Low Computational Overhead Research Articles

Dehaze-UNet: A Lightweight Network Based on UNet for Single-Image Dehazing

Numerous extant image dehazing methods based on learning improve performance by increasing the depth or width, the size of the convolution kernel, or using the Transformer structure. However, this will inevitably introduce many parameters and increase the computational overhead. Therefore, we propose a lightweight dehazing framework: Dehaze-UNet, which has excellent dehazing performance and very low computational overhead to be suitable for terminal deployment. To allow Dehaze-UNet to aggregate the features of haze, we design a LAYER module. This module mainly aggregates the haze features of different hazy images through the batch normalization layer, so that Dehaze-UNet can pay more attention to haze. Furthermore, we revisit the use of the physical model in the network. We design an ASMFUN module to operate the feature map of the network, allowing the network to better understand the generation and removal of haze and learn prior knowledge to improve the network’s generalization to real hazy scenes. Extensive experimental results indicate that the lightweight Dehaze-UNet outperforms state-of-the-art methods, especially for hazy images of real scenes.

AHA-BV: Access and handover authentication protocol with batch verification for satellite–terrestrial integrated networks

At present, the rapid development of satellite capabilities has prompted the proposal of satellite–terrestrial integrated networks (STIN), which solves the problem of limited signal coverage of terrestrial cellular networks, further promotes the globalization process, and realizes global data sharing and on-demand use. However, due to the high openness of satellite-to-ground links in STIN, users are vulnerable to attacks such as eavesdropping, replay, tampering, and impersonation when requesting access to satellite nodes and obtaining subscription services. To ensure the security and reliability, many authentication protocols have been proposed, but there are still some shortcomings, such as high authentication overhead, vulnerability to certain attacks. In addition, for inter-satellite handovers caused by the highly dynamic topology of satellites, the computational overhead of existing handover authentication mechanisms is too high to be applied to frequent inter-satellite handover scenarios in STIN. To address the above issues, in this paper, we propose a new access and handover authentication protocol with batch verification for STIN, namely the AHA-BV protocol. The AHA-BV protocol not only realizes mutual authentication and key negotiation between users and satellite access points without the participation of the network control center, but also ensures the conditional anonymity of users during the access authentication phase. Furthermore, the lightweight batch verification mechanism reduces the risk of computing bottlenecks when resource-constrained satellites receive a large number of access authentication requests. Not only that, the AHA-BV protocol can also achieve sustained trust in subscription services from STIN with low computational overhead during the inter-satellite handover authentication phase. Formal and heuristic security analysis show that the AHA-BV protocol can meet the security requirements of STIN. Performance analysis indicates that the AHA-BV protocol has low authentication overhead while ensuring security, and is more suitable for users under satellite dynamic topology to access and obtain subscription services from STIN.

Blockchain-Based Unbalanced PSI with Public Verification and Financial Security

Private set intersection (PSI) enables two parties to determine the intersection of their respective datasets without revealing any information beyond the intersection itself. This paper particularly focuses on the scenario of unbalanced PSI, where the sizes of datasets possessed by the parties can significantly differ. Current protocols for unbalanced PSI under the malicious security model exhibit low efficiency, rendering them impractical in real-world applications. By contrast, most efficient unbalanced PSI protocols fail to guarantee the correctness of the intersection against a malicious server and cannot even ensure the client’s privacy. The present study proposes a blockchain-based unbalanced PSI protocol with public verification and financial security that enables the client to detect malicious behavior from the server (if any) and then generate an irrefutable and publicly verifiable proof without compromising its secret. The proof can be verified through smart contracts, and some economic incentive and penalty measures are executed automatically to achieve financial security. Furthermore, we implement the proposed protocol, and experimental results demonstrate that our scheme exhibits low online communication complexity and computational overhead for the client. At the same time, the size of the generated proof and its verification complexity are both O(logn), enabling cost-effective validation on the blockchain.

