Sensors

15 pages, 1279 KiB

Open AccessArticle

A Stability- and Aggregation-Based Method for Heart Rate Estimation Using Photoplethysmographic Signals During Physical Activity

by Sabrina C. Crepaldi, Jiabin Wang, Fumiya Matsumoto, Hiroki Takeuchi, Tatsuhiko Watanabe and Yoshiharu Yamamoto

Sensors 2025, 25(14), 4315; https://doi.org/10.3390/s25144315 (registering DOI) - 10 Jul 2025

Abstract

In recent years, the use of photoplethysmography (PPG)-based heart rate detection has gained considerable attention as a cost-effective alternative to conventional electrocardiography (ECG) for applications in healthcare and fitness tracking. Although deep learning methods have shown promise in heart rate estimation and motion [...] Read more.

In recent years, the use of photoplethysmography (PPG)-based heart rate detection has gained considerable attention as a cost-effective alternative to conventional electrocardiography (ECG) for applications in healthcare and fitness tracking. Although deep learning methods have shown promise in heart rate estimation and motion artifact removal from PPG signals recorded during physical activity, their computational requirements and need for extensive training data make them less practical for real-world conditions when ground truth data is unavailable for calibration. This study presents a one-size-fits-all approach for heart rate estimation during physical activity that employs aggregation-based techniques to track heart rate and minimize the effects of motion artifacts, without relying on complex machine learning or deep learning techniques. We evaluate our method on four publicly available datasets—PPG-DaLiA, WESAD, IEEE_Training, and IEEE_Test, all recorded using wrist-worn devices—along with a new dataset, UTOKYO, which includes PPG and accelerometer data collected from a smart ring. The proposed method outperforms the CNN ensemble model for the PPG-DaLiA dataset and the IEEE_Test dataset and reduces the mean absolute error (MAE) by 1.45 bpm and 5.71 bpm, respectively, demonstrating that effective signal processing techniques can match the performance of more complex deep learning models without requiring extensive computational resources or dataset-specific tuning. Full article

(This article belongs to the Special Issue Feature Papers in Biomedical Sensors 2025)

► Show Figures

Figure 1

15 pages, 3836 KiB

Open AccessArticle

Porous-Cladding Polydimethylsiloxane Optical Waveguide for Biomedical Pressure Sensing Applications

by Koffi Novignon Amouzou, Alberto Alonso Romero, Dipankar Sengupta, Camila Aparecida Zimmermann, Aashutosh Kumar, Normand Gravel, Jean-Marc Lina, Xavier Daxhelet and Bora Ung

Sensors 2025, 25(14), 4311; https://doi.org/10.3390/s25144311 (registering DOI) - 10 Jul 2025

Abstract

We report a new concept of a pressure sensor fully made from polydimethylsiloxane with a solid core and porous cladding that operates through (frustrated) total internal reflection. A flexible and sensitive rectangular cross-section waveguide was fabricated via the casting and molding method. The [...] Read more.

We report a new concept of a pressure sensor fully made from polydimethylsiloxane with a solid core and porous cladding that operates through (frustrated) total internal reflection. A flexible and sensitive rectangular cross-section waveguide was fabricated via the casting and molding method. The waveguide’s optical losses can be temperature-controlled during the fabrication process by controlling the quantity of microbubbles incorporated (2% approximately for samples made at 70 °C). By controlling the precuring temperature, the microbubbles are incorporated into the waveguides during the simple and cost-effective fabrication process through the casting and molding method. For these samples, we measured good optical loss tradeoff of the order of 1.85 dB/cm, which means that it is possible to fabricate a solid-core/clad waveguide with porous cladding able to guide light properly. We demonstrated the microbubble concentration control in the waveguide, and we measured an average diameter of 239 ± 16 µm. A sensitivity to pressure of 0.1035 dB/kPa optical power loss was measured. The results show that in a biomedical dynamic pressure range (0 to 13.3 kPa), this new device indicates the critical pressure threshold level, which constitutes a crucial asset for potential applications such as pressure injury prevention. Full article

(This article belongs to the Special Issue Optical-Based Sensors and Sensing Technologies for Biomedical Applications)

► Show Figures

Graphical abstract

25 pages, 1272 KiB

Open AccessArticle

Complex Environmental Geomagnetic Matching-Assisted Navigation Algorithm Based on Improved Extreme Learning Machine

by Jian Huang, Zhe Hu and Wenjun Yi

Sensors 2025, 25(14), 4310; https://doi.org/10.3390/s25144310 (registering DOI) - 10 Jul 2025

Abstract

In complex environments where satellite signals may be interfered with, it is difficult to achieve precise positioning of high-speed aerial vehicles solely through the inertial navigation system. To overcome this challenge, this paper proposes an NGO-ELM geomagnetic matching-assisted navigation algorithm, in which the [...] Read more.

