Abstract In the automated driving community,accurately identifying distracted driving behaviors is crucial for ensuring driving safety and advancing vehicle automation levels. This paper proposes a lightweight distracted driving behavior recognition model based on the KRCA knowledge distillation framework. The ConvNeXt-SqueezeNet LW lightweight teacher and student architecture,which integrates four types of knowledge including hard labels,soft labels,sample relationships,and interlayer features,significantly improves recognition accuracy and demonstrates enhanced robustness in low-light and occlusion scenarios. Experiments on the ASU and SFD3 datasets demonstrate that the optimized lightweight student network SqueezeNet LW achieves accuracies of 99.83% and 99.52%,respectively,while substantially reducing model parameters (187.43K) and computational cost (2.543M FLOPs). This research provides a viable solution for real-time distraction detection in automotive systems,bridging the gap between lightweight model constraints and high-accuracy requirements. Our model is available at: https://github.com/lgf1329/Enhancing-Lightweight-Distracted-Driving-Behavior-Recognition-via-KRCA-KnowledgeDistillation.git.

BackgroundWith the growing demand for dementia-friendly care environments, existing smart room designs often lack systematic methods to translate user needs into functional solutions.ObjectiveThis study proposes an integrated Kano Model, Analytic Hierarchy Process, and Quality Function Deployment (Kano-AHP-QFD) framework to optimize the design of smart living rooms for dementia patients in nursing homes.MethodThrough literature reviews and user interviews, 22 user requirements were identified and categorized using the Kano model. The AHP was then employed to prioritize these requirements, with "behavior recognition (e.g., falls, wandering)" emerging as the top priority, assigned a weight of 0.3622. Subsequently, the QFD method translated these weighted requirements into design functions via the House of Quality, resulting in a set of optimized smart living room designs.ResultsThe study demonstrates that the integration of Kano-AHP-QFD provides a structured and data-driven approach to systematically address the complex needs of dementia patients, enhancing the scientific rigor and practicality of smart room design. User satisfaction scores improved from 61.655 to 80.663 after implementing the optimized smart living room designs.ConclusionsThe proposed framework offers valuable insights for designers, care providers, and policymakers aiming to improve the quality of life for elderly individuals with dementia. It is also applicable to various cognitive impairment care scenarios such as rehabilitation centers and assisted living facilities, and can provide scientific references for the environmental design of other special user groups.

Preserved social behavior recognition in patients with epilepsy.

Problematic behaviors can occur without distress: A person-centered analysis of behavioral, substance-related, and mental health indicators.

DB-SBR: A dual-backbone model for student behavior recognition

MVim: A high-accuracy, lightweight neural network for real-time driver behavior recognition

Multimodal driver behavior recognition based on frame-adaptive convolution and feature fusion

Behavior recognition of spatial non-cooperative targets via geometric feature extraction: A random forest approach with deep neural network comparison

(1) Background: With the continuous development of intelligent education, classroom behavior recognition has become increasingly important in teaching evaluation and learning analytics. In response to challenges such as occlusion, scale differences, and fine-grained behavior recognition in complex classroom environments, this paper proposes an improved YOLOv11-ASV detection framework; (2) Methods: This framework introduces the Adaptive Spatial Pyramid Network (ASPN) based on YOLOv11, enhancing contextual modeling capabilities through block-level channel partitioning and multi-scale feature fusion mechanisms. Additionally, VanillaNet is adopted as the backbone network to improve the global semantic feature representation; (3) Conclusions: Experimental results show that on our self-built classroom behavior dataset (ClassroomDatasets), YOLOv11-ASV achieves 81.5% mAP50 and 62.1% mAP50–95, improving by 1.6% and 2.9%, respectively, compared to the baseline model. Notably, performance shows significant improvement in recognizing behavior classes such as “reading” and “writing” which are often confused. The experimental results validate the effectiveness of the YOLOv11-ASV model in improving behavior recognition accuracy and robustness in complex classroom scenarios, providing reliable technical support for the practical application of smart classroom systems.