Efficient inter partitioning of versatile video coding based on supervised contrastive learning

Nowadays, Versatile Video Coding (VVC) has achieved a superior performance than previous video coding standard (High Efficiency Video Coding). The Quadtree with Nested Multi-Type Tree (QTMT) coding block structure can enhance the coding performance. Nevertheless, this technique also leads to the significantly increasing complexity of VVC inter coding. Therefore, complexity optimization is an urgent problem to be optimized in the market application of VVC. To solve this issue, we propose a Supervised-Contrastive-Learning-based Inter Partitioning (SCLIP) method in this paper. Firstly, we define the above complexity optimization problem as a supervised classification task. Next, we develop a SCLIP Estimation Network (SCLIPEst-Net) with a supervised contrastive learning module and a classification module. After training on a newly established dataset, the SCLIPEst-Net can reasonably predict the mode partitioning. Finally, we propose an overall SCLIP algorithm that effectively determines the inter partitions of VVC with a low computational overhead. Experimental results indicate that our method achieves 45.14% average Time Saving (TS) with a 2.40% Bjøntegaard Delta Bit Rate (BDBR) in Random Access (RA), outperforming the benchmarks.

Deep reinforcement learning for multi-objective game strategy selection

Multi-objective game (MOG) is a fundamental model for the decision-making problems in which each player must consider multi-dimensional payoffs that reflect different objectives. Typically, solving MOG involves refining the set of equilibrium strategies, which is also known as MOG strategy selection (MOGS). However, existing MOG algorithms only allow one metric for MOGS, which limits the application in real-world scenarios where the players may have different preferences over multiple metrics. In this paper, we first develop a preference-based MOGS framework to encompass multiple metrics with different preferences in MOGS. Based on the framework, we introduce the concept of comprehensive evaluation value (CEV) to evaluate the quality of a strategy set given the preference of each metric. Using CEV as a reward signal, we formulate the problem of finding the optimal strategy set as a Markov decision process, and use deep reinforcement learning to train a policy for MOG strategy selection given the metrics and the corresponding preferences. Specifically, we combine a rational strategy filtering procedure with a Transformer-based encoder–decoder policy network to refine the strategies given the preferences, and then we use a revised REINFORCE algorithm to train the policy network. Besides, we introduce variable beam search decoding to improve the quality of a rollout by keeping track of the most promising strategy sets and choosing the best one. We benchmark our algorithm on the MOG instances generated by GAMUT, and extensive experiments demonstrate that our algorithm can generate the strategy set significantly better than the state-of-the-art baselines with lower computational overhead given different preferences. Furthermore, we compare our approach on real-world problems, showing the great advantages in both performance and runtime.

Efficient Cyber Security Framework for IoT Using Machine Learning Algorithms

Through the network's infrastructure, the IoT can impart perception, recognition, and remote control to inanimate objects. Due to IoT's characteristics, it's possible to integrate the real world into a digital system, which improves precision, productivity, and bottom line. While traditional internet infrastructure comprises of highly capable computers and servers, IoT gadgets have limited processing power and storage space. The authentication and key agreement system SecureAuthKey is very lightweight. The suggested technique is meant to solve the security and privacy problems that plague modern constraint-based CPS programmes. A lightweight approach for authenticating cyber-physical objects is one of the expected outcomes. Adaptation and autonomy in cyber-physical systems need not be compromised by a lack of trust or security in users' data or the devices themselves, according to a new type of security algorithm. To provide flexibility and scalability, a new middleware module has been developed on the Raspberry platform to facilitate communication between CPS-based devices and services. Secure Internet of Things (IoT) evaluation methodology that may be used in a variety of contexts and is user-focused. With the suggested system, the two CPS units will be able to authenticate one another. When a network grows larger, the possibility of an attack rises. Therefore, the IoT network is much more susceptible to attacks than conventional networks. As the number of connected devices grows, so do the number of potential threats to their security. To protect the IoT ecosystem from current threats, cutting-edge technology is essential. The proposed approach must be very efficient and have low computational overheads. It creates random session keys to protect wireless transmissions. The system is protected from a variety of cyber threats. The current constraint-based CPS system requires a new lightweight security solution.

