International Journal of Sensor Networks (20 papers in press)

Regular Issues

• Optimizing Power Management in Wireless Sensor Networks Using Machine Learning: An Experimental Study on Energy Efficiency  
  by Mohammed Amine Zafrane, Ahmed Ramzi Houalef, Miloud Benchehima 
  Abstract: Wireless sensor networks (WSNs) have emerged as essential components across various fields. Comprising small, self-sustaining devices known as "Nodes," they play a critical role in data collection and analysis. However, ensuring optimal longevity without compromising data collection timeliness is a fundamental challenge. Regular data aggregation tasks, while essential, consume substantial energy resources. Furthermore, constraints in computation power, storage capacity, and energy supply pose significant design challenges within the Wireless Sensor Network domain. In pursuit of optimizing energy efficiency and extending the operational lifetime of nodes through artificial intelligence, we have developed a prototype for data collection to create a comprehensive dataset. Our approach leverages both current and precedent measurements, triggering data transmission only in the presence of significant changes. This intelligent strategy minimizes unnecessary communication and conserves energy resources. Based on the
  Keywords: Artificial intelligent; WSN; Power optimization; data acquisition.
  DOI: 10.1504/IJSNET.2024.10068162
   
  
• Named Entity Recognition for Function Point Descriptions in Software Cost Estimation Processes  
  by Boyan Zhao, Xiaofei Zou, Shijie Xin, Di Liu 
  Abstract: With the advancement of software technology, the industrys informatisation level has improved, but the growing size and complexity of software have raised costs. Consequently, assessing software project costs early is crucial. Function point analysis, the primary method for cost evaluation, quantifies functional elements like external data inputs and outputs to measure software size from the users perspective. However, it heavily relies on manual effort, especially in extracting function point descriptions, leading to errors and inefficiency. This paper proposes an entity recognition model to address these challenges, integrating a BiLSTM-CRF framework with CNN layers and hierarchical learning. A domain-specific dictionary is developed to enhance the models performance. Experimental results show that the proposed method outperforms BERT by improving accuracy by 0.42% and recall by 1.04%. The method achieves 95.37% accuracy in entity recognition for a sensor data system, demonstrating its effectiveness and reliability in software cost evaluation.
  Keywords: Software cost evaluation; Named entity recognition; Convolutional neural network; Bidirectional long-short memory network; Hierarchical learning.
  DOI: 10.1504/IJSNET.2025.10070067
   
  
• Advanced Air Quality Prediction Modelling using Intelligent Optimisation Algorithm in Urban Regions  
  by Wendi Tan, Zhisheng Li 
  Abstract: Ambient air contamination is a significant environmental challenge threatening human well-being and quality of life, especially in urban areas. Technical breakthroughs in artificial intelligence models offer more accurate air quality predictions by analysing significant data sources, including meteorological factors like humidity, wind speed, and pollution data. Also, existing methods of air quality prediction often lack due to their dependency on statistical models that may not adequately capture the complexities of environmental data like pollutants and meteorological factors. Thus, the research introduces a grey wolf optimised variational autoencoder to enhance the air quality prediction by effectively capturing complex relationships in environmental data. The model acquires the probabilistic nature of variational latent representations from historical air quality input data and prevents overfitting. The relevant features are selected using the grey wolf technique, identifying the appropriate variables to enhance the data quality. Additionally, it optimises critical hyperparameters like learning rates and greedy layer sizes, leading to better convergence during model training and improved performance in air quality index prediction. Experimental results demonstrate improved prediction accuracy, reduced error rate, and faster convergence.
  Keywords: Air Quality Index; Prediction Model; Optimization; Artificial Intelligence; Urban Region; Environmental factors.
  DOI: 10.1504/IJSNET.2025.10070252
   
  
• LSB-DSN: Sensor-Assisted Deep Learning for Robust English Speech Recognition  
  by Aili Tang, Dezi Zeng 
  Abstract: In modern communication, the English speech recognition system is essential for improving the personalised user experience and global communication. The recognition systems use sensor devices and deep learning techniques to ensure the system's robustness in all diverse environments. The traditional system efficiency is reduced due to accents, varying pronunciations, and limited contextual considerations. The paper introduces the lion-swarm boosted deep sesame networks, a new speech recognition framework that fuses sensor technologies and deep learning to improve accuracy and robustness. This model combines acoustic signals with sensor-based inputs from accelerometers, gyroscopes, and electromyography devices to capture the delicate speech-related modulations for better recognition across diversified environments. The hierarchical attention mechanism and lion swarm optimisation enable optimal feature selection, reducing the recognition error and computational overhead. The experiments show that it achieves a 9.5% word error rate in clean conditions, a low cross-entropy loss of 0.65, and 100 ms of processing latency far superior to baseline models for noisy environments. The proposed framework can adapt to different accents and pronunciations, making it a strong solution for real-world applications in speech recognition.
  Keywords: English Speech Recognition; signal modulation; sensor devices; subnetworks; lion swarm optimization; similarity measures; deep sesame networks.
  DOI: 10.1504/IJSNET.2025.10070349
   