In complex environments where satellite signals may be interfered with, it is difficult to achieve precise positioning of high-speed aerial vehicles solely through the inertial navigation system. To overcome this challenge, this paper proposes an NGO-ELM geomagnetic matching-assisted navigation algorithm, in which the Northern Goshawk Optimization (NGO) algorithm is used to optimize the initial weights and biases of the Extreme Learning Machine (ELM). To enhance the matching performance of the NGO-ELM algorithm, three improvements are proposed to the NGO algorithm. The effectiveness of these improvements is validated using the CEC2005 benchmark function suite. Additionally, the IGRF-13 model is utilized to generate a geomagnetic matching dataset, followed by comparative testing of five geomagnetic matching models: INGO-ELM, NGO-ELM, ELM, INGO-XGBoost, and INGO-BP. The simulation results show that after the airborne equipment acquires the geomagnetic data, it only takes 0.27 µs to obtain the latitude, longitude, and altitude of the aerial vehicle through the INGO-ELM model. After unit conversion, the average absolute errors are approximately 6.38 m, 6.43 m, and 0.0137 m, respectively, which significantly outperform the results of four other models. Furthermore, when noise is introduced into the test set inputs, the positioning error of the INGO-ELM model remains within the same order of magnitude as those before the noise was added, indicating that the model exhibits excellent robustness. It has been verified that the geomagnetic matching-assisted navigation algorithm proposed in this paper can achieve real-time, accurate, and stable positioning, even in the presence of observational errors from the magnetic sensor. Full article

(This article belongs to the Section Navigation and Positioning)

► Show Figures

Figure 1

23 pages, 9575 KiB

Open AccessArticle

Infrared and Visible Image Fusion via Residual Interactive Transformer and Cross-Attention Fusion

by Liquan Zhao, Chen Ke, Yanfei Jia, Cong Xu and Zhijun Teng

Sensors 2025, 25(14), 4307; https://doi.org/10.3390/s25144307 (registering DOI) - 10 Jul 2025

Abstract

Infrared and visible image fusion combines infrared and visible images of the same scene to produce a more informative and comprehensive fused image. Existing deep learning-based fusion methods fail to establish dependencies between global and local information during feature extraction. This results in [...] Read more.

Infrared and visible image fusion combines infrared and visible images of the same scene to produce a more informative and comprehensive fused image. Existing deep learning-based fusion methods fail to establish dependencies between global and local information during feature extraction. This results in unclear scene texture details and low contrast of the infrared thermal targets in the fused image. This paper proposes an infrared and visible image fusion network to address this issue via the use of a residual interactive transformer and cross-attention fusion. The network first introduces a residual dense module to extract shallow features from the input infrared and visible images. Next, the residual interactive transformer extracts global and local features from the source images and establishes interactions between them. Two identical residual interactive transformers are used for further feature extraction. A cross-attention fusion module is also designed to fuse the infrared and visible feature maps extracted by the residual interactive transformer. Finally, an image reconstruction network generates the fused image. The proposed method is evaluated on the RoadScene, TNO, and M3FD datasets. The experimental results show that the fused images produced by the proposed method contain more visible texture details and infrared thermal information. Compared to nine other methods, the proposed approach achieves superior fusion performance. Full article

(This article belongs to the Section Sensing and Imaging)

► Show Figures

Figure 1

22 pages, 590 KiB

Open AccessReview

ROS-Based Navigation and Obstacle Avoidance: A Study of Architectures, Methods, and Trends

by Zhe Wei, Sen Wang, Kangyelin Chen and Fang Wang

Sensors 2025, 25(14), 4306; https://doi.org/10.3390/s25144306 (registering DOI) - 10 Jul 2025

Abstract

With the widespread adoption of the Robot Operating System (ROS), technologies for autonomous navigation in mobile robots have advanced considerably. ROS provides a modular navigation stack that integrates essential components, such as SLAM, localisation, global path planning, and obstacle avoidance, forming the foundation [...] Read more.

With the widespread adoption of the Robot Operating System (ROS), technologies for autonomous navigation in mobile robots have advanced considerably. ROS provides a modular navigation stack that integrates essential components, such as SLAM, localisation, global path planning, and obstacle avoidance, forming the foundation for applications including service robotics and autonomous driving. Nonetheless, achieving safe and reliable navigation in complex and dynamic environments remains a formidable challenge, due to the need for real-time perception of moving obstacles, sensor fusion requirements, and the demand for robust and efficient algorithms. This study presents a systematic examination of the ROS-based navigation stack and obstacle-avoidance mechanisms. The architecture and implementation principles of the core modules are analysed, along with a comparison of the features and application suitability of common local planners such as the Dynamic Window Approach (DWA) and Timed Elastic Band (TEB). The key technical challenges in autonomous navigation are summarised, and recent advancements are reviewed to outline emerging trends in ROS-based systems, including integration with deep learning, multi-robot coordination, and real-time optimisation. The findings contribute to a deeper theoretical understanding of robotic navigation and offer practical guidance for the design and development of autonomous systems. Full article

