Internet of things (IoT) is an important technology that has numerous applications in many industries. Since IoT involves sensing of data and communication of this data between IoT nodes and IoT sensors, it is vulnerable to cyberattacks from malicious users. Anomaly detection is an important cybersecurity technique that can help identify abnormal data traffic patterns. In this work, a comparative evaluation of unsupervised anomaly detection techniques using feature reduction is presented for IoT traffic analysis. A realistic IoT attacks dataset developed by the Canadian Institute for Cybersecurity is used in this regard. The attacks considered in this dataset include denial of service HTTP flooding attacks. We utilize four machine learning techniques, namely, isolation forest, local outlier factor (LOF), one‐class support vector machine (one‐class SVM), and elliptic envelope on the considered dataset. Extensive simulation results show that one‐class SVM has the best performance in terms of recall and F1 score. On the other hand, isolation forest shows the balanced performance for all the considered metrics.
Nowadays, fall detection systems are attracting attention in the literature due to the pressing need to alert caregivers promptly when an older adult or individual experiences a fall. Real‐time detection of falls may significantly reduce the risks of severe consequences. Falls among elderly individuals carry substantial health implications and can place a significant strain on the healthcare system in developing countries. While fall detection methods such as context‐aware and wearable devices offer promising solutions, they have limitations in accuracy and usability. However, recent advancements integrate sensor technologies and machine learning algorithms that led to promising developments in fall detection systems, with innovative approaches to improve accuracy and response time. This study examines the key challenges, sensors, and general architecture of fall detection. The study reviewed recent literature about the advancements and ongoing research issues in fall detection. The study discusses the data collection mechanism, feature extraction method, and various machine learning algorithms to consider for developing a fall detection system. The results show that there is no effect in choosing multiple sensors over single sensor. This review offers insight into advancements in fall detection technologies and highlights recommendations for future research and developments.
Formaldehyde is a widely used chemical in pathology laboratories and a known occupational hazard with established carcinogenic potential. This study presents the design and validation of a low-cost, Internet of Things (IoT)-based monitoring system for real-time formaldehyde detection in indoor clinical environments. The device integrates a ZE08-CH2O electrochemical gas sensor, an ESP32-WROOM-32 microcontroller, and an LCD interface, all supported by embedded firmware for continuous environmental sampling. Field testing was conducted at a national pediatric pathology laboratory, where four critical work zones, grossing station, sample reception, tissue processor, and reagent storage, were monitored over a fixed time interval. Results from the custom-built system were compared with a certified commercial monitor (Yvelines HTO 131) using Bland-Altman analysis to assess agreement. Mean differences across zones were within +/- 0.03 mg/m3, with 95% limits of agreement (LoA) ranging from -0.05 to +0.06 mg/m3 and standard deviations of the differences ranging from 0.36% to 2.08%, supporting the system's suitability for real-time monitoring while also enabling the identification of concentration levels approaching occupational exposure thresholds, highlighting its relevance for occupational exposure monitoring in clinical settings. The total system cost remained below 60 USD, enabling scalable deployment in settings with limited technical infrastructure. The results support the clinical applicability of embedded formaldehyde-sensing technologies as accessible tools for occupational safety in low-resource healthcare environments.
Many significant and transmissible diseases, including human immunodeficiency virus (HIV)/acquired immune deficiency syndrome (AIDS), COVID‐19, hepatitis, and influenza, are caused by viruses. The rapid replication ability of these viruses is the primary reason for their widespread transmission, resulting in consequences such as loss of life at both the individual and community levels. To prevent the spread of these diseases, reliable diagnostic tools and practical approaches are essential. HIV is an immune system‐attacking retrovirus that weakens the human body’s ability to resist infection and illness by destroying T cells. This review presents a comprehensive and integrated analysis of nanostructured electrochemical biosensors for rapid HIV detection, providing a detailed overview of development strategies that incorporate the application of nanomaterials and immobilization techniques, as well as the underlying sensing mechanism. It also provides details of possible available HIV biomarkers within the states of HIV infection cycles for electrochemical sensing, while comparing electrochemical methods with conventional methods such as enzyme‐linked immunosorbent assay (ELISA), enzyme immunoassay test (EIA), and Western blot. This work adopts an interface and fabrication‐centric perspective to highlight the challenges, including reproducibility, selectivity, and scalability, associated with the development of electrochemical sensors for HIV detection. This work establishes the guiding framework and rationale for designing and transitioning principles to develop an electrochemical HIV biosensor.
