Bovine milk adulteration with coconut milk poses a significant threat to food safety, as both liquids are visually similar yet nutritionally distinct. This study presents an integrated analytical framework combining excitation–emission matrix (EEM) fluorescence spectroscopy with chemometric and deep learning techniques to detect and quantify coconut milk adulteration in bovine milk across nine concentration levels (0–100% v/v). Parallel factor analysis (PARAFAC) resolved two dominant fluorescent components, tryptophan (λ ex/em: 290/350 nm) and riboflavin (λ ex/em: 450/525 nm), whose scores decreased monotonically with increasing adulteration, confirming their role as key chemical biomarkers. For quantitative prediction, PLSR and 1D-CNN models were developed using emission spectra at three excitation wavelengths, with best performances achieved at 450 nm (PLSR: R2P = 0.97, RMSEP = 5.00%; 1D-CNN: R2P = 0.94, RMSEP = 6.75%). A lightweight 2D-CNN utilizing full EEM contour maps as image inputs outperformed all quantitative models, achieving R2P = 0.99, RMSEP = 2.36%, and RPD = 12.97, demonstrating the advantage of preserving the full two-dimensional fluorescence topology over discrete wavelength selection. These results confirm that EEM combined with 2D-CNN provides a highly accurate and non-destructive tool for dairy authentication.

Infrared–visible image fusion aims to integrate structural details, natural appearance, and thermal target information from two source modalities, thereby improving visual perception in complex scenes. However, under challenging conditions such as low illumination, noise, low contrast, and overexposure, existing methods often struggle to stably preserve cross-modal shared features (CMSF) while effectively highlighting single-modal specific features (SMSF). In addition, the absence of real fusion labels limits effective supervised learning. To address these issues, this paper proposes a unified semantic-guided fusion network, termed USF-Net, which jointly models the shared and specific features of infrared and visible images under a unified semantic representation and dynamically adjusts the fusion strategy according to imaging contexts. Specifically, the Shared Feature Alignment and Enhancement (SFAE) module is designed to strengthen consistent modeling of common features across modalities, while the Specific Feature Reweighting Fusion (SFRF) module selectively enhances modality-specific features to achieve stable and controllable fusion. Moreover, the constructed real fusion labels are incorporated into the loss function for collaborative training. Experimental results on multiple public datasets demonstrate that USF-Net achieves superior fusion performance under diverse complex imaging conditions.

This paper will introduce a radio frequency system to track the location of a stent designed to work inside a human artery. The stent is designed as a hemostasis aid tool for emergency situations where common surgical equipment, such as fluoroscopy systems, is not available, such as on the battlefield. In the application of interest, the stent must be guided to the correct location to achieve effective hemostasis and prevent complications. The locating approach uses the radiation pattern from the transmitter as the reference. When the transmitting frequency changes over a certain range, the measurement amplitude from a receiver depends on its relative location with respect to the transmitter. However, when the input frequency is unequal to the resonance frequency, the radiation pattern varies in an unpredictable way. To solve this problem, a deep learning model was trained to recognize variations in the radiation pattern and predict the receiver’s location as one of the classes in the reference grid. The deep learning model also reduces the impact of noise and disturbing signals, which effectively improves the system’s robustness.

HighlightsWhat are the main findings?CysAuNPs are used for colorimetric detection of As (III) and Hg (II) Pin serum albumin.XRF enables the qualitative detection of As (III) and H Hg (II) ions in serum albumin.The binding of As (III) and Hg (II) ions resulted in serum albumin conformational changes.LOD, as low as 0.02 ppm for Hg (II), demonstrates the method’s sensitivity.The accurate prediction of As (III) and Hg (II) for > seven months without recalibrations.What are the implications of the main findings?New alternative for rapid detection of As (III) and Hg (II) ions in physiologically relevant biological samples.This detection method allows for improved accessibility through lower costs and ease of use.A continued interest in developing a low-cost, rapid screening method for quantifying Hg (II) and As (III) in biological samples stems from the toxic effects of human exposure to these heavy metal ions. This study reports the use of cysteine-capped gold nanoparticles (CysAuNPs) for chemical sensing, colorimetric detection, and quantification of As (III) and Hg (II) ions in human serum albumin (HSA) under physiological conditions. Zeta potential measurements indicated that the CysAuNPs have a negative surface charge, which was decreased in the presence of HSA and reversed to a positive value upon binding of As (III) and Hg (II) metal ions. Circular dichroism (CD) spectroscopy revealed changes in HSA conformation upon binding to As (III) and Hg (II) ions. X-ray fluorescence enables rapid qualitative screening for As (III) and Hg (II) ions before colorimetric quantification. The figures of merit (R2 ≥ 0.940) and the low detection limits (0.05 ppm for As (III) ions and 0.02 ppm for Hg (II)) in serum albumin demonstrate the high sensitivity of the method. The developed calibration curves correctly quantified the concentration of As (III) and Hg (II) ions of independently prepared test validation samples in HSA with an accuracy of ≥95% over a period of seven months without recalibrations, demonstrating the stability of CysAuNPs in solution and the robustness of the method for analysis of As (III) and Hg (II) ions in serum albumin.