(This article belongs to the Special Issue Innovative Synergies: Robotics, AI, and Sensor Technologies in Field Autonomous Systems)

► Show Figures

Figure 1

19 pages, 1186 KiB

Open AccessArticle

Synthetic Patient–Physician Conversations Simulated by Large Language Models: A Multi-Dimensional Evaluation

by Syed Ali Haider, Srinivasagam Prabha, Cesar Abraham Gomez-Cabello, Sahar Borna, Ariana Genovese, Maissa Trabilsy, Bernardo G. Collaco, Nadia G. Wood, Sanjay Bagaria, Cui Tao and Antonio Jorge Forte

Sensors 2025, 25(14), 4305; https://doi.org/10.3390/s25144305 (registering DOI) - 10 Jul 2025

Abstract

Background: Data accessibility remains a significant barrier in healthcare AI due to privacy constraints and logistical challenges. Synthetic data, which mimics real patient information while remaining both realistic and non-identifiable, offers a promising solution. Large Language Models (LLMs) create new opportunities to generate [...] Read more.

Background: Data accessibility remains a significant barrier in healthcare AI due to privacy constraints and logistical challenges. Synthetic data, which mimics real patient information while remaining both realistic and non-identifiable, offers a promising solution. Large Language Models (LLMs) create new opportunities to generate high-fidelity clinical conversations between patients and physicians. However, the value of this synthetic data depends on careful evaluation of its realism, accuracy, and practical relevance. Objective: To assess the performance of four leading LLMs: ChatGPT 4.5, ChatGPT 4o, Claude 3.7 Sonnet, and Gemini Pro 2.5 in generating synthetic transcripts of patient–physician interactions in plastic surgery scenarios. Methods: Each model generated transcripts for ten plastic surgery scenarios. Transcripts were independently evaluated by three clinically trained raters using a seven-criterion rubric: Medical Accuracy, Realism, Persona Consistency, Fidelity, Empathy, Relevancy, and Usability. Raters were blinded to the model identity to reduce bias. Each was rated on a 5-point Likert scale, yielding 840 total evaluations. Descriptive statistics were computed, and a two-way repeated measures ANOVA was used to test for differences across models and metrics. In addition, transcripts were analyzed using automated linguistic and content-based metrics. Results: All models achieved strong performance, with mean ratings exceeding 4.5 across all criteria. Gemini 2.5 Pro received mean scores (5.00 ± 0.00) in Medical Accuracy, Realism, Persona Consistency, Relevancy, and Usability. Claude 3.7 Sonnet matched the scores in Persona Consistency and Relevancy and led in Empathy (4.96 ± 0.18). ChatGPT 4.5 also achieved perfect scores in Relevancy, with high scores in Empathy (4.93 ± 0.25) and Usability (4.96 ± 0.18). ChatGPT 4o demonstrated consistently strong but slightly lower performance across most dimensions. ANOVA revealed no statistically significant differences across models (F(3, 6) = 0.85, p = 0.52). Automated analysis showed substantial variation in transcript length, style, and content richness: Gemini 2.5 Pro generated the longest and most emotionally expressive dialogues, while ChatGPT 4o produced the shortest and most concise outputs. Conclusions: Leading LLMs can generate medically accurate, emotionally appropriate synthetic dialogues suitable for educational and research use. Despite high performance, demographic homogeneity in generated patients highlights the need for improved diversity and bias mitigation in model outputs. These findings support the cautious, context-aware integration of LLM-generated dialogues into medical training, simulation, and research. Full article

(This article belongs to the Special Issue Feature Papers in Smart Sensing and Intelligent Sensors 2025)

► Show Figures

Figure 1

16 pages, 377 KiB

Open AccessArticle

Towards Improved Eye Movement Biometrics: Investigating New Features with Neural Networks

by Katarzyna Harezlak and Ewa Pluciennik

Sensors 2025, 25(14), 4304; https://doi.org/10.3390/s25144304 (registering DOI) - 10 Jul 2025

Abstract

Providing protected access to many everyday-used resources is becoming increasingly necessary. Research on applying eye movement for this purpose has been conducted for many years. However, due to technological advancements and the lack of stable solutions, subsequent explorations remain valid. The presented work [...] Read more.

