Existing chaotic image encryption schemes have made progress in improving keystream randomness and dynamic substitution; however, the coupling between the underlying chaotic dynamics, S-box construction, and the overall encryption architecture remains limited, especially for high-resolution color images. To address this issue, this paper proposes a color image encryption method that integrates a CML-ECA neurodynamic chaotic system with a TV-BST-based permutation-diffusion framework. Specifically, an ECA-CML composite chaotic model is first established, in which a local-entropy adaptive coupling strategy is combined with Wilson-Cowan excitatory-inhibitory neurodynamic feedback to modulate the Logistic-sine control parameters and lattice states, thereby generating spatiotemporal chaotic sequences with enhanced sensitivity, entropy, and finite-precision robustness. The resulting chaotic flow is then used to construct an initial S-box through a Fisher-Yates shuffle, which is further optimized by a genetic mechanism under joint cryptographic objectives to obtain a high-quality chaotic S-box. At the architecture level, the TV-BST module performs key-dependent row-column permutation for global scrambling, while each color channel undergoes forward and backward chained diffusion together with S-box substitution. Experimental results show that the proposed method produces nearly uniform histograms, low adjacent-pixel correlation, high key sensitivity, and NPCR/UACI values close to their theoretical expectations. In addition, the scheme exhibits robustness against common disturbances such as noise contamination and cropping attacks. These results indicate that the proposed co-designed framework provides an effective and secure solution for color image encryption.

Contact-free sensing based on WiFi channel state information (CSI) has shown considerable potential for human activity recognition and indoor localization. However, jointly addressing these two tasks remains challenging because raw CSI signals usually suffer from high-dimensional channel redundancy, task-irrelevant variations, and temporally entangled multi-scale fluctuations. To address these issues, this paper proposes a dual-task learning framework that emphasizes task-aligned subspace construction and structured temporal decomposition. Specifically, a Multi-task Aligned Re-ranking Subspace Principal Component Analysis (MARS-PCA) module is designed to re-rank principal components according to their discriminative relevance to both activity recognition and localization, thereby retaining a compact CSI representation that is more consistent with the dual-task objective. In addition, a multi-level wavelet decomposition front-end is introduced to separate CSI temporal responses into sub-band components, allowing transient activity-related dynamics and relatively stable location-related patterns to be represented more explicitly. The refined and decomposed features are then modeled by a lightweight temporal prediction module with channel-wise task regulation. Experiments on a public WiFi CSI dataset show that the proposed method achieves good performance in both activity recognition and indoor localization.

Underwater object detection for real-world robotic systems requires both strong robustness against severe visual degradation and efficient inference under resource-constrained edge deployment settings. However, existing methods often improve detection accuracy at the cost of increased computational complexity, making them difficult to deploy in practice. In this paper, we propose SvelteF3M-YOLO, a lightweight underwater object detector that jointly addresses degradation-aware representation and deployment efficiency within a unified framework. The model integrates a streamlined Frequency-Fused Feature Module (F3M) to enhance high-frequency structural cues while suppressing low-frequency interference, together with a hardware-friendly lightweight neck (SvelteNeck) to enable efficient multi-scale feature propagation. Extensive experiments on three representative underwater datasets, including SCoralDet, TrashCan-Instance, and URPC2019, show that SvelteF3M-YOLO maintains competitive mAP50–95 with reduced parameters and GFLOPs. Real-time deployment experiments on an NVIDIA Jetson Orin Nano Super 8GB further show that the proposed model achieves a mean latency of 4.52 ms and a throughput of 241.63 FPS under TensorRT FP16 inference, compared with 4.56 ms and 239.96 FPS for the YOLO11n baseline. Meanwhile, on the TrashCan-Instance dataset, SvelteF3M-YOLO improves mAP50–95 from 0.680 to 0.695 over YOLO11n. These results indicate that the proposed framework improves underwater detection accuracy while maintaining real-time edge deployment efficiency, providing a practical solution for resource-constrained underwater perception.