ABSTRACT The precise detection of low‐amplitude and spatially variant biomechanical signals continues to present a substantial technological challenge. Ferroelectrets, with their exceptional piezoelectric coefficients and dynamic response characteristics, represent a compelling alternative for developing high‐sensitivity pressure sensors and deformable energy harvesting systems. Herein, we present a highly sensitive poly(vinylidene fluoride) (PVDF) ferroelectret film featuring a cross‐scale pore structure fabricated using organic dimethylhexanediol (DMHD) crystals as sacrificial templates, sandwiched between two PVDF layers, thus forming a piezoelectric‐ferroelectret‐ piezoelectric (PFP) three‐layer thin film. The engineered porous architecture enables the substantial net charge storage and creates an oriented space charge network, where subtle mechanical loading (<1 N) induces pronounced charge displacements that generate strong local electric field variations. As a result, the optimized PEP devices exhibit a superior piezoelectric coefficient of 650 pC·N −1 , and high sensitivity of 595.85 mV·kPa −1 . As a proof‐of‐concept, the PFP device is seamlessly integrated into a facial mask, facilitating accurate recognition of respiratory behaviors. With the assistance of both a Convolutional Neural Network (CNN) and a Bidirectional Long Short‐Term Memory network (BiLSTM), a PEP‐based smart mask can recognize respiratory tracts and multiple breathing patterns with a classification accuracy of up to 100%. This study pioneers a high‐efficacy respiratory monitoring device via ferroelectret ultra‐sensitivity, showing transformative prospects for clinical diagnostics and daily healthcare implementation.

This study proposes a novel Multi Curvelet Transformer Network (MCTN) for fine-grained human behavior recognition in dynamic video scenarios. A key challenge in this field lies in accurately identifying human actions under adverse conditions such as motion blur, occlusion, and varying illumination. To address this, we introduce a motion blur restoration module leveraging the curvelet transform to enhance motion image clarity, thereby improving downstream behavior detection. Furthermore, we enhance the Transformer architecture by embedding curvelet-based multi-scale attention mechanisms, which significantly improve the model's ability to extract spatial-temporal features at different resolutions. The proposed network also adopts a multi-curvelet transform structure to deepen semantic representation. Experimental results on benchmark datasets, including an action recognition dataset and the MSCOCO dataset, demonstrate that MCTN achieves superior performance, reaching a mean average precision (mAP) of 0.822. These results underscore the potential of MCTN in real-time intelligent video analysis and human-computer interaction applications.

In the breeding scene, limited by the small number of samples and environmental interference such as illumination occlusion, sika deer behavior recognition still faces challenges such as insufficient feature representation and weak cross-scale modeling ability. To this end, this study builds a lightweight improved model SDB-YOLO based on YOLOv11n. Firstly, the FPSC module is proposed to enhance the correlation between multi-scale features through the shared convolution mechanism, so as to significantly improve the quality of feature fusion under the condition of small samples. Secondly, the Ghost feature generation and dynamic convolution strategy are introduced into the C3k2 module to construct the C3_GDConv structure, so as to strengthen the fine-grained behavior pattern modeling ability and reduce redundant calculations. In addition, the CBAM attention mechanism is added to the neck of the network to further improve the ability of key information extraction and enhance the discrimination of feature expression. Finally, the EfficientHead was used to replace the original detection head to obtain a more robust training process and higher detection accuracy in small-sample scenarios. Experimental results show that SDB-YOLO achieves 90.2% detection accuracy with only 4.3 GFLOPs of calculation, which achieves significant performance improvement compared with YOLOv11n, and verifies the effectiveness and lightweight advantages of the proposed method in small-sample special animal behavior recognition tasks.

The integration of large-scale renewable energy sources has increased the complexity of operation and maintenance in modern power systems, causing on-site substation operation and maintenance activities to exhibit stronger continuity and dynamics, and thereby placing higher demands on real-time operational perception and safety judgment. However, existing behavior recognition methods have difficulty accurately identifying operational states in complex scenarios involving continuous actions, partial occlusions, and fine-grained manipulations. To address these challenges, this paper proposes a safety behavior recognition method for substation operations based on a dual-path spatiotemporal network. Personnel localization is achieved using YOLOv8, while behavior classification is performed through the SlowFast framework. In the Slow pathway, an ECA attention mechanism is integrated with residual structures to enhance the representation of sustained operational postures. In the Fast pathway, a multi-path excitation residual network is introduced to fuse temporal, channel, and motion information, improving the multi-scale representation of local action variations. Furthermore, to mitigate the issue of class imbalance in substation operation data, Focal Loss based on binary cross-entropy is incorporated to adaptively down-weight easily classified samples. Experimental results demonstrate that the proposed method achieves a recognition accuracy of 87.77% and an F1-score of 85.56% across multiple operation scenarios. The results further indicate improved recognition stability and adaptability, supporting safe substation operation and maintenance in renewable energy-integrated power systems.

