In the Internet of Things (IoT) environment, to maintain the global consistency and generalization capability of federated learning (FL) models with good data privacy, a personalized FL algorithm is proposed based on a composite feature extractor (CFE), i.e., FedCFE. The FedCFE framework consists of three core modules: a CFE, a conditional generator, and a knowledge distillation mechanism. By modularly combining global with local feature extractors, the proposed CFE enables device-specific personalization without disrupting global knowledge sharing. Meanwhile, the representative pseudo-data is produced by a conditional generator, which facilitates effective knowledge distillation from client models to the server model without requiring direct access to private data. Extensive experimental results demonstrate that the FedCFE consistently achieves competitive or superior performance across various heterogeneous data scenarios. On the highly non-IID CIFAR-100 dataset, the FedCFE improves classification accuracy by 2.19% compared with the state-of-the-art algorithms. These results indicate that the FedCFE provides an effective, lightweight, and privacy-preserving solution to realizing collaborative intelligence in resource-constrained and communication-unstable IoT environments.

As the application of image emotion recognition technology grows increasingly widespread, emotional data faces potential privacy risks during transmission and storage. Images depicting negative emotions are particularly prone to revealing an individual is psychological state and sensitive information. Therefore, effective security protection for such images is of practical importance. This paper design a chaotic system and use CNN-based recognition to select the images requiring enhanced protection. First, a novel second-order memristor is designed and coupled with a neuron model to construct a chaotic system (SOM-Rulkov). Analysis of its phase diagram, bifurcation diagram, and Lyapunov exponent spectrum indicates that SOM-Rulkov exhibits rich dynamical characteristics, which can provide a pseudo-random key stream with good performance for encryption algorithms. Then, using a CNN to recognise emotions in images, employing the identified negative emotion images as encryption images to enhance data security during transmission and storage. During the encryption process, This paper proposes an enhanced diffusion structure with a parallel pool mechanism that enables four-directional diffusion to improve encryption efficiency. Experimental results show that the proposed scheme achieves strong security and high speed for emotional image privacy protection.

To address the high computational costs of full-frame encryption and the risk of exposing sensitive locations in partial encryption, this paper proposes a video selective encryption and steganography scheme based on object detection and image inpainting. First, YOLOv8 is employed to achieve real-time and accurate detection of human targets in video frames. Then, the LIS-HMC hyperchaotic map and a new chaotic-driven interframe chain modulation (CDICM) strategy, combined with a designed row-column interchange and Roller confusion algorithm, are applied to selectively encrypt the target regions. Next, the globally and locally consistent image completion (GLCIC) algorithm is used to restore the background panoramically, eliminating visual discontinuities. Meanwhile, based on the Walsh-Hadamard transform (WHT), a multi-round embedding (MRE) steganography strategy is developed to hide the encrypted information within the restored background. Experimental results show that the encrypted data achieve an information entropy of 7.9925, a steganographic capacity of 0.75 bpp, and a PSNR above 44.91 dB after data embedding, demonstrating that the proposed method provides a new solution for video privacy protection that balances security, real-time performance, and visual naturalness.

Orbital angular momentum (OAM) technology can provide an additional degree of freedom in the spatial domain and solve the problem of spectrum shortage in the sixth generation (6G) networks. In this paper, a near-field OAM channel model based on the electromagnetic information (EMI) theory is proposed, in which the dyadic Green’s function approach and discrete Fourier basis functions are utilized to accurately describe the characteristics of near-field channel and OAM signal propagation in practical scenarios, respectively. To improve the performance of misaligned transmission, a near-field OAM misalignment precoding scheme is presented according to the characteristics of the proposed channel model. Specifically, phase errors are compensated and the optimal power is assigned according to the channel condition, which can mitigate the channel capacity loss caused by the the inter-mode interference resulting from the misalignment. Numerical results show that the benefit brought by the polarization along the propagation direction is predominantly significant at short distances, and decreases rapidly to zero as the distance increases. Results also show that the presented scheme outperforms the conventional fixed parameter scheme and effectively reduces the effects of the misalignment in practical scenarios.