Providing protected access to many everyday-used resources is becoming increasingly necessary. Research on applying eye movement for this purpose has been conducted for many years. However, due to technological advancements and the lack of stable solutions, subsequent explorations remain valid. The presented work is one of such studies. Two methods of biometric identification based on eye movements that utilize neural networks have been developed. In the first case, a feature vector was constructed from a 100-element time series depicting eye movement dynamics, which included velocity, acceleration, jerk, their point-to-point percentage changes, and frequency-domain representations. The same eye movement dynamic features were used in the second method, but this time, statistical values were calculated based on the previously defined time series. Long Short-Term Memory (LSTM) and dense networks were used in the user identification task in the first and second approaches, respectively. In the exploration, the publicly available GazeBase dataset was used, from which data collected for the ‘jumping point’ stimulus were chosen. The obtained results are very promising, with an accuracy of 96% for the LSTM model and the time series feature vector set and 76% for the second method. They were achieved over a three-year time span of eye movement recordings; however, different time periods were investigated, as well as various numbers of stimulus positions. Full article

(This article belongs to the Special Issue New Trends in Biometric Sensing and Information Processing)

► Show Figures

Figure 1

18 pages, 3088 KiB

Open AccessArticle

Incremental Multi-Step Learning MLP Model for Online Soft Sensor Modeling

by Yihan Wang, Jiahao Tao and Liang Zhao

Sensors 2025, 25(14), 4303; https://doi.org/10.3390/s25144303 (registering DOI) - 10 Jul 2025

Abstract

Industrial production often involves complex time-varying operating conditions that result in continuous time-series production data. The traditional soft sensor approach has difficulty adjusting to such dynamic changes, which makes model performance less optimal. Furthermore, online analytical systems have significant operational and maintenance costs [...] Read more.

Industrial production often involves complex time-varying operating conditions that result in continuous time-series production data. The traditional soft sensor approach has difficulty adjusting to such dynamic changes, which makes model performance less optimal. Furthermore, online analytical systems have significant operational and maintenance costs and entail a substantial delay in measurement output, limiting their ability to provide real-time control. In order to deal with these challenges, this paper introduces a multivariate multi-step predictive multilayer perceptron regression soft-sensing model, referred to as incremental MVMS-MLP. This model incorporates incremental learning strategies to enhance its adaptability and accuracy in multivariate predictions. As part of the method, a pre-trained MVMS-MLP model is developed, which integrates multivariate multi-step prediction with MLP regression to handle temporal data. Through the use of incremental learning, an incremental MVMS-MLP model is constructed from this pre-trained model. The effectiveness of the proposed method is demonstrated by benchmark problems and real-world industrial case studies. Full article

(This article belongs to the Section Industrial Sensors)

► Show Figures

Figure 1

15 pages, 2750 KiB

Open AccessArticle

Gait Environment Recognition Using Biomechanical and Physiological Signals with Feed-Forward Neural Network: A Pilot Study

by Kyeong-Jun Seo, Jinwon Lee, Ji-Eun Cho, Hogene Kim and Jung Hwan Kim

Sensors 2025, 25(14), 4302; https://doi.org/10.3390/s25144302 (registering DOI) - 10 Jul 2025

Abstract

Gait, the fundamental form of human locomotion, occurs across diverse environments. The technology for recognizing environmental changes during walking is crucial for preventing falls and controlling wearable robots. This study collected gait data on level ground (LG), ramps, and stairs using a feed-forward [...] Read more.

Gait, the fundamental form of human locomotion, occurs across diverse environments. The technology for recognizing environmental changes during walking is crucial for preventing falls and controlling wearable robots. This study collected gait data on level ground (LG), ramps, and stairs using a feed-forward neural network (FFNN) to classify the corresponding gait environments. Gait experiments were performed on five non-disabled participants using an inertial measurement unit, a galvanic skin response sensor, and a smart insole. The collected data were preprocessed through time synchronization and filtering, then labeled according to the gait environment, yielding 47,033 data samples. Gait data were used to train an FFNN model with a single hidden layer, achieving a high accuracy of 98%, with the highest accuracy observed on LG. This study confirms the effectiveness of classifying gait environments based on signals acquired from various wearable sensors during walking. In the future, these research findings may serve as basic data for exoskeleton robot control and gait analysis. Full article

(This article belongs to the Special Issue Wearable Sensing Technologies for Human Health Monitoring)

► Show Figures

Figure 1

18 pages, 1549 KiB

Open AccessArticle

Characteristics of Post-Exercise Lower Limb Muscle Tremor Among Speed Skaters

by Szymon Kuliś, Przemysław Pietraszewski and Bianca Callegari

Sensors 2025, 25(14), 4301; https://doi.org/10.3390/s25144301 (registering DOI) - 10 Jul 2025

Abstract

Physiological tremor analysis is a practical tool for assessing the neuromuscular impacts of sport-specific training. The purpose of this study was to examine and compare the physiological characteristics of lower limb resting postural tremor in athletes from Poland’s national speed skating team, following [...] Read more.

