Radio Frequency Identification (RFID)-based human activity recognition (HAR) holds strong potential for ubiquitous sensing due to its capabilities of identity-awareness, privacy-preservation, and long-term passive monitoring. However, most existing methods are constrained to offline classification, which significantly limits their deployment in real-time scenarios. In this paper, we propose RF-OnlineHAR, a real-time HAR framework that enables online, frame-by-frame recognition directly on streaming RFID signals. To better capture short-term temporal characteristics, we introduce a novel Tag Activity Representation Frame (TARF) that jointly encodes the dynamics of physical-layer signals and the irregularity of tag responses. Furthermore, we present the Memory-Augmented Streaming Transformer Encoder (MASTE), which integrates local attention with dynamic memory to capture long-range dependencies across sliding windows, enabling context-aware real-time recognition. Experimental results demonstrate that RF-OnlineHAR achieves 98.61% accuracy on the offline CWNU-RDA dataset, outperforming existing state-of-the-art methods, and attains 94.72% mAP on RF-StreamHAR, a dedicated real-time dataset constructed for continuous HAR evaluation. The proposed framework contributes toward bridging the gap between algorithmic development and practical deployment of RFID-based activity recognition systems, and shows promising potential for applications in areas such as smart healthcare and eldercare.

Multivariate time series forecasting (MTSF) is of immense significance in extensive domains, such as traffic analysis and weather forecasting. Despite the accuracy of MTSF has made significant progress in recent years, most research efforts typically require prohibitive spatio-temporal overhead, especially when dealing with higher-dimensional data. This limitation hinders the scalability of the model for real-world applications. To address this issue, we propose LSSM, a Low-Rank State Space Model that conducts multivariate time series forecasting at lower computation costs while effectively capturing intricate long-range dependencies and variable correlations within data. We apply the KL-constraint during training which enhances both the generalization of the model and the association between subsequences. Experiment results on four benchmark datasets verify the superiority of our approach compared with the state-of-the-art baselines.

In recent years, wireless radio-frequency identification (RFID) technology has shown great potential in human activity recognition (HAR) due to its passive sensing capability, low deployment cost, and strong privacy protection. However, before its widespread application in real-world scenarios, challenges such as heterogeneous data distributions, scarcity of labeled samples, and data security must be addressed. In this paper, we propose Fed-Siamese, a federated meta-learning based HAR system designed to overcome these challenges through several novel contributions. To address the challenge of data distribution heterogeneity across clients, we first model each client’s activity recognition task as an independent learning problem. Building on this, we propose a federated training algorithm and a trusted parameter aggregation strategy based on KL divergence, all within the framework of federated learning and the Reptile meta-learning optimization algorithm. Our approach enables the distributed training of a Siamese network, thereby mitigating the risk of data leakage inherent in centralized batch training. Furthermore, we design a personalized fusion fine-tuning algorithm to balance adaptation efficiency and recognition performance, enabling the system to quickly adapt to new clients using a few labeled samples. Experimental results show that the proposed system outperforms the baselines in recognition performance on both RFID activity datasets, while maintaining stable model convergence under byzantine and backdoor attacks.

In recent years, Large Language Models (LLMs) have demonstrated exceptional performance in code-generation tasks. However, under enterprise scenarios where private APIs are pre-built, general LLMs often fail to meet expectations. Existing approaches are confronted with drawbacks of high resource consumption and inadequate handling of multi-API tasks. To address these challenges, we propose EpiGEN, an Efficient multi-Api code GENeration framework under enterprise scenario. It consists of three core modules: Task Decomposition Module (TDM), API Retrieval Module (ARM), and Code Generation Module (CGM), in which Langchain played an important role. Through a series of experiments, EpiGEN shows good acceptability and readability, compared to fully fine-tuned LLM with a larger number of parameters. Particularly, in medium and hard level tasks, the performance of EpiGEN on a single-GPU machine even surpasses that of a fully fine-tuned LLM that requires multi-GPU configuration. Generally, EpiGEN is model-size agnostic, facilitating a balance between the performance of code generation and computational requirements.