Face normalization aims to enhance the robustness and effectiveness of face recognition systems by mitigating intra-personal variations in expressions, poses, occlusions, illuminations, and domains. Existing methods face limitations in handling multiple variations and adapting to cross-domain scenarios. To address these challenges, we propose a novel Unified Multi-Domain Face Normalization Network (UMFN) model, which can process face images with various types of facial variations from different domains, and reconstruct frontal, neutral-expression facial prototypes in the target domain. As an unsupervised domain adaptation model, UMFN facilitates concurrent training on multiple datasets across domains and demonstrates strong prototype reconstruction capabilities. Notably, UMFN serves as a joint prototype and feature learning framework, enabling the simultaneous extraction of domain-agnostic identity features through a decoupling mapping network and a feature domain classifier for adversarial training. Moreover, we design an efficient Heterogeneous Face Recognition (HFR) network that fuses domain-agnostic and identity-discriminative features for HFR, and introduce contrastive learning to enhance identity recognition accuracy. Empirical studies on diverse cross-domain face datasets validate the effectiveness of our proposed method.

Considering the extremely complex physiological environment within neurons, there is feedback from autapse currents as well as the influence of external electromagnetic radiation. In this paper, a second-order memristor is constructed based on the definition of a generic memristor, which two intermediate variables are used to simultaneously model the effects of electromagnetic radiation and autapse on Rulkov neuron, called SOM-Rulkov neuron. The analysis of Lyapunov Exponent spectrum(LEs), bifurcation diagrams, phase diagrams, and iterative diagrams with different parameters that SOM-Rulkov has various types of periodic and chaotic firing patterns and high complexity. In particular, homogeneous extreme multistability is demonstrated with different initial conditions, and the phenomenon is more suitable for image encryption. Furthermore, the SOM-Rulkov map is implemented on the DSP platform. Finally, the SOM-Rulkov map is applied to encrypt the 3D model, which is essentially a sequence generated by homogeneous multistability and the vertex coordinates of the 3D model for different position xor operations. When decrypting, users with different levels of keys can access different visualizations. Experiments show that the scheme has strong security and low time cost.

With the rapid development of the Internet, face recognition technology is widely used, which makes the protection of face database especially important. To protect the recognized faces, a multiface image compression encryption (MFICE) scheme is designed based on the electromagnetic radiation Ktz neuron (ERKN). Since only faces are to be encrypted, they are first extracted. Then the face images are compressed by using semi-tensor product compressed sensing (STP-CS) algorithm, and the compressed images are integrated into a large cube, i.e., a 3-D cube. After that, interface confusion algorithm, 3-D shuffling algorithm, and 3-D diffusion algorithm are sequently performed by using chaotic sequences generated by iteration of ERKN, and finally the ciphertext image cube is obtained. The proposed scheme is evaluated, and it performs well in terms of feasibility and security.

With the popularity of surveillance devices, the security of surveillance video has attracted much attention, and three key issues need to be solved. The videos cannot be synchronized with their encryption effects, while full encryption does not meet the current trend of lightweight algorithms, and customized encryption for multiple specific targets is rarely seen. Inspired by this, a lightweight batch video frame awareness multitarget encryption scheme based on a novel discrete Tabu learning neuron (DTLN) and YoloV5 is designed in this article, the DTLN is in hyperchaotic state within a great range of parameters, which ensures the diversity of key selection and security. At the same time, the coexistence of homogeneous attractors is found, and such attractors are difficult to be successfully recognized by the parameter recognition algorithm, which increases the difficulty for the attacker to obtain the key. The mosaic tracking scheme designed by YoloV5 network can lightweightly encrypt batch frames of multiple types and targets, and users can also customize the encryption targets according to their needs. The simulation results show that the encryption scheme can realize lightweight encryption of multitype and multitarget, and performs well in all the security performance indexes, and has certain advantages compared with other video encryption schemes in terms of performance and functionality.

Face normalization is a critical technique for improving the robustness and generalizability of face recognition systems by reducing intra-personal variations arising from expressions, poses, occlusions, illuminations, and domain shifts. Existing normalization methods, however, often lack the flexibility to handle multi-factorial variations and exhibit limited cross-domain adaptability. To address these challenges, we propose a Unified Multi-Domain Face Normalization Network (UMFN), which is designed to process facial images with diverse variations from various domains and reconstruct frontal, neutralized facial prototypes in the target domain. As an unsupervised domain adaptation model, the UMFN facilitates concurrent training across multiple cross-domain datasets and demonstrates robust prototype reconstruction capabilities. Notably, the UMFN functions as a joint prototype and feature learning framework, extracting domain-agnostic identity features through a decoupling mapping network and adversarial training with a feature domain classifier. Furthermore, we design an efficient Heterogeneous Face Recognition (HFR) network that integrates these domain-agnostic features and the identity-discriminative features extracted from normalized prototypes, enhanced by contrastive learning to improve identity recognition accuracy. Empirical evaluation on multiple cross-domain benchmark datasets validate the effectiveness of the UMFN for face normalization and the superiority of the HFR network for heterogeneous face recognition.