EPFL-DAC: Enhancing Privacy in Federated Learning with Dynamic Aggregation and Clipping

Federated learning achieves data privacy protection by uploading only users’ local updates. However, recent research has shown that privacy data can still be inferred from local updates. Privacy-preserving federated learning also faces challenges related to the confidentiality and privacy of local updates, the robustness of dynamic membership schemes, and the determination of appropriate clipping threshold in differentially private federated learning. To address these issues, this paper presents the EPFL-DAC scheme: Enhancing Privacy in Federated Learning with Dynamic Aggregation and Clipping. Firstly, Paillier homomorphic encryption is employed to securely aggregate local updates, ensuring that the server can only access the aggregated information. Secondly, a dynamic robustness algorithm is designed to ensure the robustness of the aggregation scheme when user states change. Lastly, a dynamic threshold determination method is introduced to introduce more accurate noise, while also providing resistance against collusive attacks in differentially private settings. Experimental results demonstrate the effectiveness, low computational overhead, and preservation of accuracy in the dynamic aggregation scheme. Additionally, the proposed dynamic threshold determination method exhibits good performance, especially at higher privacy protection levels and larger initial clipping threshold values.

Cross-Attention Enhanced Pyramid Multi-Scale Networks for Sensor-Based Human Activity Recognition.

Human Activity Recognition (HAR) has recently attracted widespread attention, with the effective application of this technology helping people in areas such as healthcare, smart homes, and gait analysis. Deep learning methods have shown remarkable performance in HAR. A pivotal challenge is the trade-off between recognition accuracy and computational efficiency, especially in resource-constrained mobile devices. This challenge necessitates the development of models that enhance feature representation capabilities without imposing additional computational burdens. Addressing this, we introduce a novel HAR model leveraging deep learning, ingeniously designed to navigate the accuracy-efficiency trade-off. The model comprises two innovative modules: 1) Pyramid Multi-scale Convolutional Network (PMCN), which is designed with a symmetric structure and is capable of obtaining a rich receptive field at a finer level through its multiscale representation capability; 2) Cross-Attention Mechanism, which establishes interrelationships among sensor dimensions, temporal dimensions, and channel dimensions, and effectively enhances useful information while suppressing irrelevant data. The proposed model is rigorously evaluated across four diverse datasets: UCI, WISDM, PAMAP2, and OPPORTUNITY. Additional ablation and comparative studies are conducted to comprehensively assess the performance of the model. Experimental results demonstrate that the proposed model achieves superior activity recognition accuracy while maintaining low computational overhead.

An Efficient Lightweight Authentication Scheme for Smart Meter

With the rapid development of the information age, smart meters play an important role in the smart grid. However, there are more and more attacks on smart meters, which mainly focus on the identity authentication of smart meters and the security protection of electricity consumption data. In this paper, an efficient lightweight smart meter authentication scheme is proposed based on the Chinese Remainder Theorem (CRT), which can realize the revocation of a single smart meter user by publishing a secret random value bound to the smart meter identity. The proposed scheme not only protects the security of smart meter electricity consumption data by using encryption, but also resists identity attacks from both internal and external adversaries by using hash functions and timestamps. Experiment shows that the proposed scheme has lower computation overhead and communication overhead than other authentication schemes and is more suitable for smart meter authentication.

Improved YOLOv8 Model for Lightweight Pigeon Egg Detection.

In response to the high breakage rate of pigeon eggs and the significant labor costs associated with egg-producing pigeon farming, this study proposes an improved YOLOv8-PG (real versus fake pigeon egg detection) model based on YOLOv8n. Specifically, the Bottleneck in the C2f module of the YOLOv8n backbone network and neck network are replaced with Fasternet-EMA Block and Fasternet Block, respectively. The Fasternet Block is designed based on PConv (Partial Convolution) to reduce model parameter count and computational load efficiently. Furthermore, the incorporation of the EMA (Efficient Multi-scale Attention) mechanism helps mitigate interference from complex environments on pigeon-egg feature-extraction capabilities. Additionally, Dysample, an ultra-lightweight and effective upsampler, is introduced into the neck network to further enhance performance with lower computational overhead. Finally, the EXPMA (exponential moving average) concept is employed to optimize the SlideLoss and propose the EMASlideLoss classification loss function, addressing the issue of imbalanced data samples and enhancing the model's robustness. The experimental results showed that the F1-score, mAP50-95, and mAP75 of YOLOv8-PG increased by 0.76%, 1.56%, and 4.45%, respectively, compared with the baseline YOLOv8n model. Moreover, the model's parameter count and computational load are reduced by 24.69% and 22.89%, respectively. Compared to detection models such as Faster R-CNN, YOLOv5s, YOLOv7, and YOLOv8s, YOLOv8-PG exhibits superior performance. Additionally, the reduction in parameter count and computational load contributes to lowering the model deployment costs and facilitates its implementation on mobile robotic platforms.