As data-intensive applications continue to scale reconfigurable array processors (RAPs), electrical networks-on-chip (NoCs) are increasingly constrained by energy-delay bottlenecks due to RC-delay constraints. Hybrid optoelectronic NoCs (HONoCs) suffer from a fundamental medium-selection dilemma: optical circuit switching incurs microsecond-scale setup overheads for long flows, whereas static distance thresholds fail to capture the spatiotemporal heterogeneity of traffic, causing wavelength waste for bursty flows and congestion diffusion under non-stationary loads. This paper presents an adaptive switching framework that is aware of spatiotemporal locality. We introduce the Temporal-Spatial Locality Index (TSLI) to classify flows into Electrophilic (EF), Photophilic (PF), and Hybrid-sensitive (HF) categories, and propose Cross-layer Congestion Entropy (CCE) to unify electrical and optical resource states. Based on these metrics, an Adaptive Medium Selection State Machine (AMSSM) dynamically switches among Electro-Dominant (EDM), Electro-Optical Synergistic (EOSM), and Optical-Dominant (ODM) modes, while a Weighted Multi-dimensional Medium Matching (WMMM) algorithm performs fine-grained channel selection. A Predictive Optical Path Provisioning (POPP) mechanism further amortizes setup latencies via trend-aware pre-establishment. Evaluation on an 8 × 8 mesh HONoCs demonstrates 22% higher saturation throughput, 38% lower energy-delay product (EDP), and 57% reduction in average latency under non-stationary traffic, compared to static thresholds. The proposed mechanisms provide a theoretical foundation and engineering paradigm for efficient on-chip interconnects.

Major depressive disorder (MDD) is a prevalent and severe mental disorder, and EEG-based automated detection has become a promising approach for auxiliary screening diagnosis. In this work, we propose a novel multiscale spatiotemporal graph neural network for MDD detection from multichannel EEG signals. Specifically, a left–right hemispheric partitioning prior is used to encode brain functional organization. Based on this partitioning, adaptive graphs are then constructed and graph message passing is performed to model intra-hemispheric interactions. The approach not only incorporates brain functional organization into the learning process but also enhances the extraction of discriminative features related to depressive brain dynamics. The proposed method was validated in a cross-subject scenario on a private resting-state EEG dataset including 54 adult participants (27 MDD patients and 27 healthy controls; age range: 27–48 years). Experimental results on the dataset achieve an accuracy of 92.21%, surpassing the baseline models. Meanwhile, ablation experiments demonstrate the effectiveness of our proposed method.

HighlightsWhat are the main findings?A Multi-Stage Feature Selection (MSFS) framework integrating Pearson correlation, mutual information, and Boruta was developed to optimize high-dimensional airborne LiDAR point cloud features for individual-tree DBH estimation.The proposed MSFS–XGBoost model significantly improved prediction accuracy, achieving an of 0.901 and an RMSE of 1.647 cm, outperforming DTR, RFR, and GBM models.What are the implications of the main findings?The proposed feature optimization strategy effectively reduces redundancy in LiDAR-derived features and enhances model stability for forest structural parameter estimation.The MSFS–XGBoost framework provides a reliable approach for accurate individual-tree DBH estimation and supports refined forest resource monitoring using airborne LiDAR data.Diameter at breast height (DBH) is a fundamental structural parameter for forest inventory and ecological analysis. However, field-based measurements (e.g., diameter tape surveys) are labor-intensive and inefficient for large-scale applications. Airborne light detection and ranging (LiDAR) provides an efficient alternative for individual-tree DBH estimation. Nevertheless, LiDAR-derived features—defined as statistical descriptors of point cloud structure and radiometric properties—are typically high-dimensional and redundant, which may degrade model performance. To address this issue, this study proposes an integrated framework combining Multi-Stage Feature Selection (MSFS) and Extreme Gradient Boosting (XGBoost) for DBH estimation. A total of 104 variables, including LiDAR-derived features (height, density, intensity, and canopy structure metrics) and structural parameters (tree height, crown diameter, and crown area), were used as predictors. The MSFS framework was applied to progressively reduce feature redundancy and identify an optimal subset, which was then used to train the XGBoost model. The results demonstrate that the MSFS–XGBoost model achieved the best performance, with a coefficient of determination (R2) of 0.901 and a root mean square error (RMSE) of 1.647 cm. Compared with models using the original feature set, R2 increased by 0.384 and RMSE decreased by 1.146 cm. These findings indicate that the proposed framework effectively improves DBH estimation accuracy and provides a reliable approach for individual-tree parameter estimation and large-scale forest resource monitoring using airborne LiDAR data.

