Jiaquan Yan

Existing multi-color space guided techniques for underwater image enhancement (UIE) fail to take the advantages of the XYZ color space for preserving underwater image details, meanwhile, existing UIE datasets, typically containing low-quality reference images of distorted colors and blurred structures, lead to inaccurate enhancement mapping between low-quality and high-quality images. To overcome these above limitations, we propose a Multi-Color Space Fusion Network (MCSF-Net) for UIE. The MCSF-Net incorporates a Multi-dimensional Feature Fusion Block (MFFB) and weighted feature fusion scheme to effectively integrate complementary features from both XYZ and RGB color spaces. Moreover, we establish a Large-Scale Mixed UIE dataset (LSMU) by using nine no-reference metrics to filter out low-quality reference images from eight public UIE datasets, enabling more effective network learning. Extensive experiments on mainstream datasets demonstrate that the proposed method outperforms several leading approaches in both color restoration and detail enhancement of various underwater images. The code and dataset for MCSF-Net will be available at https://github.com/WYJGR/MCSF-Net.

Objective. Signal denoising methods based on deep learning have been extensively adopted on electroencephalogram devices. However, they are unable to deploy on edge-based portable or wearable (P/W) electronics due to the high computational complexity of the existed models. To overcome such issue, we propose an edge-based lightweight Kalman filter network (EKFNet) that does not require manual prior knowledge estimation. Approach. Specifically, we construct a multi-scale feature fusion module to capture multi-scale feature information and implicitly compute the prior knowledge. Meanwhile, we design an adaptive gain estimation module that incorporates long short-term memory and sequential channel attention module to dynamically predict the Kalman gain. Furthermore, we present an optimization strategy utilizing operator fusion and constant folding to reduce the model’s computational overhead and memory footprint. Main results. Experimental results show that the EKFNet reduces the sum of the square of the distances by at least 12% and improves the cosine similarity by at least 2.2% over the state-of-the-art methods. Besides, the model optimization shortens the inference time by approximately 3.3×. The code of our EKFNet is available at https://github.com/cathnat/EKFNet. Significance. By integrating Kalman filter with deep learning, the approach addresses the parameter-setting challenges in traditional algorithms while reducing computational overhead and memory consumption, which exhibits a good tradeoff between algorithm performance and computing power.

In this paper, a novel time-frequency signature using resonance-based sparse signal decomposition (RSSD), phase space reconstruction (PSR), time-frequency distribution (TFD) and manifold learning is proposed for feature extraction of ship-radiated noise, which is called resonance-based time-frequency manifold (RTFM). This is suitable for analyzing signals with oscillatory, non-stationary and non-linear characteristics in a situation of serious noise pollution. Unlike the traditional methods which are sensitive to noise and just consider one side of oscillatory, non-stationary and non-linear characteristics, the proposed RTFM can provide the intact feature signature of all these characteristics in the form of a time-frequency signature by the following steps: first, RSSD is employed on the raw signal to extract the high-oscillatory component and abandon the low-oscillatory component. Second, PSR is performed on the high-oscillatory component to map the one-dimensional signal to the high-dimensional phase space. Third, TFD is employed to reveal non-stationary information in the phase space. Finally, manifold learning is applied to the TFDs to fetch the intrinsic non-linear manifold. A proportional addition of the top two RTFMs is adopted to produce the improved RTFM signature. All of the case studies are validated on real audio recordings of ship-radiated noise. Case studies of ship-radiated noise on different datasets and various degrees of noise pollution manifest the effectiveness and robustness of the proposed method.

Whale sounds may mix several elements including whistle, click, and creak in the same vocalization, which may overlap in time and frequency, so it leads to conventional signal separation techniques challenging to be applied for the signal extraction. Unlike conventional signal separation techniques which are based on the frequency bands, such as WT and EMD, tunable‐Q wavelet transform (TQWT) can separate the objected signal into particular components with different structures according to its oscillation property and eliminate in‐band noise using the basis pursuit method. Considering the characteristics of oscillatory and transient impulse, we propose a novel signal separation method for whale whistle and click extraction. The proposed method is performed by the following two steps: first, TQWT is used to construct the dictionary for sparse representation. Secondly, the whale click and whistle construction are performed by designing the basis pursuit denoising (BPD) algorithm. The proposed method has been compared with one of the popular signal decomposition techniques, i.e., the EMD method. The experimental results show that the proposed method has a better performance of click and whistle signal separation in comparison with the EMD algorithm.

Under the complex oceanic environment, robust and effective feature extraction is the key issue of ship radiated noise recognition. Since traditional feature extraction methods are susceptible to the inevitable environmental noise, the type of vessels, and the speed of ships, the recognition accuracy will degrade significantly. Hence, we propose a robust time-frequency analysis method which combines resonance-based sparse signal decomposition (RSSD) and Hilbert marginal spectrum (HMS) analysis. First, the observed signals are decomposed into high resonance component, low resonance component, and residual component by RSSD, which is a nonlinear signal analysis method based not on frequency or scale but on resonance. High resonance component is multiple simultaneous sustained oscillations, low resonance component is nonoscillatory transients, and residual component is white Gaussian noises. According to the low-frequency periodic oscillatory characteristic of ship radiated noise, high resonance component is the purified ship radiated noise. RSSD is suited to noise suppression for low-frequency oscillation signals. Second, HMS of high resonance component is extracted by Hilbert-Huang transform (HHT) as the feature vector. Finally, support vector machine (SVM) is adopted as a classifier. Real audio recordings are employed in the experiments under different signal-to-noise ratios (SNRs). The experimental results indicate that the proposed method has a better recognition performance than the traditional method under different SNRs.