  
• ESFM-Net: Low-dose CT Artifact Suppression Method based on Edge-Guided Spatial-Frequency Mutual Network  
  by Xueying Cui, Weisen Song, Xiaoling Han, Lizhong Jin, Hong Shangguan, Xiong Zhang 
  Abstract: Deep learning has shown superior performance in low-dose CT artefact suppression. However, the existing deep network has a limited perception of edge and texture information, and the transformer-based approaches have high computational complexity in calculating self-attention. To alleviate these issues, an artefact suppression method based on an edge-guided spatial-frequency mutual network (ESFM-Net) is proposed, which can enhance edge and texture information with low computational complexity. Specifically, an edge decoder is designed to supplement multi-scale edge details for reconstructing high-frequency in a single-encoder dual-decoder. For obtaining rich edge and texture information, a spatial-frequency feature extraction module is developed to obtain local spatial and global frequency features with little computational complexity. Considering the complementarity of information, a spatial-frequency mutual module is further constructed to enhance the feature representation capability by adaptive fusion. High and low-frequency features with different scales are also gradually fused through a multi-scale fusion module to obtain final denoised images. The comparative experiments and ablation results show the superior performance of our method in edges, texture preservation, and artefact suppression.
  Keywords: Low Dose CT; Artifact suppression; Edge guidance; Spatial-frequency mutual fusion.
  DOI: 10.1504/IJSNET.2025.10070503
   
  
• PULP-Lite: a More Light-weighted Multi-core Framework for IoT Applications  
  by Yong Yang, Yuyu Lian, Yanxiang Zhu, Shun Li, Wenhua Gu, Ming Ling 
  Abstract: The increasing volume of data generated by real-time applications and sensors places significant performance demands on processors. Single-core processors are constrained by the inherent limitations of their architecture in terms of parallel processing capability, making it challenging to handle real-time applications. To address this, we propose parallel ultra low power lite (PULP-Lite), a tightly coupled multi-core on-chip system to efficiently handle near-sensor data analysis in the Internet of Things endpoint devices. PULP-Lite uses low-latency interconnections and an innovative address-mapping mechanism to connect central processing units (CPU), ensuring high-performance processing while maintaining flexibility with a lightweight multi-core programming. We evaluate a field-programmable gate array (FPGA) implementation of PULP-Lite with 8 cores, showing a speedup of 6
  Keywords: multi-core optimization; parallel processing; computer systems organization.
  DOI: 10.1504/IJSNET.2025.10070743
   
  
• ToI-Model: Trustworthy Objects Identification Model for Social-Internet-of-Things (S-IoT)  
  by Rahul Gaikwad, Venkatesh R 
  Abstract: The social-internet-of-things (SIoT) paradigm integrates social concepts into IoT systems. Identifying trustworthy SIoT objects, as well as managing trust, are essential for promoting cooperation among them. The current state-of-the-art methods inadequately quantify the trustworthiness of SIoT objects and fails to evaluate trustworthiness of SIoT objects. This paper comprehensively considers specific features of SIoT objects and integrates them with the theory of social trust. The proposed Trustworthy-objects identification model (ToI-Model) captures comprehensive trust, proficiency, readiness, recommendation, reputation, honesty and excellence metrics for identifying trustworthy objects in SIoT. Service requester (SR) uses trust score of service providers (SP) before initiating service delegation. A series of experiments are conducted to evaluate the proposed trust models effectiveness in the successful completion of services, convergence, accuracy, and resilience against deceitful activities. Results of experiment shows that trust model identifies trustworthy service provider that has 19.89% more trust score and a 27.61% less latency than state-of-the-art models.
  Keywords: Trustworthy; Proficiency; Readiness; Recommendation; Reputation; Honesty; and Excellence.
  DOI: 10.1504/IJSNET.2025.10070868
   