(1) Background: Breast cancer screening remains limited by mammography, particularly in younger women, in dense breast tissue, and in the detection of interval cancers. The PHI-BRA Smart Bra was developed as a wearable, non-invasive device combining thermography and bioimpedance for frequent breast monitoring. This first-in-human study aimed to assess the feasibility and in vivo diagnostic performance of the PHI-BRA system in discriminating between women with and without breast lesions. (2) Methods: A prospective feasibility study was conducted between March 2023 and February 2024. A calibration cohort (n = 15) was used to define the discrimination model, followed by an analysis cohort (n = 26; 13 with breast lesions and 13 without). Thermal and bioimpedance signals were acquired using the PHI-BRA device. Diagnostic performance was evaluated using receiver operating characteristic (ROC) analysis, with mammography as the reference standard. (3) Results: In the analysis cohort, the temperature-based model achieved an area under the ROC curve (AUC) of 80.8% (95% CI [63.2–98.3]). At the optimal threshold, sensitivity was 84.6% (95% CI [61.5–100]) and specificity was 76.9% (95% CI [53.8–100]). Exploratory bioimpedance analyses showed lower sensitivity but high specificity, mainly limited by sensor contact stability. No adverse events were reported. (4) Conclusions: This first-in-human study demonstrates an initial exploration of the feasibility and safety of a wearable thermography-based approach for breast lesion discrimination. The results support further clinical validation of a multimodal wearable system as a complementary tool to existing breast cancer screening strategies.

In recent years we have seen large language models (LLMs) demonstrating robust reasoning capabilities comparable to human performance. This makes them increasingly appealing for driver assistance, where adaptation to dynamic human context is essential. Yet, research in this area remains fragmented, often focusing on isolated applications, lacking utilization of LLM’s full potential to deliver integrated, context-specific support and action. This survey synthesizes recent advancements in LLM-driven occupant monitoring systems, focusing on their capabilities for interpreting driver states and acting appropriately, enabling a new generation of intelligent driver assistance. We critically examine pioneering frameworks, benchmarks, and foundational datasets that employ techniques like reasoning chains, multimodality, and human-in-the-loop feedback to create personalized and safe driving experiences. We lay out the current trends, limitations, and emerging patterns, in addition to a novel human-centered evaluation of the field, providing researchers with a roadmap towards transparent and trustworthy in-cabin systems that bridge safety with driver experience.

Withthe widespread deployment of intelligent terminals, mobile payment platforms, and Internet of Things devices, security systems are being progressively transformed from traditional transaction outcome analysis toward an intelligent perception paradigm centered on user behavior, device states, and environmental context. To address the challenges of multimodal data heterogeneity, non-independent and identically distributed data across nodes, and the difficulty of centralized modeling under privacy constraints in distributed scenarios, an artificial intelligence-driven federated multimodal security perception framework, namely FMS-LLM, is proposed. At its core, the framework introduces a Non-IID adaptive federated fusion mechanism that achieves dual-level alignment—structural alignment via parameter-level masks and semantic alignment via feature consistency constraints—to effectively mitigate cross-node distribution discrepancies. Additionally, an LLM-driven semantic enhancement module is developed, utilizing trend-guided token selection and inertia-suppression to map low-level sensing features into high-level risk semantic representations, thereby supporting logical reasoning and explainable decision-making. This framework takes user behavioral sensing data, device state information, environmental context data, and transaction behavior data as inputs, and constructs an integrated security analysis pipeline of “perception–collaboration–reasoning”. Experimental results on the distributed multimodal security perception task demonstrate that the proposed method achieves an Accuracy of 91.62%, a Precision of 91.04%, a Recall of 90.37%, an F1-score of 90.70%, and a ROC-AUC of 94.73%, consistently outperforming baseline methods including Logistic Regression, Random Forest, LSTM, the centralized multimodal deep model, FedAvg, FedProx, and MOON. Under strongly Non-IID conditions, when , the model still maintains an Accuracy of 88.47% and an F1-score of 87.11%, demonstrating stronger cross-node robustness. The ablation study further indicates that the complete model attains the best classification performance while reducing communication cost to 18.92 MB/Round. These results demonstrate that the proposed method can effectively fuse multi-source sensing information under privacy-preserving conditions and support intelligent security perception tasks with higher accuracy, stronger robustness, and improved interpretability.