This paper focuses on a new heterogeneous prototype learning (HPL) problem, which aims at reconstructing the variation-free and identity-preserved prototype in the target domain from a contaminated input image in the source domain. Most existing heterogeneous face synthesis (HFS) methods are unsuitable for HPL, as these methods focus on performing accurate image-to-image translation with facial details unaltered, but cannot effectively remove the input facial variations. In this paper, we propose an identity-aware cycle-consistent network, dubbed IAC2N, for image-to-prototype transformation across domains. To address HPL, IAC2N designs three effective losses, i.e., prototype adversarial loss, label information guided identity loss, and prototype learning cycle loss, in its objective. The first loss is used for transferring the domain style as well as removing the universal facial variations. The latter two losses are used for maintaining the identity consistency during HPL from an explicit and an implicit perspectives, respectively. Furthermore, IAC2N is a joint learning framework that is able to learn the identity feature for the contaminated image via its encoder-decoder structural generator in order to perform heterogeneous face recognition (HFR). Extensive experiments on various heterogeneous face datasets demonstrate the effectiveness of IAC2N in both tasks of HPL and HFR.

In this article, a novel locally active memristor (LAM) model is designed and its characteristics are studied in detail. Then, the LAM model is applied to couple FitzHugh-Nagumo (FHN) and Hindmarsh-Rose (HR) neuron. The simple neuron network is built to emulate connection of separate neurons and transmission of information from FHN neuron to HR neuron. The equilibrium point about this FHN-HR model is analyzed. Under the influence of varied parameters, dynamical characteristics for the model are explored with various analysis methods, including phase diagram, time series, bifurcation diagram, and Lyapunov exponent spectrum (LEs). The spectral entropy (SE) complexity and sequence randomness of the model are studied. In addition to observing chaotic and periodic attractors, multiple types of attractor coexistence and particular state transition phenomena are also found in the coupled FHN-HR model. Furthermore, geometric control is used for modulating the amplitude and offset of attractor and neuron firing signals, involving amplitude control and offset control. Finally, DSP implementation is finished, proving digital circuit feasibility of the FHN-HR model. The research imitates the coupling and information transmission between different neurons and has potential applications to secrecy or encryption.

With the rapid development of the internet worldwide, network security issues are becoming increasingly prominent. Network intrusion detection systems (NIDSs) play a vital role in ensuring computer network security due to their ability to identify potential network threats. Despite considerable research efforts, deploying NIDSs on resource-constrained devices has been challenging. To reduce the imposed computational cost and model storage requirements, in this paper, we propose a novel lightweight NIDS model. In this model, patterns of normal and malicious actions are learned via a finite Dirichlet mixture model (DMM) in the context of the extended stochastic variational inference (ESVI) framework. With the proposed method, both the parameter estimation and model selection processes can be simultaneously addressed in a unified Bayesian framework. A great number of experiments conducted on three publicly available datasets demonstrate that the proposed model not only achieves comparable classification performance to that of detection models based on several well-studied finite mixture modeling, traditional machine learning (ML) and promising deep learning (DL) algorithms but also significantly reduces the required training and detection time. Extensive experimental results validate that the proposed model is a feasible and efficient lightweight intrusion detection model.

Abstract The secrecy rate of two‐way untrusted relay networks with imperfect channel state information based on SWIPT is investigated when multiple relays harvest energy from two sources. Despite assisting in information forwarding, the relays are considered untrusted in that they might attempt to eavesdrop on confidential information. To interfere with eavesdropping by untrusted relays, sources‐based friendly cooperative jamming is introduced. A joint power allocation and time switching strategy has been studied to maximize the sum secrecy rate of the system under total block transmission duration and power constraints. The genetic algorithm (GA) is introduced to optimize the proposed joint power allocation and time switching (JPTs) scheme. In addition, two suboptimal schemes are proposed: the power allocation and time switching scheme for individual power constraints and the power allocation scheme for equal time allocation. Simulation results demonstrate that the proposed joint power allocation and time switching strategy performs better in contributing to the total secrecy rate of the system.

A new image encryption scheme is proposed with a combination of block scrambling, closed-loop diffusion, and DNA molecular mutation. The new chaotic block scrambling mechanism is put forward to replace the traditional swapping rule by combining the rectangular-ambulatory-plane cyclic shift with the bidirectional random disorganization. The closed-loop diffusion strategy is designed to form a feedback system, which improves the anti-interference capacity of the algorithm. To further destroy the blocks characteristics and eliminate the correlations among adjacent blocks, two efficient methods of DNA molecular mutation are adopted in the mutation stage. Moreover, the proposed algorithm possesses a large key space and the keys are highly related with the plaintext image. Experimental results demonstrate that the suggested image encryption strategy is practicable and has strong ability against a variety of common attacks.