Recent developments in IoT technology provide new solutions for field‐based monitoring of plant health. Among the many rapid detection techniques applied to the perception layer of IoT, electrical feature‐based detection technology stands out due to its distinct advantages such as low cost, ability to extract internal plant information, and immunity to external light sources. This article provides an overview of on‐site detection methods for plant health based on electrical features. Basic experimental and research methods for plant health detection are presented in the dimensions of sample collection and preprocessing, electrical feature extraction, and data processing. The article then classifies and elaborates on these methods based on four key research subjects: plant ontology, symbiotic bacteria, pathogens, and the microenvironment. For each subject, it introduces the underlying detection principles, traces technological developments, and analyzes the current state of applications, major scientific challenges, and prevailing research hotspots. This article aims to integrate research results from various fields from the perspective of plant health detection, extract research results with inherent correlations from different disciplines such as agriculture, food, environment, detection, and data analysis, and connect them through the development of electrical detection technology, providing interdisciplinary references and perspectives for researchers in this field.
As the fastest‐growing economic country in the Asia Pacific, Bangladesh needs to be technologically advanced and adaptive in every sector to foster its development. In recent times, Internet of Things (IoT) has shown its potential for improving productivity and efficiency, thus contributing to a country’s economy. With technological advances, the need for low‐power, low‐cost, and secured IoT solutions has become prominent, especially for a developing country like Bangladesh. Long‐range wide‐area network (LoRaWAN) is an IoT solution that meets those criteria and provides exceptional long‐range coverage in addition. In this article, the possibilities of LoRaWAN technology have been analyzed from the perspective of Bangladesh. Initially, a technical review has been performed on its background, network topology, and compared with other leading technologies. Then a design architecture of the LoRaWAN’s end‐node and gateway has been presented. The article also investigates the possible deployment scenario and future scopes of LoRaWAN in Bangladesh. Finally, a LoRaWAN system has been demonstrated using two nodes in an urban scenario and experimentally evaluated data reception, received signal strength, and LoRa packet loss or delay over two different communication ranges. Based on the experimentation conducted, a maximum coverage of 1.094 km has been estimated by the log‐distance probabilistic approach.
Road traffic accidents (RTAs) remain a major global safety challenge, requiring reliable automated detection systems to support rapid emergency response, traffic management, and smart city surveillance. This study presents a comparative evaluation of deep learning models for road accident detection using heterogeneous visual traffic sources, namely CCTV images and dashcam videos. A unified evaluation framework is introduced to systematically compare traditional machine learning, standard deep learning, and hybrid deep learning models within a consistent experimental setting. Publicly available datasets comprising several 1000 annotated CCTV images and thousands of dashcam video clips were used, covering diverse traffic scenarios, including urban and highway environments, day and night conditions, weather variations, and multiple camera viewpoints. The evaluated models were grouped into baseline approaches, consisting of Support Vector Machine, 2D Convolutional Neural Network (2D-CNN), R-CNN, and Long Short-Term Memory (LSTM) networks, and a proposed hybrid VGG16-LSTM model. This categorization reflects the complementary strengths of spatial and temporal feature extraction for static image-based and dynamic video-based accident detection. For CCTV image analysis, the 2D-CNN achieved the best performance, with 99% accuracy and a 98% F1-score. For dashcam video analysis, the VGG16-LSTM model outperformed competing approaches, achieving 99.53% accuracy and a high area under the receiver operator characteristic (ROC) curve. The findings demonstrate the effectiveness of convolutional models for spatial accident representation and the importance of temporal modeling for video-based detection. The proposed cross-modality framework provides practical insights into model suitability across different traffic data sources and has potential integration into Intelligent Transportation Systems. Future research should examine cross-dataset validation, event-level detection, and robustness in uncontrolled real-world environments.
