Nonlinear blind source separation (BSS) is one of the unsolved problems in unsupervised learning, because the solutions are highly non-unique when there is no prior information for the mixing functions. In this paper, we present a novel approach to tackle the ill-posedness of the nonlinear BSS problem with a few assumptions. The derivation of our algorithm is inspired by the idea of an efficient layer-by-layer representation to approximate the nonlinearity. Once such a representation is built, a final output layer is constructed by solving a convex optimization problem. Relying on the multi-layer architecture, the algorithm transforms a time-invariant nonlinear BSS to the local linear problem with a tolerable computational cost. Then, the projected data can break the nonlinear problem down into the version of a generalized joint diagonalization problem in the feature space. Importantly, the parameters and forms of polynomials depend solely on the input data, which guarantee the robustness of the structure. We thus address the general problem without being restricted to any specific mixture or parametric model. Experiments show that the proposed algorithm has a higher estimation accuracy on audio data sets from the real world for separating various nonlinear mixtures.

Change detection based on synthetic aperture radar (SAR) images is an important application in the remote sensing technology field. However, the lack of labeled data has been a difficult problem in SAR image detection, especially for pixel-level change detection. In this letter, we propose a novel unsupervised change detection algorithm, which improves the detection accuracy by exploring features from both spatial and frequency domains of SAR images. In particular, firstly clustering is used as pre-classification to obtain pseudo-labels, and then by incorporating classifiers and pseudo-labels in terms of feature learning, a novel unsupervised detection algorithm is proposed. In order to improve the sensitivity of the algorithm to changed details and enhance the anti-noise ability of the change detection network, the attention mechanism (AM) is integrated into the network to fully extract important spatial structure information. Moreover, a multi-domain fusion module is proposed to integrate spatial and frequency domain features into complementary feature representations. This module contains multi-region features weighted by the channel-spatial AM and deep features filtered out by the gated linear units (GLU) in the frequency domain. To verify the effectiveness of the proposed algorithm, it is compared against the other four SAR image change detection algorithms using three real datasets. The experimental results show that the proposed method outperforms the other four algorithms in terms of percent correct classification (PCC) and Kappa coefficient (KC).

Synthetic aperture radar (SAR) plays a crucial role in maritime surveillance due to its capability for all-weather, all-day operation. However, SAR ship recognition faces challenges, primarily due to the imbalance and inadequacy of ship samples in publicly available datasets, along with the presence of numerous outliers. To address these issues, this paper proposes a SAR ship classification method based on text-generated images to tackle dataset imbalance. Firstly, an image generation module is introduced to augment SAR ship data. This method generates images from textual descriptions to overcome the problem of insufficient samples and the imbalance between ship categories. Secondly, given the limited information content in the black background of SAR ship images, the Tokens-to-Token Vision Transformer (T2T-ViT) is employed as the backbone network. This approach effectively combines local information on the basis of global modeling, facilitating the extraction of features from SAR images. Finally, a Squeeze-and-Excitation (SE) model is incorporated into the backbone network to enhance the network’s focus on essential features, thereby improving the model’s generalization ability. To assess the model’s effectiveness, extensive experiments were conducted on the OpenSARShip2.0 and FUSAR-Ship datasets. The performance evaluation results indicate that the proposed method achieves higher classification accuracy in the context of imbalanced datasets compared to eight existing methods.

Object tracking based on RGB images may fail when the color of the tracked object is similar to that of the background. Hyperspectral images with rich spectral features can provide more information for RGB-based trackers. However, there is no fusion tracking algorithm based on hyperspectral and RGB images. In this paper, we propose a reliability-guided aggregation network (RANet) for hyperspectral and RGB tracking, which guides the combination of hyperspectral information and RGB information through modality reliability to improve tracking performance. Specifically, a dual branch based on the Transformer Tracking (TransT) structure is constructed to obtain the information of hyperspectral and RGB modalities. Then, a classification response aggregation module is designed to combine the different modality information by fusing the response predicted through the classification head. Finally, the reliability of different modalities is also considered in the aggregation module to guide the aggregation of the different modality information. Massive experimental results on the public dataset composed of hyperspectral and RGB image sequences show that the performance of the tracker based on our fusion method is better than that of the corresponding single-modality tracker, which fully proves the effectiveness of the fusion method. Among them, the RANet tracker based on the TransT tracker achieves the best performance accuracy of 0.709, indicating the effectiveness and superiority of the RANet tracker.

Despite wearable electronic devices have enormous potential applications, the recording quality has not sufficiently met the requirements for fetal disease detection. In this paper, we aim to propose a novel framework for fetal ECG extraction, where the hybrid approach mainly considers the combination of a novel adaptive maternal QRS removal (AQRSR) algorithm and a tensor-based joint blind source separation (JBSS) approach. To adapt to the dynamic change of the ECG signal and suppress the powerful maternal heartbeat, AQRSR is explored with an adaptive template, which is generated by a model consisting of a generator, discriminator, and transformer. By replacing the oldest segmentation with the newest cycle, the approach is capable to quickly adapt to the estimated output to match the new input. In addition, due to significant cross-over between the target signal and the noise, even if the maternal ECG removal processing is implemented, the residual signal inevitably contains fetal ECG, part of maternal ECG, and noise. As a consequence of this, a novel joint blind source separation that incorporates tensor decomposition is formulated to separate fetal heartbeat from long-term recordings contaminated with maternal movements noise and heartbeat interference. By dividing the collected data into multiple segments, this method can extract fetal ECG signals by exploiting the signals that are not only statistically independent within each segment but can be dependent on different segments. Experimental results show that even if the recorded signals are contaminated by the noise, the proposed framework has comparable better performance on fetal ECG extraction.