  
• Adaptive Energy-Efficient Task Offloading and Resource Management in UAV-Assisted Mobile Edge Networks using Dynamic DRL  
  by Y. Zhou, H. Cao, J. Duan, H. Qing, Amin Mohajer 
  Abstract: Next-generation aerial edge networks must support delay-critical and computation-intensive services in highly dynamic wireless environments. This paper introduces a distributed control architecture that integrates flight-aware workload distribution, predictive mobility mapping, and adaptive edge resource slicing. The model combines spatiotemporal learning with an enhanced policy gradient mechanism for joint optimisation of service placement, bandwidth provisioning, and UAV trajectory scheduling. By integrating predictive modelling of user mobility via recurrent neural structures and embedding temporal attention, the system anticipates regional demand fluctuations and proactively reconfigures aerial coverage. A dual-critic actor-learner structure ensures stable policy evolution under hybrid discrete-continuous action spaces. Extensive evaluations across diverse network densities and traffic dynamics reveal that the proposed scheme improves task finalization rates by over 30%, sustains autonomous operation via harvested energy, and consistently outperforms existing baselines in spectral efficiency and decision latency. These findings position the framework as a robust foundation for real-time orchestration in scalable, mission-adaptive aerial edge infrastructures.
  Keywords: UAV-Enabled Edge Intelligence; Predictive Mobility Modeling; Spatiotemporal Resource Orchestration; Policy-Driven Task Offloading; Dynamic Edge Slicing.
  DOI: 10.1504/IJSNET.2025.10071611
   
  
• Adaptive Offloading in Multi-Access Edge Networks via Hierarchical Federated Learning and Real-time System Adaptation  
  by J. Wang, Q. Liang, Amin Mohajer 
  Abstract: Achieving ultra-reliable real-time digital twin (DT) adaptation in mobile edge environments requires intelligent orchestration of computation and communication under user heterogeneity and dynamic mobility. This paper introduces GADENet, a Graph Attention-enhanced Digital twin Evolution Network, that fuses graph neural modelling, multi-agent actor-critic learning, and hierarchical federated personalisation to enable seamless digital representations of user equipment (UE) in distributed edge networks. At its core, GADENet employs a GAT-assisted Multi-Agent Deep Deterministic Policy Gradient (MADDPG) framework to jointly learn optimal DT migration and personalization strategies across edge servers, guided by real-time traffic topologies and resource interdependencies. Each DT model is modularized into generalizable and adaptive subspaces, trained collaboratively through a three-tier edge-cloud federated loop and refined using localized attention-based updates. For efficient mobility handling, we propose a parameter-sliced DT relay protocol that selectively migrates the minimal personalisation subset across servers, leveraging learned action-value functions to minimize response latency. Extensive simulations on CIFAR-based datasets and synthetic edge workloads demonstrate that GADENet achieves up to 30% reduction in interaction latency and significantly boosts modelling fidelity versus strong federated and DRL-based baselines. This work offers a principled blueprint for intelligent DT deployment under the constraints of 6G and next-gen IoT fabrics.
  Keywords: Digital Twin Modeling; Graph Attention Networks; Multi-Agent Deep Reinforcement Learning; Federated Learning; Intelligent Network Orchestration.
  DOI: 10.1504/IJSNET.2025.10071733
   
  
• A Decoupling Algorithm for Three-Dimensional Electric Field Sensors Based on Extreme Learning Machines Optimised by Bat Algorithm  
  by Wei Zhao, Zhizhong Li 
  Abstract: During measuring the spatial electric field intensity using a three-dimensional electric field sensor, due to the electric field components coupling effect caused by the electric field distortion, a certain coupling error exists in the electric field intensity components measurement. Aiming at the problem of insufficient decoupling accuracy of the traditional extreme learning machine method, an optimised extreme learning machine method based on the combination of maximum inter-class variance and the bat algorithm is proposed to decouple the three dimensional electric field sensor. The Bat algorithm optimised the extreme learning machine methods optimal initial weight and threshold. The maximum inter-class variance method was used to analyse the inherent coupling characteristics of the sensor. The coupling effect was classified according to the varying coupling contribution degree. The traditional extreme learning machine decoupling network was extended. The calibration experiments and decoupling calculations show that the extreme learning machine algorithm optimised by the bat algorithm and maximum inter-class variance can effectively reduce the error, which is between the electric field components obtained by the model calculation and the actual electric field components, and can effectively reduce the interference generated by the inter-dimensional coupling effect of the sensor, and further improve the measurement accuracy of the electric field intensity.
  Keywords: bat algorithm; decoupling; extreme learning machine; maximum inter-class variance; three-dimensional electric field sensor.
  DOI: 10.1504/IJSNET.2025.10072003
   