By enabling the detection of molecular, biochemical, and physiological signals at micro- and nano-scales, bio-nano sensor networks (BNSNs) have emerged as a transformative sensing paradigm for environmental surveillance, biomedical monitoring, and healthcare. However, accurate real-time anomaly detection in BNSNs remains a critical challenge due to molecular noise, bio-interference, signal drift, and the severe energy and computational limitations inherent to bio-nano sensors. This article proposes a lightweight, sensor-aware artificial intelligence (AI) framework for real-time anomaly detection under sixth-generation (6G)-enabled mobile edge computing (MEC) architectures, specifically designed to address these bio-nano sensing constraints. The proposed TinyCNN-GRU framework integrates tiny convolutional neural networks (Tiny-CNNs) with gated recurrent units (GRUs), optimized via posttraining 8-bit quantization and structured pruning for deployment on resource-constrained edge nodes. Unlike conventional IoT-centric solutions, the framework explicitly addresses bio-nano sensing characteristics and edge hardware limitations. Experiments conducted on a synthetic bio-nano sensor dataset (20,000 instances) and the publicly available MIT-BIH Arrhythmia Database demonstrate up to 96.7% detection accuracy, 35% reduction in inference latency, and 48% reduction in model size compared to baseline edge AI models. A comprehensive ablation study confirms that 8-bit quantization reduces model size by 48% with less than 0.5% accuracy loss, validating hardware adaptability for ultra-low-power bio-nano sensing platforms. Comparisons against established anomaly detection baselines-including LSTM, Autoencoder, and Isolation Forest-further confirm the superiority of the proposed framework for next-generation 6G bio-nano sensor deployments.
With the rapid development of frontier applications, such as robotic dexterous hands and minimally invasive surgery, there is an urgent demand for miniature six-axis force/torque (F/T) sensors that combine ultrasmall dimensions with high load capacity. While soft tactile sensors excel at texture recognition, they often lack the structural rigidity required for high-precision closed-loop force control. To address this, the study introduces a novel miniature six-axis F/T sensor (& Fcy; 9.5 mm in diameter and 17 mm in height) featuring a stacked vertical structure, designed to serve as the "skeletal" proprioception node in a hierarchical sensing architecture. By optimizing the utilization of vertical space, force transmission paths, and stress distribution, this design achieves a high measurement range (150 N in the F-z direction and 1 N m in the M-z direction) within an extremely compact form factor. A backpropagation (BP) neural network optimized by particle swarm optimization (PSO) was employed for decoupling. The sensor exhibits a nonlinearity error of <0.4% F.S., a hysteresis error of <0.8% F.S., a repeatability error of <0.4% F.S., and a coupling error of <3% F.S. The proposed design demonstrates competitive static performance among representative miniature multiaxis F/T sensors of similar scale and provides a practical route toward ultrasmall, high-robustness multidimensional force sensors. Given its ultracompact dimensions, the present prototype is well suited for dexterous robotic hands operating in space-constrained environments; broader applications in minimally invasive surgery, haptic systems, or aerospace scenarios would require dedicated dynamic, fatigue, and, where relevant, biocompatibility and sterilization validation.
This study presents a simulation-driven digital twin (DT) prototype framework, designed around a wireless sensor network (WSN) style pipeline, to study congestion and vehicular emissions at an urban signalized junction. The architecture follows a sensing-to-edge workflow: detector-like traffic measurements (counts and speeds) are published through an MQTT middleware, synchronized into time-aligned snapshots, and used to update a SUMO-based microscopic model that acts as the virtual twin. In the present study, sensing is emulated within the simulator, and no roadside deployment is carried out. Three signal control strategies are evaluated: an uncontrolled (random release) baseline, a fixed-time plan, and a priority-based plan with simple queue-responsive green extensions. 1 h simulation experiments are conducted for the junction demand setting, and performance is assessed using waiting time, queue length, and HBEFA-based emission estimates for five pollutants. Results show that lightweight state-aware timing rules can substantially reduce delays and stabilize queues relative to the uncontrolled baseline in this configuration, while higher electric vehicle (EV) penetration yields marked reductions in NOx and PMx in the emission model. At the edge layer, K-means clustering is applied to speed and count features to label links into low, medium, and high congestion regimes and to highlight congestion hotspots for operator interpretation. Since the demand is synthetic and sensing is emulated, the findings are indicative and not calibrated for any specific city. The prototype provides a practical base for future extensions using optimization- or learning-based control and field calibration.