Physiological tremor analysis is a practical tool for assessing the neuromuscular impacts of sport-specific training. The purpose of this study was to examine and compare the physiological characteristics of lower limb resting postural tremor in athletes from Poland’s national speed skating team, following both sprint and endurance workouts. The study included 19 male, well-trained, elite athletes (with a mean age of 18 ± 3.1 years, body mass of 71.4 ± 10.1 kg, height of 178.5 ± 9.0 cm, and training experience of 12.6 ± 2.8 years) and a control group of 19 physically active but non-athlete men (with a mean age of 19 ± 2.3 years, body mass of 78.9 ± 12.1 kg, and height of 181.5 ± 11.0 cm). This group was assessed under resting conditions to provide baseline reference values for physiological tremor and to evaluate whether the neuromuscular tremor response is specific to trained athletes. Tremor amplitude and frequency were measured using an accelerometer, with data log-transformed to normalize the power spectrum distribution. Key findings indicate a significant effect of training condition on tremor amplitude in the low-frequency range (L(2_5); F_(1,18) = 38.42; p < 0.0001; η_p² = 0.68) and high-frequency range (L(9_14); F_(1,36) = 19.19; p < 0.0001; η_p² = 0.51). Post hoc analysis showed that tremor amplitude increased significantly after both sprint (p < 0.001) and endurance training (p < 0.001) compared to rest. No significant differences were observed between sprint and endurance training conditions for L(2_5) (p = 0.1014), but sprint training resulted in a greater increase in tremor in the high-frequency range (L(9_14); p < 0.0001). Tremor frequency (F(2_5) and F(9_14)) also increased significantly post-training. Notably, no differences were observed between limbs, indicating symmetrical neuromuscular adaptation. These findings highlight the utility of tremor analysis in monitoring neuromuscular fatigue and performance in speed skaters. Future research should explore the application of this method in broader athletic populations and evaluate its potential integration into training programs. Full article

(This article belongs to the Section Wearables)

► Show Figures

Figure 1

18 pages, 736 KiB

Open AccessArticle

Collaborative Split Learning-Based Dynamic Bandwidth Allocation for 6G-Grade TDM-PON Systems

by Alaelddin F. Y. Mohammed, Yazan M. Allawi, Eman M. Moneer and Lamia O. Widaa

Sensors 2025, 25(14), 4300; https://doi.org/10.3390/s25144300 (registering DOI) - 10 Jul 2025

Abstract

Dynamic Bandwidth Allocation (DBA) techniques enable Time Division Multiplexing Passive Optical Network (TDM-PON) systems to efficiently manage upstream bandwidth by allowing the centralized Optical Line Terminal (OLT) to coordinate resource allocation among distributed Optical Network Units (ONUs). Conventional DBA techniques struggle to adapt [...] Read more.

Dynamic Bandwidth Allocation (DBA) techniques enable Time Division Multiplexing Passive Optical Network (TDM-PON) systems to efficiently manage upstream bandwidth by allowing the centralized Optical Line Terminal (OLT) to coordinate resource allocation among distributed Optical Network Units (ONUs). Conventional DBA techniques struggle to adapt to dynamic traffic conditions, resulting in suboptimal performance under varying load scenarios. This work suggests a Collaborative Split Learning-Based DBA (CSL-DBA) framework that utilizes the recently emerging Split Learning (SL) technique between the OLT and ONUs for the objective of optimizing predictive traffic adaptation and reducing communication overhead. Instead of requiring centralized learning at the OLT, the proposed approach decentralizes the process by enabling ONUs to perform local traffic analysis and transmit only model updates to the OLT. This cooperative strategy guarantees rapid responsiveness to fluctuating traffic conditions. We show by extensive simulations spanning several traffic scenarios, including low, fluctuating, and high traffic load conditions, that our proposed CSL-DBA achieves at least 99% traffic prediction accuracy, with minimal inference latency and scalable learning performance, and it reduces communication overhead by approximately 60% compared to traditional federated learning approaches, making it a strong candidate for next-generation 6G-grade TDM-PON systems. Full article

(This article belongs to the Special Issue Recent Advances in Optical Wireless Communications)

► Show Figures

Figure 1

15 pages, 3481 KiB

Open AccessArticle

Rolling Bearing Degradation Identification Method Based on Improved Monopulse Feature Extraction and 1D Dilated Residual Convolutional Neural Network

by Chang Liu, Haiyang Wu, Gang Cheng, Hui Zhou and Yusong Pang

Sensors 2025, 25(14), 4299; https://doi.org/10.3390/s25144299 (registering DOI) - 10 Jul 2025

Abstract

To address the challenges of extracting rolling bearing degradation information and the insufficient performance of conventional convolutional networks, this paper proposes a rolling bearing degradation state identification method based on the improved monopulse feature extraction and a one-dimensional dilated residual convolutional neural network [...] Read more.