High generalization identification method based on MI-SI distributed optical fiber sensor and video signals

A pattern recognition method for intrusion signals is introduced, leveraging distributed optical fiber sensors and convolutional neural networks. To enhance the model's generalization capabilities, two strategies are employed. Firstly, the sensor structure is refined, incorporating the strengths of both the Sagnac and Michelson interferometers to create a hybrid design with a wide frequency response. Comparative experiments confirm that this exceptional frequency response significantly boosts the model's generalization power. Secondly, to address the limitations of distributed fiber optic sensors, a multimodal approach is proposed, integrating video frame difference and fiber optic pattern recognition. The trained convolutional neural network model exhibits low computational overhead, robust generalization, and practical applicability. When applied to a dataset collected in diverse scenarios not included in the model training, the model achieved a CNR of 0% and a FPR of only 0.26% for the intrusion behavior of climbing fences.

ESVFL: Efficient and secure verifiable federated learning with privacy-preserving

Federated learning has been widely applied as a distributed machine learning method in various fields, allowing a global model to be trained by sharing local gradients instead of raw data. However, direct sharing of local gradients still carries the risk of privacy data leakage, and the malicious server might falsify aggregated result to disrupt model updates. To address these issues, a lot of privacy-preserving and verifiable federated learning schemes have been proposed. However, existing schemes suffer from significant computation overhead in either encryption or verification. In this paper, we present ESVFL, an efficient and secure verifiable federated learning scheme with privacy-preserving. This scheme can simultaneously achieve low computation overhead for encryption and verification on the user side. We design an efficient privacy-preserving method to encrypt the users’ local gradients. Using this method, the computation and communication overheads of encryption on the user side is independent of the number of users. Users can efficiently verify the correctness of aggregated results returned by the cloud servers using cross-verification. During the verification process, there is no interaction among users and no additional computation is required. Furthermore, we also construct an efficient method to address the issue of user dropout. When some users drop out, online users do not incur any additional computation and communication overheads, while guaranteeing the correctness of the aggregated result of online users’ encrypted gradients. The security analysis and the performance evaluation demonstrate that ESVFL is secure and can achieve efficient encryption and verification.

Data enhancement method based on cyclic generation adversarial network

This study introduces a novel data augmentation technique employing Cycle Generative Adversarial Networks (CycleGAN) to mitigate the challenges posed by the paucity of image datasets in deep learning domains. Through the adept training of a CycleGAN model, this method substantially enriches image datasets, thereby enhancing the efficiency of deep learning models in target detection tasks. Distinct from conventional approaches, our strategy incorporates advanced activation functions, relative discriminators, and residual connections, which collectively foster greater image diversity and mitigate mode collapse, all while maintaining a low computational overhead. Evaluations conducted on diverse datasets, including MNIST, Synthetic Aperture Radar (SAR), and medical blood cell images, demonstrate the method's superior augmentation capabilities compared to traditional Deep Convolutional GAN (DCGAN) techniques, underscoring its efficacy and potential utility in preprocessing for deep learning applications.

Towards Large Certified Radius in Randomized Smoothing Using Quasiconcave Optimization

Randomized smoothing is currently the state-of-the-art method that provides certified robustness for deep neural networks. However, due to its excessively conservative nature, this method of incomplete verification often cannot achieve an adequate certified radius on real-world datasets. One way to obtain a larger certified radius is to use an input-specific algorithm instead of using a fixed Gaussian filter for all data points. Several methods based on this idea have been proposed, but they either suffer from high computational costs or gain marginal improvement in certified radius. In this work, we show that by exploiting the quasiconvex problem structure, we can find the optimal certified radii for most data points with slight computational overhead. This observation leads to an efficient and effective input-specific randomized smoothing algorithm. We conduct extensive experiments and empirical analysis on CIFAR-10 and ImageNet. The results show that the proposed method significantly enhances the certified radii with low computational overhead.