Addressing three-component errors in fluxgate magnetometers, this study introduces a novel error correction approach based on particle swarm optimization with simulated annealing (PSO-SA). Through an exhaustive analysis of error sources, a robust correction model is formulated. PSO-SA seamlessly integrates a simulated annealing mechanism into the conventional PSO algorithm, enhancing its capability to evade local optima and achieve superior global search proficiency. The method's viability is rigorously validated via comprehensive numerical simulations. An experimental framework is meticulously devised to rectify three-component errors in fluxgate magnetometers using particle swarm optimization (PSO), particle swarm-genetic algorithm (PSO-GA), and the proposed PSO-SA. Results distinctly reveal the remarkable superiority of PSO-SA over PSO in terms of correction efficacy, yielding a substantial 38% enhancement in calibration precision and a notable 20% escalation in calibration speed compared to PSO-GA. However, it is prudent to acknowledge that although the corrective outcomes closely approximate the values garnered from standard magnetometer measurements, they do not warrant exact representation of the authentic magnetic field. Consequently, the algorithmic methodology cannot entirely substitute the manual vector calibration technique. Nonetheless, this innovative approach presents a swift and precise corrective strategy devoid of the prerequisite for specialized laboratory setups. It showcases accelerated convergence and heightened accuracy, thus rendering it an apt contender for meticulous fluxgate magnetometer calibration in pragmatic real-world scenarios.
The development of electrochemical sensors (ECSs) with enhanced sensitivity has emerged as a focal point in contemporary sensor technology research due to the important biochemical roles of hydrogen peroxide (H2O2) in the industrial and biological fields. In this paper, a high-performance ECS based on laser-irradiated carbon nanotube (LICNT) electrode as the working electrode (ECS-LICNT) for sensitive detection of H2O2 is designed and fabricated to achieve long-term monitoring of trace concentrations of H2O2. The thickness and surface microcluster structure of LICNT electrodes were optimized by precise control of laser energy density and CNT suspension dose. The ECS-LICNT, designed with a dual mechanism combining surface laser-induced oxidation to increase active point density and ordered microcluster structure to enhance effective active area, possesses a high sensitivity of 475 mu A & centerdot;mM(-1)& centerdot;cm(-2) in a wide range (10-9000 mu M), surpassing most reported pure carbon-based sensors. In addition, the sensor exhibits excellent stability (current decay <5% within 7200 s) and reproducibility (RSD = 2.3%, n = 7), maintaining 95.8% of its initial response even after 28 days of storage. The ECS demonstrates sufficient sensitivity to accurately determine H2O2 concentration in the presence of interfering substances such as ascorbic acid (AA), glucose (Glu), and uric acid (UA). It can be foreseen that this sensor provides a novel strategy for designing efficient ECSs and has broad application prospects in the fields of biomedicine and food safety detection.
Accurate measurement of photoinduced temperature rise (Delta T) in optically absorbing thin films is important for understanding light-matter interactions, preventing laser-induced damage, and validating photothermal models. However, existing techniques often rely on contact probes or calibration-intensive procedures. Here we present a noncontact method for estimating Delta T using an atomic force microscopy (AFM) cantilever as a mechanical sensor of photoacoustic (PA) loading. The optical excitation is tuned to a molecular absorption resonance and intensity-modulated at the cantilever fundamental resonance. Periodic thermoelastic expansion of the heated surface generates an acoustic pressure wave that drives the cantilever oscillation. A closed-form relation is derived that converts the measured on-resonance cantilever oscillation amplitude into Delta T using known cantilever, acoustic, and thermoelastic parameters. The method is demonstrated on a 500 nm poly(methyl methacrylate) (PMMA) film excited at 1733 cm-1, where the inferred temperature rise is Delta T approximate to 1.6 K. A three-dimensional finite-element model coupling electromagnetic absorption and heat transfer predicts Delta T approximate to 1.4 K, in good agreement with the experiment. These results demonstrate a simple route for estimating optically induced temperature rise using standard AFM instrumentation and provide a practical tool for photothermal characterization of thin films.