Passenger-gathering behavior often triggers safety incidents such as stampedes due to overcrowding, posing significant challenges to public order maintenance and passenger safety. Traditional early warning algorithms for passenger-gathering behavior typically perform only global modeling of image appearance, neglecting the analysis of individual passenger actions in practical 3D physical space, leading to high false-alarm and missed-alarm rates. To address this issue, we decompose the modeling process into two stages: human pose estimation and gathering behavior recognition. Specifically, the pose of each individual in 3D space is first estimated from images, and then fused with global features to complete the early warning. This work focuses on the former stage and aims to develop an accurate and efficient human pose estimation model capable of real-time inference on resource-constrained devices. To this end, we propose a 3D human pose estimation framework that integrates a hybrid spatio-temporal Transformer with three collaborative agents. First, a reinforcement learning-based architecture search agent is designed to adaptively select among Global Self-Attention, Window Attention, and External Attention for each block to optimize the model structure. Second, a feedback optimization agent is developed to dynamically adjust the search process, balancing exploration and convergence. Third, a quantization agent is employed that leverages quantization-aware training (QAT) to generate an INT8 deployment-ready model with minimal loss in accuracy. Experiments conducted on the Human3.6M dataset demonstrate that the proposed method achieves a mean per joint position error (MPJPE) of 42.15 mm with only 4.38 M parameters and 19.39 GFLOPs under FP32 precision, indicating substantial potential for subsequent gathering behavior recognition tasks.

With the rapid advancement of 5G communication and computing technologies, wireless sensing has demonstrated extensive application prospects in human vital sign monitoring. This article systematically reviews noncontact vital sign monitoring technologies based on Wi-Fi and millimeter-wave radar, covering their fundamental principles, typical applications, and key challenges. First, the fundamental mechanisms of wireless sensing are introduced, including the use of Channel State Information (CSI) in multipath environments and the micro-Doppler effect in Frequency Modulated Continuous Wave (FMCW) radar. Then, key technologies and developments in typical applications—such as driver behavior recognition, gesture interaction, and heart rate monitoring—are examined in detail. Key challenges in practical applications are summarized, including weak signal detection, clutter suppression, and model generalization. Finally, future development trends are discussed, including multimodal fusion, intelligent algorithm optimization, and device miniaturization. This review aims to provide technical references and insights for future development to researchers in related fields.

Aquaculture monitoring increasingly relies on computer vision to evaluate fish behavior and welfare under farming conditions. This dataset was collected in a commercial recirculating aquaculture system (RAS) integrated with hydroponics in Queretaro, Mexico, to support the development of robust visual models for Nile tilapia (Oreochromis niloticus). More than ten hours of underwater recordings were curated into 31 clips of 30 s each, a duration selected to balance representativeness of fish activity with a manageable size for annotation and training. Videos were captured using commercial action cameras at multiple resolutions (1920 × 1080 to 5312 × 4648 px), frame rates (24–60 fps), depths, and lighting configurations, reproducing real-world challenges such as turbidity, suspended solids, and variable illumination. For each recording, physicochemical parameters were measured, including temperature, pH, dissolved oxygen and turbidity, and are provided in a structured CSV file. In addition to the raw videos, the dataset includes 3520 extracted frames annotated using a polygon-based JSON format, enabling direct use for training object detection and behavior recognition models. This dual resource of unprocessed clips and annotated images enhances reproducibility, benchmarking, and comparative studies. By combining synchronized environmental data with annotated underwater imagery, the dataset contributes a non-invasive and versatile resource for advancing aquaculture monitoring through computer vision.