In the field of indoor human behavior recognition, traditional deep learning methods often underperform in adequately capturing temporal features and core information in RFID behavior data, leading to suboptimal predictive performance. We note that human behavior is primarily composed of complex action sequences, where certain sequences formed by Received Signal Strength Indicator (RSSI) values may exhibit more representative and information-rich patterns. Therefore, we propose SATCN: an innovative model that leverages Temporal Convolutional Networks (TCN) and self-attention mechanisms for advanced RFID-based human behavior recognition. The SATCN model employs TCN to adeptly capture the inherent temporal dependencies within behavior data, while integrating self-attention mechanisms to focus on high-value features in the data. Experimental validation on RFID indoor human behavior recognition dataset reveals that SATCN achieves an accuracy of 94.51%, a 10.83 percentage point increase over existing models, highlighting SATCN’s potential in enhancing the accuracy of RFID-based indoor human behavior recognition.

Amidst the rapid advancement of Internet of Things (IoT) technology, achieving precise indoor localization has emerged as a pivotal research area. Localization algorithms relying on Radio Frequency Identification (RFID) received signal strength indicator (RSSI) have gained widespread adoption in numerous indoor positioning systems due to their straightforward implementation and cost-effectiveness. However, in indoor settings, challenges like building obstructions and multipath effects often lead to signal reception failures by RFID antennas, consequently compromising the reliability of positioning outcomes. Recent research has approached indoor localization as a regression problem, employing deep learning models for analysis and prediction. But most current indoor localization models primarily focus on either spatial or temporal features within RSSI data, leading to suboptimal localization outcomes. To tackle these challenges, this paper proposes an enhanced methodology that leverages Generative Adversarial Networks (GAN) to impute missing RSSI data. Additionally, Convolutional Neural Networks (CNN) are utilized to extract spatial domain features, while Long Short-Term Memory Networks (LSTM) are employed for extracting temporal domain features. Ultimately, this paper designs a novel model, GCLA, which integrates an Attention mechanism with a location coding strategy to fuse features for precise location prediction. Experimental results show that the proposed GCLA model can obtain stable localization results after a short training on a small number of datasets.

Performing cross-domain activity recognition using a few samples is a challenge in the field of radio frequency identification (RFID) sensing. Due to multipath effect, RF signals collected from different domains exhibit unbalanced label distribution and heterogeneous signal distribution, which leads to domain shift problem. Moreover, constrained by cost, the labeled data available for training in each domain is limited. In this paper, we propose RFRN, an RFID activity recognition system based on Relation Network, and improve the recognition accuracy through several designs. First, the structure of Relation Network is improved to extract activity-related features, filter domain-related interference and then match different activities without additional fine-tuning. Second, a new task generation strategy is proposed to make full use of source domains and enable the model to experience more tasks. Experiments demonstrate that RFRN can effectively handle the domain shift problem and adapt well to new domains with a few samples. With one and five samples of each activity, RFRN outperforms the baselines by at least 7.3% and 4.0% on a real-world dataset with balanced label distribution and around 22.2% and 7.7% on a dataset with unbalanced label distribution, respectively.

Unsupervised domain adaptation (UDA) methods have achieved significant success in the cross-modality medical image segmentation tasks. However, due to the inherent properties of medical images, i.e., the large distribution gap between different modalities and the low intensity contrast between different categories of organ structures, it is challenging for current UDA methods to perform well on the boundary regions. In this paper, we propose a boundary contrast domain adaptation framework for cross-modality medical image segmentation. Concretely, we consider the same category prototypes as positive samples to reduce category-level distribution differences between domains, and boundary features as negative samples to improve the discriminative ability of ambiguous boundary regions. And we construct mixed samples by bidirectional cross-domain cutmix for self-training to further reduce the domain gap. Moreover, we dynamically assign weights to different parts of the pseudo-labels to prevent model degradation. Experimental results show our proposed method outperforms the state-of-the-art methods.