In order to overcome the problems of the bluish-green tone of color, bad contrast, and bad texture of underwater pictures, we introduced a two-step lightweight enhancement algorithm called WAWB-IDCP. The algorithm uses the wavelength-based white balance and an enhanced dark channel post-module, which contribute to the correct color correction and optimization of image quality, respectively. It solves the problem of color distortion and artifacts in blocks seen in traditional DCP algorithms. Experiments with multi-dimensional references were performed on three classic algorithms (Gray-world, CLAHE, DCP). The experimental results on the UIEB dataset prove that our algorithm has the highest subjective visual performance and also performs well on other quantitative measures, like the standard deviation of the algorithm of 44.71 and color cast control of 11.07. Furthermore, the SIFT feature experiment proves that it has a great capacity for recovering details and is also noise-resilient. The algorithm is highly performing and can be used in preprocessing underwater images.

Consumer electronic applications, including smart-home safety monitoring, non-contact indoor healthcare, indoor activity recognition, and context-aware location-related services, are increasingly expected to provide low-cost, privacy-friendly, and unobtrusive sensing capabilities for both user-state awareness and indoor location awareness. Wi-Fi Channel State Information (CSI) offers a promising sensing modality for such scenarios by exploiting existing wireless infrastructure without relying on cameras, wearable devices, or dedicated ranging hardware, thereby reducing hardware dependence and user burden. This paper proposes LBA-TCN, a lightweight task-aware architecture for joint activity recognition and indoor localization using Wi-Fi CSI. Rather than treating activity recognition and indoor localization as two isolated tasks, LBA-TCN learns a shared CSI representation through progressive evidence refinement, including local CSI pattern extraction, channel recalibration, temporal context modeling, and lightweight temporal evidence aggregation. The shared-backbone design enables activity-state and location-related predictions within a single forward pass, thereby avoiding duplicated feature extraction in separate single-task models and improving the accuracy–efficiency trade-off for edge-side consumer sensing. Experimental results show that LBA-TCN achieves competitive joint-sensing performance while maintaining a compact parameter scale. From a consumer-electronics perspective, the proposed framework provides a practical product-level advantage: a single compact Wi-Fi CSI model can support both activity awareness and indoor location awareness, reducing the need for additional sensing hardware and repeated model inference. These results indicate that, under the evaluated settings, LBA-TCN can serve as a practical foundation for consumer-electronics-oriented indoor sensing systems that require low-cost, less intrusive, and multi-functional perception capabilities.

Balancing positioning accuracy and deployment cost in indoor environments remains challenging. RSS-based Wi-Fi fingerprinting has emerged as a popular, low-cost approach; however, RSS is highly variant due to signal attenuation and multipath effects, which substantially degrade localization accuracy. This paper proposes CR-AP (Capsule Routing on AP-Centric Heterogeneous Graphs), a novel method based on capsule network for RSS-based Wi-Fi fingerprinting. Offline, we extract three types of relations from the radio map-AP-AP cosine-similarity kNN, RP-RP co-occurrence, and RP-AP visibility-to construct a heterogeneous graph. Online, the measurement is mapped into a capsule space, followed by one round of message passing over the graph and multi-round conditional dynamic routing to produce AP soft selection and RP soft weights, which are then fed into a lightweight coordinate regressor. We further introduce a reconstruction loss to align the soft weights with the actual observation. On the SODIndoorLoc dataset, CR-AP achieves a mean error of 2.07 m and a P90 of 3.94 m. In AP-missing experiments, CR-AP maintains strong robustness when less than 50% of APs are missing. Cross-scene evaluation on CETC331 also demonstrates good generalization. Compared with multiple baselines and recent models, CR-AP consistently delivers superior performance across several metrics.