TurboBlom: A light and resilient key predistribution scheme with application to Internet of Things.

In the Internet of Things (IoT), there are often devices that are computationally too constrained to establish a security key using traditional key distribution mechanisms such as those based on the Diffie-Hellman key exchange. To address this, current solution commonly rely on key predistribution schemes (KPSs). Among KPSs, the Blom scheme provably provides the highest resilience against node capture attacks. This, however, comes at high computational overhead, because the Blom scheme requires many multiplications over a large finite field. To overcome this computational overhead, we present TurboBlom, a novel amendment of the Blom scheme. TurboBlom circumvents the need for field multiplications by utilizing specialized generator matrices, such as random zero-one matrices. We demonstrate that, through this approach, TurboBlom can significantly reduce the computational overhead of the Blom scheme by orders of magnitude. In our next key finding, we demonstrate that TurboBlom offers a level of resilience against node capture that is virtually on par with the Blom scheme. Notably, we prove that the gap between the resilience of the two schemes is exponentially small. These features of TurboBlom (i.e., low computational overhead and high resilience) make it suitable for computationally constrained devices. Such devices exist in abundance in IoT, for example, as part of Low Power and Lossy Networks (LLNs). To demonstrate a sample application of TurboBlom, we show how to use it to enable sender authentication in the Routing Protocol for LLNs (RPL), a standard routing protocol for IoT.

MLSTL-WSN: machine learning-based intrusion detection using SMOTETomek in WSNs

In the domain of cyber-physical systems, wireless sensor networks (WSNs) play a pivotal role as infrastructures, encompassing both stationary and mobile sensors. These sensors self-organize and establish multi-hop connections for communication, collectively sensing, gathering, processing, and transmitting data about their surroundings. Despite their significance, WSNs face rapid and detrimental attacks that can disrupt functionality. Existing intrusion detection methods for WSNs encounter challenges such as low detection rates, computational overhead, and false alarms. These issues stem from sensor node resource constraints, data redundancy, and high correlation within the network. To address these challenges, we propose an innovative intrusion detection approach that integrates machine learning (ML) techniques with the Synthetic Minority Oversampling Technique Tomek Link (SMOTE-TomekLink) algorithm. This blend synthesizes minority instances and eliminates Tomek links, resulting in a balanced dataset that significantly enhances detection accuracy in WSNs. Additionally, we incorporate feature scaling through standardization to render input features consistent and scalable, facilitating more precise training and detection. To counteract imbalanced WSN datasets, we employ the SMOTE-Tomek resampling technique, mitigating overfitting and underfitting issues. Our comprehensive evaluation, using the wireless sensor network dataset (WSN-DS) containing 374,661 records, identifies the optimal model for intrusion detection in WSNs. The standout outcome of our research is the remarkable performance of our model. In binary classification scenarios, it achieves an accuracy rate of 99.78%, and in multiclass classification scenarios, it attains an exceptional accuracy rate of 99.92%. These findings underscore the efficiency and superiority of our proposal in the context of WSN intrusion detection, showcasing its effectiveness in detecting and mitigating intrusions in WSNs.

DROPSYS: Detection of ROP attacks using system information

As modern return-oriented programming (ROP) attacks have become more sophisticated, preventing or detecting these attacks is essential for real-world exploit defense. As an alternative to many defense techniques of ROP attacks that require software modification and hardware assistance, researchers have proposed ROP defense techniques using hardware performance counters (HPCs) to mitigate concerns about additional cost and compatibility issues. However the existing HPC data-based ROP detection techniques typically suffer from low detection performance mainly because of the non-deterministic nature of HPCs.To address these issues, we propose DROPSYS, an enhanced detection of ROP attacks using system information. DROPSYS is based on the detection of the abnormal change of system information that takes place during ROP attacks. Differing from the existing techniques, DROPSYS harnesses not only HPC data, but also system utilization data to mitigate the non-deterministic nature of HPCs. Using both HPCs of processors and system utilization of operating systems makes transparent operation without requiring any modifications to the protected programs. DROPSYS uses a long short-term memory-based variational autoencoder to effectively analyze the multivariate time-series HPC data and system utilization data for better detection performance. DROPSYS also performs feature selection for low computational overhead while maintaining the attack detection performance.In our experiments with real-world ROP exploits, DROPSYS successfully detected ROP code execution in all tested programs. Evaluation results show that DROPSYS effectively captures the behaviors and effects of ROP attacks and can detect the attacks with a 0.028% false positive rate. The accuracy of DROPSYS is 95.3%, and its F1 score is 94.9%—a figure much higher than those of existing techniques that utilize only HPC data.