This study proposes an intelligent surveillance framework that integrates image preprocessing, illuminance-adaptive object detection, multi-object tracking, and pedestrian abnormal behavior recognition to address the rapid degradation of image recognition performance under low-illuminance street lighting conditions. In the preprocessing stage, image quality was enhanced by correcting color distortion and contour loss, while in the detection stage, illuminance-based loss weighting was applied to maintain high detection sensitivity even in dark environments. During the tracking process, a Kalman filter was employed to ensure inter-frame consistency of detected objects. In the abnormal behavior recognition stage, temporal motion patterns were analyzed to detect events such as falls and prolonged inactivity in real time. The experimental results indicate that the proposed method maintained an average detection accuracy of approximately 0.9 and adequate tracking performance in the 80% range under low-illuminance conditions, while also exhibiting stable recognition rates across various weather environments. Although slight performance degradation was observed under dense fog or highly crowded scenes, such limitations are expected to be mitigated through sensor fusion and enhanced processing efficiency. These findings experimentally demonstrate the technical feasibility of a real-time intelligent recognition system for nighttime street lighting environments.

Deep learning-based animal activity recognition (AAR) achieves promising performance but remains constrained by its reliance on large labeled datasets. While pre-training offers a viable path toward reducing annotation dependency, existing approaches like transfer learning primarily utilize single-species labeled data, neglecting the substantial publicly available unlabeled data across species. To address this gap, this study introduces a contrastive learning based self-supervised learning framework that effectively leverages cross-species unlabeled data to mitigate annotation scarcity in AAR. The approach involves two stages: (1) Self-supervised pre-training of a patch time series Transformer (PatchTST) based encoder using cross-species unlabeled data with a time-frequency consistency objective to learn transferable motion representations; and (2) Fine-tuning on species-specific labeled data, where the pre-trained encoder serves as a feature extraction backbone in a novel classification model that integrates intra-axis local dynamics via PatchTST-based encoder and cross-axis global patterns via depth-wise separable convolution. Experimental results demonstrate that our approach significantly enhances recognition performance under limited labeled data, achieving improvements of 4.79% in accuracy and 4.57% in F1-score over a baseline trained from scratch, while also improving discrimination of semantically similar behaviors and maintaining robustness with reduced samples. This work establishes a new direction for scalable and label-efficient animal behavior monitoring.

Primary lung cancer encompasses heterogeneous histopathological subtypes with distinct clinical and radiological features. Accurate correlation of CT patterns with histology can improve diagnostic assessment and contribute to earlier recognition of aggressive tumor behavior. A retrospective study was conducted on 33 patients (25 males, 8 females; mean age 67.3 years) with histologically confirmed primary lung cancer. Demographic factors, smoking history, histological subtype, and CT characteristics—including lesion size, lobar and segmental distribution, necrosis, and metastatic spread—were analyzed. Radiological staging was performed according to the TNM system, and findings were correlated with histopathology. Results: Histology: Adenocarcinoma was the most frequent subtype (51.5%), followed by squamous cell carcinoma (27.3%), with fewer cases of large cell, small cell, mixed, and plano-cellular carcinomas. Adenocarcinoma predominated in both smokers and non-smokers, whereas squamous carcinoma was more frequent among smokers. Tumor size and location: Mean lesion diameter was 5.9 cm (range: 1.5–13 cm). The most common sites were the right lower lobe (33.3%) and right upper lobe (30.3%), particularly in posterior basal and posterior segments. Necrosis: Present in 54.5% of tumors, necrosis was more frequent in squamous carcinomas (44.4%) compared with adenocarcinomas (33.3%), while adenocarcinomas were more often necrosis-free (75%). Metastases: Distant spread occurred in 27.3% of patients, predominantly to adrenal glands (44.4%) and brain (22.2%), with additional involvement of liver (11.1%) and lymph nodes/contralateral lung (55.6%). Adenocarcinoma accounted for two-thirds of metastatic cases. Staging: Advanced disease predominated, with stage IIIB (27.3%) and stage IVA (18.2%) most frequent. Only 24.2% of patients were diagnosed at early operable stages (IA–IIB). CT–pathology correlation demonstrated that adenocarcinoma is the predominant subtype across smoking categories, whereas squamous carcinoma more frequently exhibited necrosis. Most patients presented with large tumors and advanced disease, with metastases reflecting typical spread patterns. These findings underscore the diagnostic value of CT in characterizing histological subtypes, staging, and guiding clinical management, while reinforcing the need for earlier detection and screening strategies.

Two-Way Classroom Interaction Analysis via a Coupled ConvNeXt–Multimodal Transformer for Fine-grained Behavior Recognition