A nonlinear multi-image encryption scheme is proposed by combining the reality-preserving discrete fractional angular transform with the deoxyribonucleic acid sequence operations. Four approximation coefficients of the four images are extracted by performing the two-dimensional lifting wavelet transform. Then, the four approximation coefficients are synthesized to generate a real-valued output with the reality-preserving discrete fractional angular transform. Finally, based on the deoxyribonucleic acid operation and the Logistic-sine system, the real-valued intermedium output will be encrypted to yield the final ciphertext image. To enhance the security of the image encryption algorithm, the initial value of the chaotic system is calculated by the 256-bit binary sequence, which is obtained by taking the statistics information of the plaintext images as the input of SHA-256. Deoxyribonucleic acid sequence operations, as nonlinear processes, could help to improve the robustness of the cryptosystem. Simulation results and security analysis demonstrate the effectiveness of the image encryption algorithm and the capability of withstanding various common attacks.

The secrecy performance of a wireless‐powered relaying system is investigated in the presence of imperfect channel state information. To improve the system secrecy performance, the destination splits a part of power to transmit the cooperative interference signal while the source transmits confidential information to the relay. The relay is assumed to be equipped with a non‐linear energy harvester and to harvest energy from the source and destination for information processing. A joint power splitting and time allocation scheme is studied to maximise the system secrecy rate. The joint optimisation problem is dealt with by the interior‐point method. For the sake of comparison, the equal time allocation scheme and the equal power splitting scheme are also investigated. The numerical results verify that the proposed scheme could achieve higher secrecy rate.

Abstract Cooperative communication can improve the performance of communication system under multi‐path fading conditions by relaying. If there is an eavesdropper in communication system, the secrecy capacity of the system will decrease. Sending artificial noise can enhance the secrecy capacity of communication system. A novel scheme combining relay selection with artificial noise for amplify‐and‐forward cooperative communication system in the presence of an eavesdropper is designed, which seeks the relay with the highest signal‐to‐noise ratio. The location of optimal relay node is found out in this scheme by the relay selection scheme based on distance. However, there is not always a relay node at the optimal location. If there is no relay at the optimal location, then the suboptimal relay node can be found out by drawing circles centered on the optimal location. After that, the optimal relay forwards the signals and sends the artificial noise in the null space of the legitimate channel to confuse the eavesdropper. A close‐form expression for maximizing the secrecy capacity is derived, and it is used as the objective function to select the optimal or suboptimal relay node. The algorithm complexity of the relay selection scheme based on distance is lower than that of the relay selection scheme based on instantaneous channel states. Moreover, it can achieve higher secrecy capacity compared with the scheme without artificial noise. Simulations are conducted to validate the theoretical analyses, and the results demonstrate the validity and reliable security of the scheme. Copyright © 2016 John Wiley & Sons, Ltd.

Hash functions are often used to protect the integrity of information. In general, the design of hash functions should satisfy three standards: pre-image resistance, second pre-image resistance and collision resistance. The design of hash functions in the literature assumes that the messages to be transmitted are equally probable. In this paper, we focus on the pre-image resistance and investigate the problem of mapping design for hash function utilizing the unequal occurrence probabilities of the messages. We first present a necessary condition for the optimal mapping and then introduce a heuristic algorithm. Simulation experiments are carried out to evaluate the performance of the proposed new design. It is shown that the probability of successful attack can be significantly reduced compared with the conventional design. Our algorithm can be useful in scenarios where the attacker has limited ability or time to estimate the probability distribution of the messages. To our best knowledge, this work is the first attempt of making use of the message distribution in designing hash functions for information security.

We present a new method for image encryption with discrete factional cosine transform (DFrCT) and chaos. DFrCT holds the particular properties which the conventional discrete cosine transform (DCT) hasnpsilat, that is its fraction. Chaos functions have extreme sensitivity to the initial conditions. Logistic map is a simple equation of chaos functions. XOR is first operated between the original image and logistic map. Then the chaotic image is transformed with DFrCT two times using different keys successively by rows and by columns. Based on this method, the image can encrypted effectively, also, the transmission of the encrypted image with DFrCT and chaos is faster than with fractional Fourier transform (DFrFT) and chaos. The computer simulations are presented to verify the validity of the proposed method, such as the mean square error (MSE) between the original images and the decrypted images.