Toward social media forensics through development of iOS analyzers for evidence collection and analysis

SummarySocial media usage in mobile phones has increased substantially in recent times, and they are a critically important source of a forensics investigation. In this paper, we have developed Python‐based forensic analyzers that are integrated with the open‐source tool Autopsy. The proposed analyzers find forensic artifacts from the three most widely used social media messaging applications, that is, WhatsApp, Instagram, and Facebook Messenger. This research focuses on finding forensic artifacts stored by these social media applications on an iOS device. These analyzers extract data critical for a forensic investigation such as text messages, media attachments, sender and receiver details, timestamps, contact information, and other related forensics data from the full file system image of iOS devices. These Python‐based plugins extract the required data from the social media applications' databases and present the evidential artifacts in a human‐readable format. We integrated these analyzers into the Autopsy Forensics tool and showcased the gathered evidence so that investigators are capable to analyze the extracted information effortlessly. The data integrity is maintained by converting it into readable form without permanently altering the database format. The results prove that the proposed analyzers can successfully extract and analyze forensics data at a low computational overhead.

PREEN: An Aggregation Mechanism for Privacy-Preserving Smart-Metering Communications

The widespread use of home consumer eletronics (CEs) provides residents facility, assistance and comfort. Meanwhile, smart meters are deployed at households to optimize electricity supply and demand and improve electricity efficiency. These meters monitor, collect and aggregate the electricity consumption of the home CEs in a near-real-time way. However, such processes seriously invade consumer privacy, e.g., household occupancy and economic status. Many existing privacy-preserving schemes rely on the assumption that certain components can be trusted, which is hard to be ensured, and cannot resist against collusion attacks. In this study, we leverage differential privacy and a re-encryption model to produce a privacy-preserving aggregation mechanism, called PREEN, for smart-metering communications. PREEN simultaneously facilitates data aggregation, differential privacy, and protection against various attacks. Specifically, we adopt ElGamal algorithm to construct the re-encryption scheme and differential privacy to aggregate the consumption measurements of residential users at the gateways and control center, thereby enabling resistance against slashing privacy attacks. Strict and detailed security proofs as well as extensive performance evaluations are adopted to demonstrate that the proposed PREEN provides rigorous privacy preservation and confidentiality against diverse external and internal attacks, while remaining a comparable or even lower computational overhead and a negligible communication overhead and a reasonable accuracy in terms of the utility of differential privacy, compared to the state-of-the-art mechanisms.

Pairing-Free Certificate-Based Proxy Re-Encryption Plus Scheme for Secure Cloud Data Sharing

The popularity of secure cloud data sharing is on the rise, but it also comes with significant concerns about privacy violations and data tampering. While existing Proxy Re-Encryption (PRE) schemes effectively protect data in the cloud, challenges persist with certificate administration and key escrow. Moreover, the increasing number of users and prevalence of lightweight devices demand functional and cost-effective solutions. To address these issues, this paper presents a novel Pairing-free Certificate-Based Proxy Re-Encryption Plus scheme that leverages elliptic curve groups for improved effectiveness and performance. This scheme successfully resolves challenges related to certificate management and key escrow in traditional PRE schemes, while also introducing non-transferable and message-level fine-grained control characteristics. These enhancements bolster data security during sharing and minimize the risk of malicious information leakage. Our proposed scheme’s correctness, security, and effectiveness are rigorously verified and analyzed. The results demonstrate that the scheme achieves the chosen ciphertext security in the random oracle model. Compared to current PRE schemes, our approach offers greater advantages, lower computational overhead, and enhanced suitability for practical cloud computing applications.