Glycated hemoglobin (HbA1c), a form of hemoglobin (Hb) bound to glucose, is a widely used biomarker for hyperglycemia, providing diagnostic and monitoring value in diabetes and prediabetes. This study investigated the development of an electrochemical aptasensor for HbA1c detection based on a screen-printed carbon electrode (SPCE) modified with gold nanoparticles (AuNPs). A biotinylated aptamer was immobilized onto the electrode surface through streptavidin-biotin interaction, and the signal response was evaluated using differential pulse voltammetry (DPV) and electrochemical impedance spectroscopy (EIS) in the presence of potassium ferricyanide ([K3Fe(CN)6]) as a redox probe. The aptasensor response was characterized by a decrease in current upon HbA1c binding. Experimental parameters were systematically examined, and the optimal conditions were obtained at an aptamer concentration of 1.5 & micro;g/mL, an aptamer incubation time of 60 min, and an HbA1c incubation time of 4 min. Under these conditions, the detection limit was 1.17% and the quantification limit was 3.5%, with a dynamic range of 2%-15%. The results indicate that this approach enables sensitive and selective HbA1c detection, supporting its potential application in diabetes diagnosis and monitoring through point-of-care (PoC) testing.
Compared with traditional biometrics like fingerprints and iris recognition, human odor-based identification has attracted significant attention due to its unique advantages, including noncontact operation and high concealability. However, its relatively low long-term prediction accuracy (current state-of-the-art recognition rates are similar to 20%) limits practical applications. To explore the long-term stability of human odor, this study uses a self-developed electronic nose (eNose) system and selects three body odor sources-oral, ear, and armpit-for recognition. The experiment includes two parts: first, a 1-month baseline test using standard cross-validation, which confirms the system's capability to capture distinctive odor patterns, with single-day rates remaining high (98%-100%). Second, and most critically, a long-term prediction experiment employing a strict time-based data-splitting strategy-training exclusively on first-day samples and testing on all subsequent days-to prevent data leakage and evaluate real-world generalization. This rigorous evaluation shows that first-day armpit odor can predict identity 1 month later with 80% accuracy (overall average 70%), and monthly stability remains at similar to 60%. These findings confirm human odor's potential for cross-month identification under methodologically sound validation, providing key theoretical support for developing noncontact, long-term biometric authentication systems.
This article presents an energy-efficient intelligent edge surveillance expert system for real-time intrusion detection in resource-constrained environments. Unlike conventional continuous surveillance systems, the proposed framework introduces a motion-triggered inference mechanism using passive infrared (PIR) sensing, enabling selective activation of deep learning models. This transforms surveillance into an event-driven expert system that integrates perception, reasoning, and decision-making. The system achieves 95.54% accuracy and 93.07% F1-score, and reduces energy consumption by similar to 40% compared to continuous inference. The proposed architecture demonstrates improved detection reliability, reduced false alarms, and enhanced computational efficiency, making it suitable for scalable embedded surveillance applications.
Accurate calibration of triaxial magnetometers is critical for magnetic anomaly detection, yet existing optimization-based calibration methods often suffer from premature convergence and suboptimal parameter estimation. To address these limitations, this article proposes a Tent map and Selection-phase enhanced African Vulture Optimization Algorithm (TS-AVOA) for efficient magnetometer error correction. The proposed TS-AVOA enhances global search diversity and local convergence accuracy by integrating a Bald Eagle Search (BES)-inspired selection mechanism and embedding Tent chaotic mapping into both the exploration and exploitation phases. Numerical simulations demonstrate that the maximum absolute error between the calibrated and theoretical magnetic field amplitude is reduced to just 1.1 nT, with parameter estimation accuracy exceeding 99.9%. Comparative analysis against PSO, GA, and original AVOA confirms that TS-AVOA outperforms all baselines in convergence speed and calibration accuracy. Field experiments further validate that total geomagnetic fluctuation is suppressed from 282.47 to 9.79 nT, achieving a 28.86-fold reduction. These results highlight the practical potential of TS-AVOA in enhancing the reliability of fluxgate magnetometers for geomagnetic applications.
This study introduces MGaming EEG-BCI, a mobile brain-computer interface framework designed to enable gesture-free gaming through neural activity based on motor imagery (MI). The proposed system integrates an OpenBCI Cyton + Daisy headset for noninvasive EEG acquisition with a real-time processing pipeline, including band-pass filtering, independent component analysis (ICA), common spatial patterns (CSPs) feature extraction, and linear discriminant analysis (LDA) classification. A Firebase-based cloud infrastructure enables real-time synchronization with a Unity-developed mobile game incorporating tap-style interaction. Fifteen participants were recruited, of whom nine completed the full protocol, including training, gameplay, and user feedback. The system achieved a mean classification accuracy of 55.29%, with peak performance reaching 98.82%. Most participants successfully exceeded the predefined gameplay score threshold. User feedback indicated strong interest in future use, despite challenges related to calibration time and device comfort. The MGaming EEG-BCI demonstrates the feasibility of gesture-free, EEG-based mobile gaming and highlights its potential for inclusive and accessible interaction. The system provides a foundation for future developments in accessible entertainment, rehabilitation, and advanced human-computer interaction (HCI).