To address the challenges of extracting rolling bearing degradation information and the insufficient performance of conventional convolutional networks, this paper proposes a rolling bearing degradation state identification method based on the improved monopulse feature extraction and a one-dimensional dilated residual convolutional neural network (1D-DRCNN). First, the fault pulse envelope waveform features are extracted through phase scanning and synchronous averaging, and a two-stage grid search strategy is employed to achieve FCC calibration. Subsequently, a 1D-DRCNN model is constructed to identify rolling bearing degradation states under different working conditions. The experimental study collects the vibration signals of nine degradation states, including the different sizes of inner and outer ring local faults as well as normal conditions, to comparatively analyze the proposed method’s rapid calibration capability and feature extraction quality. Furthermore, t-SNE visualization is utilized to analyze the network response to bearing degradation features. Finally, the degradation state identification performance across different network architectures is compared in pattern recognition experiments. The results show that the proposed improved feature extraction method significantly reduces the iterative calibration computational burden while effectively extracting local fault degradation information and overcoming complex working condition influence. The established 1D-DRCNN model integrates the advantages of dilated convolution and residual connections and can deeply mine sensitive features and accurately identify different bearing degradation states. The overall recognition accuracy can reach 97.33%. Full article

(This article belongs to the Section Fault Diagnosis & Sensors)

► Show Figures

Figure 1

20 pages, 1521 KiB

Open AccessArticle

Poisonous Plant Prediction Using Explainable Deep Inherent Learning Model

by Ahmed S. Maklad, Ashraf Alyanbaawi, Mohammed Farsi, Hani M. Ibrahim and Mahmoud Elmezain

Sensors 2025, 25(14), 4298; https://doi.org/10.3390/s25144298 (registering DOI) - 10 Jul 2025

Abstract

The increasing global discovery of plant species presents both opportunities and challenges, particularly in distinguishing between beneficial and poisonous varieties. While computer vision techniques show promise for classifying plant species and predicting toxicity, the lack of comprehensive datasets including images, scientific names, descriptions, [...] Read more.

The increasing global discovery of plant species presents both opportunities and challenges, particularly in distinguishing between beneficial and poisonous varieties. While computer vision techniques show promise for classifying plant species and predicting toxicity, the lack of comprehensive datasets including images, scientific names, descriptions, local names, and poisonous status complicates these predictions. In this paper, we propose an Explainable Deep Inherent Learning approach that leverages advanced computer vision techniques for effective plant species classification and poisonous status prediction. The proposed Deep Inherent Learning method was validated using different explanation techniques, and Explainable AI (XAI) was employed to clarify decision-making processes at both the local and global levels. Additionally, we provide visual information to enhance trust in the proposed method. To validate the efficacy of our approach, we present a case study involving 2500 images of 50 different plant species from the Arabian Peninsula, enriched with essential metadata. This research aims to reduce the incidence of poisoning from harmful plants, thereby benefiting individuals and society. Our experimental results demonstrate strong performance, with the XAI model achieving accuracy, Precision, Recall, and F1-Score of 0.94, 0.96, 0.96 and 0.97, respectively. By enhancing interpretability, our study fosters greater trust in AI-driven plant classification systems. Full article

(This article belongs to the Section Intelligent Sensors)

► Show Figures

Figure 1

12 pages, 1029 KiB

Open AccessArticle

Does tDCS Enhance Complex Motor Skill Acquisition? Evidence from a Golf-Putting Task

by Virginia Lopez-Alonso, Gabriel López-Bermúdez, Jeffrey Cayaban Pagaduan and Jose Andrés Sánchez-Molina

Sensors 2025, 25(14), 4297; https://doi.org/10.3390/s25144297 (registering DOI) - 10 Jul 2025

Abstract

Transcranial direct current stimulation (tDCS) modulates cortical excitability, thus inducing improvements in motor learning of simple tasks. In this study, we aimed to evaluate the effect of different tDCS conditions—anodal stimulation over the motor cortex (M1), anodal and cathodal stimulation over the prefrontal [...] Read more.

Transcranial direct current stimulation (tDCS) modulates cortical excitability, thus inducing improvements in motor learning of simple tasks. In this study, we aimed to evaluate the effect of different tDCS conditions—anodal stimulation over the motor cortex (M1), anodal and cathodal stimulation over the prefrontal cortex (PFC), and sham—on the online and offline learning of a complex accuracy task (golf-putting) in novice golfers. Methods: A total of 40 young, healthy subjects (24 men, 16 women) without previous golf experience were randomly distributed in four groups receiving sham, anodal M1, anodal PFC or cathodal PFC tDCS. All subjects participated in two consecutive sessions. In the first session, they performed 15 blocks of 10 golf-putting along with tDCS stimulation. After 24 h, they performed the same task without tDCS. Results: Repeated measures ANOVA revealed a significant improvement in performance during the two consecutive golf-putting sessions regardless of the site and the stimulation conditions. Conclusion: Our findings suggest that tDCS over M1 or PFC does not confer additional benefits in the acquisition of complex, full-body motor skills such as golf-putting. Full article