  
• A Self-Distillation Approach for Enhancing Intelligence Tutoring System Math Solving Based on Large Language Models  
  by Guanlin Chen, Yuchen Jin, Wenyong Weng, Tian Li, Jianshao Wu 
  Abstract: Intelligent Tutoring Systems have demonstrated strong capabilities in supporting students' learning, particularly in solving predefined problems. However, they have a key limitation: Intelligent Tutoring Systems are designed to solve only the problems specifically programmed into the system. This paper introduces a novel approach that integrates large pre-trained models into local Intelligent Tutoring Systems to address this challenge. Specifically, we propose a method where a local large pre-trained model generates high-accuracy logical reasoning through the Chain of Thought and enhanced computational capabilities via the Program of Thought. By combining these two outputs, we generate high-quality synthetic data to train the local model, improving its ability to solve a broader range of mathematical problems, including those it has not previously encountered. Experimental results demonstrate that our approach significantly enhances both reasoning precision and computational efficiency, ultimately improving the overall performance of local Intelligent Tutoring Systems in supporting students with mathematical problem-solving.
  Keywords: Intelligent Tutoring Systems; Large Pre-trained Models; Chain of Thought; Program of Thought; Mathematical Problem-Solving; Fine-Tuning; Sensor.
  DOI: 10.1504/IJSNET.2025.10072011
   
  
• Cybersecurity, Throughput and Delay Analysis using Multiple Reconfigurable Intelligent Surfaces  
  by Faisal Alanazi 
  Abstract: The integration of Multiple Reconfigurable Intelligent Surfaces (MRIS) into modern wireless communication systems has emerged as a promising solution to enhance throughput and minimize delay. MRIS are artificial surfaces designed to dynamically manipulate the propagation of electromagnetic waves to optimize signal transmission. This paper presents an in-depth analysis of the throughput and delay performance in systems utilising MRIS. We explore the underlying mechanisms by which MRIS can improve channel conditions and reduce interference, thereby improving overall system efficiency. Through analytical models and simulations, we investigate the trade-offs between throughput enhancement and delay reduction in various MRIS-assisted communication scenarios. The results demonstrate that MRIS can effectively optimise the network's performance by providing adaptive control over the signal environment, leading to significant improvements in throughput and latency. We also study the physical layer security using multiple RIS.
  Keywords: Multiple RIS; delay analysis; throughput analysis; outage probability; 6G.
  DOI: 10.1504/IJSNET.2025.10072131
   
  
• A New Hyperchaotic Image Encryption Scheme Based on DNA Computing and SHA-512  
  by Shuliang Sun, Xiping Wang, Zihua Zhao 
  Abstract: Smartphones and digital cameras are becoming more and more widespread in the world. Massive images are generated every day in the world. They are easily transmitted on the insecure channel-Internet. Encryption technique is usually adopted to protect sensitive images during communication. A new cryptosystem is constructed by six-dimensional (6D) chaotic system and deoxyribonucleic acid (DNA) techniques. Firstly, the hash value is calculated. It keeps the encrypted result closely connected with the original image. The initial conditions of the cryptosystem are produced with the generated hash value and secret key. Secondly, the pixel is divided into four parts, forming a large matrix. Scrambling is performed on the new image. Subsequently, DNA coding, modern DNA complementary rules, DNA computing, and DNA decoding are utilised. Diffusion operation is also executed to improve the security, and the ciphered image is achieved finally. The experimental performance reveals that the designed algorithm has some advantages. It also signifies that the designed algorithm could protect against common attacks and is more secure than some existing methods.
  Keywords: hyperchaotic system; DNA computing; SHA-512.
  DOI: 10.1504/IJSNET.2025.10072159
   
