysis results. The spatial distribution characteristics of PA provide new insights and theoretical support for enhancing localization performance and applications such as anti-detection.

Neighbor Discovery is a fundamental building block in Wireless Ad Hoc Networks. But the very nature of wireless mobile networks makes it easy to abuse neighbor discovery, such as the wormhole attack which relaying the HELLO packet between two far away nodes make them establishing the neighborship, and thus disrupting the routing function. In this paper, we present a reputation system for neighbor discovery that helps the node to assess the authenticity of the neighborship. The reputation system provides a scheme based on grey theory to compute the neighboring probability which requires less sample data. From building the reputation table, the neighbor node made by the wormhole would be marked the lower value. Through the analysis and simulation, the results show that it is efficient to avoid the harm brought by the wormhole in neighbor discovery.