Recently, deep learning (DL) has achieved great success in biometric recognition. The application of DL has also led to a high demand for biometric data. However, as people attach more importance to privacy protection, biometric data have become increasingly difficult to obtain and access, leading to significant limitations in the development and application of DL-based biometric recognition. Federated learning (FL), a distributed learning technique with privacy protection, provides a potential solution to this problem. Several researchers have attempted to integrate FL into biometric recognition. These studies have shown that the introduction of FL not only solves the conflict between privacy and accessibility of biometric data but also improves the accuracy and generalizability of local recognition systems. Therefore, the combination of FL and biometric recognition techniques has become a new research hotspot. In this survey, we comprehensively review the latest advances regarding the application of FL to biometric recognition, biometric presentation attack detection and the related fields to provide new researchers with a quick and systematic overview of this emerging cross-disciplinary field. This paper also summarizes the future opportunities and challenges of this field. To our knowledge, this is the first survey that systematically organizes and analyses federated biometric recognition and related fields to provide suggestions and references for future research.

Detecting subsequent targets behind obstacles has always been a major challenge for humans. Based on the characteristic that low-frequency through-wall radar signals can penetrate obstacles and reflect on the target surface, we propose a real-time multiperson 3D pose estimation method based on multiple-input multiple-output (MIMO) through-wall radar. Our proposed approach addresses the problem of being unable to recover human poses in real-time when the human body is completely obstructed by an obstacle. This method follows a cross-modal supervision learning pipeline, first using a synchronized camera and a self-designed through-wall radar to obtain paired RGB images and through-wall radar images. Then, we use a monocular camera-based 3D pose estimation method as a teacher network to extract 3D poses from RGB images as a supervision signal. Finally, we design an end-to-end radar pose estimation network based on transformer (ERPENT) as a student network, which learns pose features from two different temporal radar signals and directly predicts the final 3D human pose. Unlike existing radar-based multiperson pose estimation methods, the proposed method in this paper does not require an additional radar target reflection region detection process, thus greatly shortening the necessary pose estimation time. In practical applications, the 3D human pose can be predicted by simply collecting through-wall radar images and inputting them into ERPENT. The experiment shows that under the condition of a 28 cm compact brick wall scene, the average positioning accuracy of human keypoints is 40.15 mm. The time cost of reconstructing a frame of the 3D pose is 0.082 s.

Data missing is a common problem in wireless sensor networks. Currently, to ensure the performance of data processing, making imputation for the missing data is the most common method before getting into sensor data analysis. In this paper, the temporal and spatial nearest neighbor values-based missing data imputation (TSNN), a new imputation based on the temporal and spatial nearest neighbor values has been presented. First, four nearest neighbor values have been defined from the perspective of space and time dimensions as well as the geometrical and data distances, which are the bases of the algorithm that help to exploit the correlations among sensor data on the nodes with the regression tool. Next, the algorithm has been elaborated as well as two parameters, the best number of neighbors and spatial–temporal coefficient. Finally, the algorithm has been tested on an indoor and an outdoor wireless sensor network, and the result shows that TSNN is able to improve the accuracy of imputation and increase the number of cases that can be imputed effectively.

Blast-related mild traumatic brain injury (bmTBI) often leads to long-term sequalae, but diagnostic approaches are lacking due to insufficient knowledge about the predominant pathophysiology. This study aimed to build a diagnostic model for future verification by applying machine-learning based support vector machine (SVM) modeling to diffusion tensor imaging (DTI) datasets to elucidate white-matter features that distinguish bmTBI from healthy controls (HC). Twenty subacute/chronic bmTBI and 19 HC combat-deployed personnel underwent DTI. Clinically relevant features for modeling were selected using tract-based analyses that identified group differences throughout white-matter tracts in five DTI metrics to elucidate the pathogenesis of injury. These features were then analyzed using SVM modeling with cross validation. Tract-based analyses revealed abnormally decreased radial diffusivity (RD), increased fractional anisotropy (FA) and axial/radial diffusivity ratio (AD/RD) in the bmTBI group, mostly in anterior tracts (29 features). SVM models showed that FA of the anterior/superior corona radiata and AD/RD of the corpus callosum and anterior limbs of the internal capsule (5 features) best distinguished bmTBI from HCs with 89% accuracy. This is the first application of SVM to identify prominent features of bmTBI solely based on DTI metrics in well-defined tracts, which if successfully validated could promote targeted treatment interventions.

Legal judgments are generally very long, and relevant information is often scattered throughout the text. To complete a legal judgment summarization, capturing important, relevant information comprehensively from a lengthy text is crucial. The existing abstractive-summarization models based on pre-trained language have restrictions on the length of an input text. Another concern is that the generated summaries have not been well integrated with the legal judgment’s technical terms and specific topics. In this paper, we used raw legal judgments as information of different granularities and proposed a two-stage text-summarization model to handle different granularities of information. Specifically, we treated the legal judgments as a sequence of sentences and selected key sentence sets from the full texts as an input corpus for summary generation. In addition, we extracted keywords related to technical terms and specific topics in the legal texts and introduced them into the summary-generation model as an attention mechanism. The experimental results on the CAIL2020 and the LCRD datasets showed that our model achieved an overall 0.19–0.41 improvement in its ROUGE score, as compared to the baseline models. Further analysis also showed that our method could comprehensively capture essential and relevant information from lengthy legal texts and generate better legal judgment summaries.

Long-term respiratory monitoring plays an extremely important role in the diagnosis of respiratory system-related diseases. It can detect diseases temporally and improve the effectiveness of treatment, which is crucial for home health monitoring. With the rapid development of radar chip technology, noncontact respiratory monitoring based on commercial millimeter-wave radar sensors has received increasing attention. However, traditional signal processing methods are difficult to extract fine respiratory waveforms from radar signals mixed with human body movements. Fortunately, deep learning provides a solution, as we can use its powerful learning ability to learn a mapping from radar signals to real respiratory waveforms. With this method, we can directly observe fine-grained respiratory waveforms and further improve the accuracy of respiratory rate detection. Therefore, we propose MM-FGRM, a respiratory monitoring system based on a commercial 77 GHz multiple input multiple output (MIMO) radar. The core of this system is a deep learning-based network called IQ-Transformer with a self-attention mechanism that aims at capturing the latent respiratory-related features from I/Q components of radar signals directly in each human body region of interest (ROI) and recovering the respiratory waveforms. We collected 12 h of data from eight subjects and conducted experiments. The experimental results show that MM-FGRM can accurately recover respiratory waveforms and provide accurate respiratory rates. In addition, we perform testing on data from two other users who do not participate in the training, and the results verify that MM-FGRM has a strong generalization ability. Our results demonstrate the feasibility of further development of home respiratory monitoring products.