Medical anomaly detection remains a critical challenge due to the heterogeneity of multimodal imaging data and the dynamic nature of physiological sensor readings. Existing fusion-based approaches, which often rely on shallow concatenation or modality-specific networks, struggle with cross-modality inconsistencies, loss of diagnostic cues during integration, and elevated false-alarm rates, thereby limiting reliable clinical deployment. Moreover, most conventional frameworks fail to jointly capture the spatial complexity of imaging modalities and the temporal dependencies of biosensor data, leading to suboptimal anomaly recognition and classification. To address these limitations, we propose the deep pattern recognition fusion network (DPRFN), a novel deep learning and pattern recognition framework that jointly exploits multimodal medical imaging and physiological sensors through an attention-driven fusion mechanism for advanced anomaly detection. The architecture integrates ResNet-50-based hierarchical feature extraction for spatial representation with a bidirectional gated recurrent unit (Bi-GRU) to model sequential dependencies in sensor signals. The attention-guided fusion module ensures optimal cross-modality harmonization, retaining clinically relevant features while improving sensitivity to subtle anomalies. The framework was validated on the Brain Tumor Segmentation Challenge 2021 (BraTS 2021) dataset (2000+ multimodal magnetic resonance imaging [MRI] scans with ground truth segmentations) and Medical Information Mart for Intensive Care-III (MIMIC-III)/PhysioNet physiological records, which provide rich spatial, volumetric, temporal, and biosignal annotations-enabling robust benchmarking of fusion-based anomaly detection systems. Experimental results demonstrate significant improvements, achieving accuracy = 97.6%, F1-score = 96.9%, area under the curve (AUC) = 0.984, and a false-positive rate = 2.1%, outperforming state-of-the-art baselines by 3%-5% across key metrics. In addition to standard evaluations, we report advanced anomaly detection parameters: structural similarity index (SSIM) = 0.912 for imaging integrity, Matthews correlation coefficient (MCC) = 0.937 for anomaly classification stability, and a diagnostic confidence index (DCI = 0.923), a novel metric introduced to assess the reliability of anomaly detection in clinical decision-making. Importantly, the reduction in false positives directly translates to fewer unnecessary clinical interventions, highlighting practical medical relevance. The results show that DPRFN delivers a robust and scalable solution for multimodal anomaly detection, significantly reducing diagnostic uncertainty while preserving critical cross-modality information. Beyond brain imaging, the framework demonstrates high adaptability and reusability, with potential applications in cardiac arrhythmia detection, neurodegenerative disorder monitoring, and other sensor-driven anomaly detection scenarios. These findings position DPRFN as a next-generation computational intelligence framework for precision diagnostic imaging and anomaly recognition.
To address the challenges of detecting the variable pose of bolster springs within the confined space of a bogie frame and identifying potential entrapment interactions between the springs and surrounding obstacles, this study proposes a method for identifying the trapped state of bolster springs based on heterogeneous information fusion. This approach improves the accuracy of pose detection and enables precise perception of trapped states through a constructed identification model. Initially, the principles of multisensor joint calibration, heterogeneous information fusion, and the construction of the bolster spring trap identification model are presented. Subsequently, the architecture and loss function of RTMDet were optimized to achieve accurate detection of the planar pose of the bolster spring based on image data. Statistical filtering and Random Sample Consensus (RANSAC) were employed for the segmentation of external spring point clouds and the initial pose estimation. Additionally, pose optimization was performed based on projection roundness and a genetic algorithm. Finally, the identification experiment for the trapped state of the bolster spring was successfully completed based on the proposed model. The results indicated that the method achieved a root mean square error (RMSE) of 1.47 mm for position and 0.84 degrees for orientation, with an accuracy of 94.4% in classifying whether the spring is trapped or not trapped. Compared to mainstream methods, this approach significantly improves the accuracy of bolster spring pose detection. This advancement provides a novel identification strategy for workpieces trapped in confined complex environments, offering potential applications in the intelligent perception system design of robots for workpiece assembly and disassembly in the equipment manufacturing industry.