The deep integration of digitalization and intelligence presents unprecedented challenges to traditional teaching models. To address common issues such as fragmented knowledge systems, weak practical links, insufficient personalized cultivation, and single evaluation mode, this study constructs a systematic reform framework of “concept–technology empowerment–mechanism innovation” based on the concept of digital-intelligence integration and vocational-undergraduate collaborative education. It proposes a four-dimensional implementation path: “driven by dynamic knowledge graphs, by project-flow simulation, supported by AI–human collaborative teaching, and guaranteed by an integrated practical teaching system.” The research specifically focuses on the construction of a higher vocational-undergraduate integrated practical teaching system, forming a progressive practical teaching closed loop of “basic skill training–comprehensive ability cultivation–innovative ability stimulation” through the construction of a cross-stage, cross-disciplinary, virtual-real combined practical platform. The results show that this system can effectively promote the transformation of theoretical knowledge into practical ability and enhance students’ comprehensive literacy and innovative spirit, providing a replicable systematic solution for higher education teaching reform under higher vocational-undergraduate cooperation.

AbstractSecure colour image transmission requires encryption schemes in which randomness generation, nonlinear substitution and diffusion are designed as a coherent whole. Conventional coupled map lattice-based methods are often constrained by fixed neighbourhood evolution and limited coupling diversity, while many chaos-based substitution-box (S-box) schemes use chaotic sequences mainly as external shuffling sources. This article develops a colour image encryption method that couples adaptive spatio-temporal chaos with dynamic finite-field substitution and coordinated permutation–diffusion. In the proposed chaotic model, lattice interactions and local evolution are regulated by the system state, allowing the generated sequences to exhibit stronger complexity, sensitivity and distribution uniformity. The resulting chaotic flow is further embedded into a dual-affine construction over GF(28), producing key-dependent bijective S-boxes with enhanced nonlinear substitution capability. For image encryption, time-varying bit-switching transform (TV-BST) permutation, Hopfield-driven time-varying coupled map lattice (HDTCML) block key mixing and dynamic S-box-based bidirectional diffusion are integrated to jointly strengthen global scrambling, local confusion and plaintext sensitivity. Experimental analyses show that the proposed scheme produces random-like ciphertexts, suppresses adjacent-pixel correlation and improves resistance to statistical analysis, differential attacks and chosen-plaintext attacks.

The advent of sixth-generation (6G) networks promises ultrahigh bandwidth, reliable connectivity, and ultralow latency, enabling large-scale Internet of Things (IoT) deployment. Network slicing is central to these capabilities, but conventional deep learning approaches often suffer from privacy risks, high computational cost, and poor energy efficiency. To address these challenges, this work proposes a federated and explainable artificial intellige (AI) framework for energy-efficient IoT slicing in 6G. Federated learning (FL) enables collaborative training without sharing raw data, preserving privacy and reducing communication overhead. A transformer-based model captures complex traffic patterns, while a hybrid swarm-intelligence optimizer balances throughput, latency, and energy consumption. SHAP-based explainability enhances transparency in slice allocation. Experiments on real and simulated traffic confirm superior performance, with the proposed framework achieving 98.42% accuracy compared to convolutional and long short-term memory network (CNN+LSTM, 92.67%) and Harris Hawks optimization with CNN+LSTM (HHO-CNN+LSTM, 95.12%). These results demonstrate a scalable, privacy-preserving, and sustainable solution for future 6G IoT ecosystems.

In this study, we propose the Frequency-domain Feature Fusion Module (F3M) to address the challenges of underwater object detection, where optical degradation—particularly high-frequency attenuation and low-frequency color distortion—significantly compromises performance. We critically re-evaluate the need for strict invertibility in detection-oriented frequency modeling. Traditional wavelet-based methods incur high computational redundancy to maintain signal reconstruction, whereas F3M introduces a lightweight “Separate–Project–Fuse” paradigm. This mechanism decouples low-frequency illumination artifacts from high-frequency structural cues via spatial approximation, enabling the recovery of fine-scale details like coral textures and debris boundaries without the overhead of channel expansion. We validate F3M’s versatility by integrating it into both Convolutional Neural Networks (YOLO) and Transformer-based detectors (RT-DETR). Evaluations on the SCoralDet dataset show consistent improvements: F3M enhances the lightweight YOLO11n by 3.5% mAP50 and increases RT-DETR-n’s localization accuracy (mAP50–95) from 0.514 to 0.532. Additionally, cross-domain validation on the deep-sea TrashCan-Instance dataset shows F3M achieving comparable accuracy to the larger YOLOv8n while requiring 13% fewer parameters and 20% fewer GFLOPs. This study confirms that frequency-domain modulation provides an efficient and widely applicable enhancement for real-time underwater perception.