  
• Advanced GNSS Positioning for Low-Cost Android Smartphones using RTS Assisted Extended Kalman Filter  
  by Yuhan Zhou, Xuanwen Wang, Shuaiyong Zheng, Xiaoqin Jin, Shuailong Chen, Jixi Liu, Heng Yang 
  Abstract: Global Navigation Satellite System (GNSS) is pervasively employed in smartphone location services. However, the positioning performance of low-cost Android smartphones is often limited by the suboptimal quality of their GNSS observation data, impeding their ability to fulfil user demands. To address this need, this paper proposes an improved positioning algorithm using RTS-EKF (Rauch-Tung-Striebel and Extended Kalman Filter). This algorithm initially enhances the raw data quality through rigorous gross error detection and elimination, coupled with Doppler smoothing. Subsequently, the RTS-EKF algorithm is employed to estimate and smooth localisation results, ultimately enhancing positioning accuracy. To validate the effectiveness of this algorithm, dynamic experiment was conducted, yielding results that demonstrate horizontal positioning accuracy exceeding 1.9 meters and vertical positioning accuracy better than 7 metres. Compared with traditional positioning algorithms, the RTS-EKF exhibits at least a 15% improvement in localisation accuracy and superior performance, thereby satisfying the high-precision requirements of low-cost smartphone users.
  Keywords: Android smartphone; extended Kalman filter (EKF); Rauch-Tung-Striebel (RTS); Global Navigation Satellite System (GNSS); navigation and positioning.
  DOI: 10.1504/IJSNET.2025.10072253
   
  
• Reinforcement Learning Approach for Quality of Coverage-Driven Mobile Charger Optimal Scheduling in Wireless Rechargeable Sensor Networks  
  by Haoran Wang, Jinglin Li, Tianhang Chen, Wendong Xiao 
  Abstract: In wireless rechargeable sensor networks (WRSNs), mobile charger (MC) scheduling is one critical issue for improving network utility and resource utilisation efficiency. Traditional charging scheduling approaches usually focus on maximising charging utility while neglecting the importance of network service performance, especially quality of coverage (QoC). In practice, QoC directly affects the integrity of network information acquisition and its effectiveness and reliability. Therefore, the QoCdriven MC optimal scheduling (QCOS) problem is studied, and then a novel reinforcement learning-based mobile charger scheduling algorithm (RL-MCS) is proposed to maximize the network QoC and achieve stable network service performance. Meanwhile, a broadcast charging mechanism is also introduced to improve the overall charging efficiency and reduce the node charging time. In RL-MCS, the real-time energy demand of nodes and the network monitoring performance are considered, which aims to achieve the equilibrium between node survivability and network QoC. In addition, an experience extraction mechanism is designed, which enables MC to make smarter and more prospective charging decisions based on the current network state. Extensive simulations show that RL-MCS significantly outperforms other approaches in improving network QoC and ensuring node survival rate
  Keywords: Wireless rechargeable sensor networks; mobile charger scheduling; reinforcement learning; quality of coverage; broadcast charging.
  
  
• A data validity and energy efficiency sensitive method for in-node multi-parameter collaborative sensing in internet of things for agriculture  
  by Zhaokang Gong, Xiaomin Li, Rihong Zhang, Yongxin Liu 
  Abstract: In the agricultural IoT, there are problems such as low data value and high energy consumption. To this end, a cloud-edge intelligence-supported intra-node multi-parameter collaborative sensing strategy that is sensitive to data validity and energy efficiency is proposed. Firstly, a node intra-node data sensing framework supported by cloud-edge intelligence is proposed, a mathematical model of multi-parameter data sensing tasks is established, and evaluation indicators such as energy efficiency are designed. Secondly, an in-node multi-parameter data sensing based on correlation is proposed. The correlation between tasks and parameters is analysed using grey correlation to quantify the data value and validity of parameters. Using edge intelligence technology, the intra-node sensors are optimised with high value data as the target. Simulation experiments show that this method is superior to traditional methods in data value density, data validity rate and energy efficiency, with improvements of 100%, 20.59% and 300% respectively.
  Keywords: multi-parameter collaborative sensing; data validity; energy efficiency; internet of things; IoT.
  DOI: 10.1504/IJSNET.2025.10070952
   