(This article belongs to the Special Issue Sensor-Based Human Motor Learning)

► Show Figures

Figure 1

32 pages, 16988 KiB

Open AccessArticle

From Photogrammetry to Virtual Reality: A Framework for Assessing Visual Fidelity in Structural Inspections

by Xiangxiong Kong, Terry F. Pettijohn II and Hovhannes Torikyan

Sensors 2025, 25(14), 4296; https://doi.org/10.3390/s25144296 (registering DOI) - 10 Jul 2025

Abstract

Civil structures carry significant service loads over long times but are prone to deterioration due to various natural impacts. Traditionally, these structures are inspected in situ by qualified engineers, a method that is high-cost, risky, time-consuming, and prone to error. Recently, researchers have [...] Read more.

Civil structures carry significant service loads over long times but are prone to deterioration due to various natural impacts. Traditionally, these structures are inspected in situ by qualified engineers, a method that is high-cost, risky, time-consuming, and prone to error. Recently, researchers have explored innovative practices by using virtual reality (VR) technologies as inspection platforms. Despite such efforts, a critical question remains: can VR models accurately reflect real-world structural conditions? This study presents a comprehensive framework for assessing the visual fidelity of VR models for structural inspection. To make it viable, we first introduce a novel workflow that integrates UAV-based photogrammetry, computer graphics, and web-based VR editing to establish interactive VR user interfaces. We then propose a visual fidelity assessment methodology that quantitatively evaluates the accuracy of the VR models through image alignment, histogram matching, and pixel-level deviation mapping between rendered images from the VR models and UAV-captured images under matched viewpoints. The proposed frameworks are validated using two case studies: a historic stone arch bridge and a campus steel building. Overall, this study contributes to the growing body of knowledge on VR-based structural inspections, providing a foundation for our peers for their further research in this field. Full article

(This article belongs to the Section Sensing and Imaging)

► Show Figures

Graphical abstract

27 pages, 27475 KiB

Open AccessArticle

LiGenCam: Reconstruction of Color Camera Images from Multimodal LiDAR Data for Autonomous Driving

by Minghao Xu, Yanlei Gu, Igor Goncharenko and Shunsuke Kamijo

Sensors 2025, 25(14), 4295; https://doi.org/10.3390/s25144295 (registering DOI) - 10 Jul 2025

Abstract

The automotive industry is advancing toward fully automated driving, where perception systems rely on complementary sensors such as LiDAR and cameras to interpret the vehicle’s surroundings. For Level 4 and higher vehicles, redundancy is vital to prevent safety-critical failures. One way to achieve [...] Read more.

The automotive industry is advancing toward fully automated driving, where perception systems rely on complementary sensors such as LiDAR and cameras to interpret the vehicle’s surroundings. For Level 4 and higher vehicles, redundancy is vital to prevent safety-critical failures. One way to achieve this is by using data from one sensor type to support another. While much research has focused on reconstructing LiDAR point cloud data using camera images, limited work has been conducted on the reverse process—reconstructing image data from LiDAR. This paper proposes a deep learning model, named LiDAR Generative Camera (LiGenCam), to fill this gap. The model reconstructs camera images by utilizing multimodal LiDAR data, including reflectance, ambient light, and range information. LiGenCam is developed based on the Generative Adversarial Network framework, incorporating pixel-wise loss and semantic segmentation loss to guide reconstruction, ensuring both pixel-level similarity and semantic coherence. Experiments on the DurLAR dataset demonstrate that multimodal LiDAR data enhances the realism and semantic consistency of reconstructed images, and adding segmentation loss further improves semantic consistency. Ablation studies confirm these findings. Full article

(This article belongs to the Special Issue Recent Advances in LiDAR Sensing Technology for Autonomous Vehicles)

► Show Figures

Figure 1

16 pages, 10934 KiB

Open AccessArticle

Visualization Monitoring and Safety Evaluation of Turnout Wheel–Rail Forces Based on BIM for Sustainable Railway Management

by Xinyi Dong, Yuelei He and Hongyao Lu

Sensors 2025, 25(14), 4294; https://doi.org/10.3390/s25144294 (registering DOI) - 10 Jul 2025

Abstract

With China’s high-speed rail network undergoing rapid expansion, turnouts constitute critical elements whose safety and stability are essential to railway operation. At present, the efficiency of wheel–rail force safety monitoring conducted in the small hours reserved for the construction and maintenance of operating [...] Read more.