Nowadays, IoT technology is developing rapidly and RFID (Radio Frequency Identification) based indoor positioning problems can be performed using deep learning and intelligent optimization algorithms. Deep models can analyze and predict the localization problem as a regression problem to achieve high accuracy positioning. Meanwhile, to ensure the accuracy of the model, we need to find excellent hyperparameters, which requires the support of optimization algorithms, but existing optimization algorithms do not allow flexible adaptation according to the optimization phase and there is room for improvement. In this paper, we propose a deep model, called CTT, and a multi-objective evolutionary algorithm (MOEA) based on a neighborhood adaptive adjustment strategy, called MOEA-NAAS. The experimental results show that CTT optimized by the NAAS algorithm is significantly more accurate and stable in the localization problem, with significant improvements in the three main metrics, proving the usability of the optimization algorithm. At the same time, the localization effect of the CTT also shows obvious advantages. In the future, the optimized algorithm can be combined with other deep models and widely used in various high-precision indoor positioning.

Wireless sensing techniques for Human Activity Recognition (HAR) have been widely studied in recent years.At present, the research on HAR based on Radio Frequency Identification (RFID) is changing from the tag attachment method to the tag non-attachment method.Affected by multipath, the current solutions in tag non-attachment scenarios mainly focus on singleobject activity recognition, which is not suitable for multi-object scenarios.To address these issues, we propose DeepMultiple, a novel tag non-attachment activity recognition model for multiobject.The model first preprocesses the raw signal with filter and phase calibration, then it applies dilated convolution in the frequency domain to extract multi-object activity features, finally ProbSparse is used to optimize the vanilla Transformer-based Encoder to enhance the activity recognition ability.We deployed a single reader and antenna for multi-object activity tracking during the experiments to reduce deployment difficulties.Extensive experimental results show that DeepMultiple can recognize ten types of multi-object activities with 98.12% precision under different challenging settings, which has excellent performance compared with several state-of-the-art methods.

Wireless sensing techniques for Human Activity Recognition (HAR) have been widely studied in recent years. At present, research on HAR based on Radio Frequency Identification (RFID) is changing from the tag attachment method to the tag non-attachment method. Affected by multipath, the current solutions in tag non-attachment scenarios mainly focus on single-object activity recognition, which is not suitable for multi-object scenarios. To address these issues, we propose STMultiple, a novel tag non-attachment activity recognition model for multi-object. The model first preprocesses the raw signal with filter and phase calibration, then it applies dilated convolution in the frequency domain to extract multi-object activity features, finally the feature pyramid structure and ProbSparse are used to optimize the vanilla Transformer-Encoder to enhance the activity recognition ability. Extensive experiments show that STMultiple can achieve recognition accuracy of up to 97.93% and down to about 90% in challenging environments ranging from two to five users, which has excellent performance compared to several state-of-the-art methods.

Depression is a common mental disease, and it is committed to promote the research of depression assessment based on physiological signals. By collecting the resting state EEG data of depressive disorder, we collected the resting state EEG data of 14 patients with depression and 17 normal people. Through the analysis, the number of troughs of each person's data was statistically analyzed, combined with the convolution neural network model. The accuracy rate of some data is 94.88% by the statistical trough number method, and the accuracy rate of the remaining part of the test data set is 85.0% through the convolution neural network model, and the final fusion accuracy rate is 86.88%. The experimental results show that the combination of statistical trough number and convolution neural network can distinguish depression patients better in resting state EEG data.

Google Maps is an excellent GIS (Geographic Information System) product, but its visual effect is unsatisfactory under the specific environment. As is known to all, visualization is an important aspect to evaluate the quality of electronic map. Firstly, we analyze factors of visual effect in the paper. Combining implementation mechanisms of RFID (Radio Frequency Identification Devices) and Google Maps, we designed the system of electronic map which owns a reasonable amount of displayed information, beautiful legends and fast reaction rate. After that, we made an example to verify the effectiveness of the system's performance.