  
• A feature-driven approach for ADAS using real-time smartphone inertial sensor data  
  by Sakshi , M.P.S. Bhatia, Pinaki Chakraborty 
  Abstract: With rapid advancement of advanced driver assistance systems (ADAS) and autonomous vehicles, ensuring safe interaction between human drivers and ADAS system has become a critical challenge for accurately predicting driver behaviour. The study addresses the problem of developing robust systems, with limited raw features from smartphone inertial measurement unit (IMU) sensors. This study proposes a feature-driven approach to predict driver behaviour using smartphone IMU data. The objective is to enhance prediction accuracy by utilising two novel features namely, statistical deep features and cross-correlated features. These features capture more intrinsic patterns in driver behaviour and improve the performance. The proposed methodology was experimented on three benchmark datasets with machine learning models like random forest (RF) and extreme gradient boosting (XGBoost). Using cross-correlated features, accuracies of 99%, 100%, and 99% were obtained. This demonstrates that this approach outperforms existing methods, capturing detailed patterns and providing more reliable predictions in complex traffic.
  Keywords: driving behaviour; deep learning; signal data; autonomous driving; feature engineering; inertial sensors; advanced driver assistance systems; ADAS; inertial measurement unit; IMU.
  DOI: 10.1504/IJSNET.2025.10071148
   
  
• Enhancing intrusion detection through anomaly detection and integrated deep learning with TabTransformer  
  by Omar Al-Harbi 
  Abstract: Intrusion detection systems (IDSs) are crucial for safeguarding network infrastructures as cybersecurity threats rapidly evolve, necessitating effective detection and response mechanisms. This paper introduces an intrusion detection model that combines anomaly detection with deep learning techniques and the TabTransformer architecture. Anomaly detection performs binary classification, identifying deviations from normal behaviour and enriching the feature set with binary predictions. For multiclass classification, the TabTransformer processes categorical data efficiently, while the convolutional neural network (CNN) and long short-term memory (LSTM) extract patterns from continuous data. Evaluations on benchmark datasets demonstrate that the proposed method surpasses baseline models, achieving superior accuracy, recall rate, and false positive rate.
  Keywords: intrusion detection; tabtransfomer; anomaly detection; deep learning; convolutional neural networ; CNN; long short-term memor; LSTM.
  DOI: 10.1504/IJSNET.2025.10071467
   
  
• Hybrid feature extraction and enhanced intrusion detection classification in industrial control networks  
  by Hanlin Chen, Entie Qi, Jin Si, Hui Yan, Tong Zhou 
  Abstract: In the rapidly developing industrial ecosystem, more and more malicious security attacks against industrial control come one after another. To address growing threats effectively, intrusion detection is essential in the multi-layer defence of communication networks. It helps prevent network attacks, policy violations, unauthorised access, and other security issues. It is very important to integrate this technology into the industrial control network. For anomaly identification, deep inspection of packets is required to extract appropriate features to identify attacks. Data usage and demand in industrial control networks are increasing daily; accurate anomaly detection with low data testing and training time is still challenging. This paper uses a hybrid feature extraction model consisting of a chi-square test, autoencoder, and principal component analysis. This paper presents a hybrid feature extraction-based intrusion detection model enhanced by a deep neural network. It is utilised to classify and identify various attack types within the non-intrusive learning packet dataset using the KDD dataset for industrial control networks, enabling an effective evaluation of intrusion detection performance. Experimental results demonstrate an average accuracy of 97.54%, a precision of 97.30%, and an F1-score of 94.01%. The model has better performance.
  Keywords: industrial control network; deep neural network; DNN; feature extraction; chi-square test; autoencoder; principal component analysis; PCA; intrusion detection.
  DOI: 10.1504/IJSNET.2025.10069603
   
  
• FSFDS: enhancing flight sensor fault diagnosis via diffusion and self-attention networks  
  by Jiaojiao Gu, Ping Gao, Xue Li, Bei Hong, Tao Sun 
  Abstract: Aircraft fault diagnosis relies on flight sensor data, with accuracy critical for safe operation. Deep neural networks (DNNs) have shown success in many fields, but their application in fault prediction faces challenges: 1) DNN-generated fault data differs significantly from real faults; 2) existing DNN-based classification models exhibit suboptimal accuracy. This paper proposes a flight sensor fault diagnosis system (FSFDS) using diffusion and attention networks. We enhance data quality with a diffusion model, introducing a scoring function for improved score matching. The generated data undergoes manual annotation and trains an attention-based diagnostic model with a weight-sharing multi-twin neural network to increase training samples. The attention mechanism extracts parameter relationships from time series, boosting accuracy. Deploying the model on an FPGA achieves high energy efficiency. Experiments show FSFDS improves diagnostic accuracy by 15.3% and speeds inference 14.23× over CPUs and 2.08 × over GPUs.
  Keywords: flight sensor; diagnosis system; diffusion; self-attention; field programmable gate array; FPGA; deep neural network; DNN.
  DOI: 10.1504/IJSNET.2025.10069837