With China’s high-speed rail network undergoing rapid expansion, turnouts constitute critical elements whose safety and stability are essential to railway operation. At present, the efficiency of wheel–rail force safety monitoring conducted in the small hours reserved for the construction and maintenance of operating lines without marking train operation lines is relatively low. To enhance the efficiency of turnout safety monitoring, in this study, a three-dimensional BIM model of the No. 42 turnout was established and a corresponding wheel–rail force monitoring scheme was devised. Collision detection for monitoring equipment placement and construction process simulation was conducted using Navisworks, such that the rationality of cable routing and the precision of construction sequence alignment were improved. A train wheel–rail force analysis program was developed in MATLAB R2022b to perform signal filtering, and static calibration was applied to calculate key safety evaluation indices—namely, the coefficient of derailment and the rate of wheel load reduction—which were subsequently analyzed. The safety of the No. 42 turnout and the effectiveness of the proposed monitoring scheme were validated, theoretical support was provided for train operational safety and turnout maintenance, and technical guidance was offered for whole-life-cycle management and green, sustainable development of railway infrastructure. Full article

(This article belongs to the Section Fault Diagnosis & Sensors)

► Show Figures

Figure 1

16 pages, 419 KiB

Open AccessArticle

Energy-Efficient Resource Allocation for Near-Field MIMO Communication Networks

by Tong Lin, Jianyue Zhu, Junfan Zhu, Yaqin Xie, Yao Xu and Xiao Chen

Sensors 2025, 25(14), 4293; https://doi.org/10.3390/s25144293 (registering DOI) - 10 Jul 2025

Abstract

With the rapid development of sixth-generation (6G) wireless networks and large-scale multiple-input multiple-output (MIMO) technology, the number of antennas deployed at base stations (BSs) has increased significantly, resulting in a high probability that users are in the near-field region. Note that it is [...] Read more.

With the rapid development of sixth-generation (6G) wireless networks and large-scale multiple-input multiple-output (MIMO) technology, the number of antennas deployed at base stations (BSs) has increased significantly, resulting in a high probability that users are in the near-field region. Note that it is difficult for the traditional far-field plane-wave model to meet the demand for high-precision beamforming in the near-field region. In this paper, we jointly optimize the power and the number of antennas to achieve the maximum energy efficiency for the users located in the near-field region. Particularly, this paper considers the resolution constraint in the formulated optimization problem, which is designed to guarantee that interference between users can be neglected. A low-complexity optimization algorithm is proposed to realize the joint optimization of power and antenna number. Specifically, the near-field resolution constraint is first simplified to a polynomial inequality using the Fresnel approximation. Then the fractional objective of maximizing energy efficiency is transformed into a convex optimization subproblem via the Dinkelbach algorithm, and the power allocation is solved for a fixed number of antennas. Finally, the number of antennas is integrally optimized with monotonicity analysis. The simulation results show that the proposed method can significantly improve the system energy efficiency and reduce the antenna overhead under different resolution thresholds, user angles, and distance configurations, which provides a practical reference for the design of green and low-carbon near-field communication systems. Full article

(This article belongs to the Special Issue Extremely Large-Scale MIMO Technologies for 6G Near-Field Communications)

► Show Figures

Figure 1

20 pages, 2037 KiB

Open AccessArticle

A UNet++-Based Approach for Delamination Imaging in CFRP Laminates Using Full Wavefield

by Yitian Yan, Kang Yang, Yaxun Gou, Zhifeng Tang, Fuzai Lv, Zhoumo Zeng, Jian Li and Yang Liu

Sensors 2025, 25(14), 4292; https://doi.org/10.3390/s25144292 (registering DOI) - 9 Jul 2025

Abstract

The timely detection of delamination is essential for preventing catastrophic failures and extending the service life of carbon fiber-reinforced polymers (CFRP). Full wavefields in CFRP encapsulate extensive information on the interaction between guided waves and structural damage, making them a widely utilized tool [...] Read more.

The timely detection of delamination is essential for preventing catastrophic failures and extending the service life of carbon fiber-reinforced polymers (CFRP). Full wavefields in CFRP encapsulate extensive information on the interaction between guided waves and structural damage, making them a widely utilized tool for damage mapping. However, due to the multimodal and dispersive nature of guided waves, interpreting full wavefields remains a significant challenge. This study proposes an end-to-end delamination imaging approach based on UNet++ using 2D frequency domain spectra (FDS) derived from full wavefield data. The proposed method is validated through a self-constructed simulation dataset, experimental data collected using Scanning Laser Doppler Vibrometry, and a publicly available dataset created by Kudela and Ijjeh. The results on the simulated data show that UNet++, trained with multi-frequency FDS, can accurately predict the location, shape, and size of delamination while effectively handling frequency offsets and noise interference in the input FDS. Experimental results further indicate that the model, trained exclusively on simulated data, can be directly applied to real-world scenarios, delivering artifact-free delamination imaging. Full article

(This article belongs to the Section Sensing and Imaging)

Journal Description

Sensors

Latest Articles

Journal Menu

Journal Browser

Highly Accessed Articles

Latest Books

E-Mail Alert

News

Topics

Conferences

Special Issues

Topical Collections

Further Information

Guidelines

MDPI Initiatives

Follow MDPI