To design and validate a real-time, low-cost edge system that detects multiple dangerous electric-bicycle riding behaviors. We formulate the task as multi-class object detection plus rider counting and implement it with an enhanced YOLO variant, SDV-YOLO. Helmet status (Helmet/No_Helmet) and hand status (Grab/No_Grab) are treated as learnable classes, while overloading (rider count > 2) is inferred by associating person and vehicle boxes after class-wise NMS. SDV-YOLO integrates (i) SPDConv down-sampling to preserve detail with fewer FLOPs, (ii) DySample adaptive up-sampling for content-aware multi-scale fusion, and (iii) a VoV-GSCSP neck for efficient channel aggregation. The processing pipeline is: frame capture -> resize to 640 x 640 -> normalization -> INT8 inference (PyTorch -> ONNX -> Hailo HEF) -> confidence filtering and NMS -> behavior logic and alert output. On a 2041-image, five-class dataset, SDV-YOLO raises recall by 10.6% over YOLOv8n, maintains mAP(50) = 0.748, and reduces FLOPs by 13.6%. Deployed on a Raspberry Pi 5 + Hailo-8L, the system reaches 122 fps (6.8 ms latency) at a hardware cost of approximate to USD 150, confirming scalability for roadside supervision.

With the rapid development of the internet, phishing attacks have become more diverse, making phishing website detection a key focus in cybersecurity. While machine learning and deep learning have led to various phishing URL detection methods, many remain incomplete, limiting accuracy. This paper proposes CSPPC-BiLSTM, a malicious URL detection model based on BiLSTM (Bidirectional Long Short-Term Memory, BiLSTM). The model processes URL character sequences through an embedding layer and captures contextual information via BiLSTM. By integrating CBAM (Convolutional Block Attention Module, CBAM), it applies channel and spatial attention to highlight key features and transforms URL sequence features into a spatial matrix. The SPP (Spatial Pyramid Pooling, SPP) module enables multi-scale pooling. Finally, a fully connected layer fuses features, and dropout regularization enhances robustness. Compared to CharBiLSTM, CSPPC-BiLSTM significantly improves detection accuracy. Evaluated on two datasets, Grambedding (balanced) and Mendeley AK Singh 2020 phish (imbalanced)—and compared with six baselines, it demonstrates strong generalization and accuracy. Ablation experiments confirm the critical role of CBAM and SPP in boosting performance.

This work proposes S2-YOLO, a lightweight marine-debris detector that integrates a parameter-free space-to-depth down-sampling block and a grouped-shuffle Slim-Neck. Compared with YOLOv8n on the TrashCan-Instance benchmark, S2-YOLO cuts GFLOPs by 14.8 % and boosts mAP50-95 by 2.8 %, providing a compact yet accurate solution for embedded cleanup robots.

Based on MOOC, the blended teaching mode of piano in university have shown significant advantages in enhancing teaching efficiency and students’ independent learning ability due to its openness and richness of resources. Traditional piano teaching is mostly a “one-to-one” mode, which is effective, but with high cost and limited coverage, it is difficult to meet diversified learning needs. This study focuses on MOOC-supported blended piano teaching in university, and proposes innovative teaching strategies and practical paths. The study shows that MOOC blended teaching provides a more efficient and flexible solution for piano courses in in university, and puts forward suggestions for optimizing resources and evaluation mechanisms, providing new ideas for innovation in music education in university.

WiFi fingerprinting has become a widely adopted solution for indoor localization due to its low deployment cost and wide availability. However, its positioning accuracy is often unsatisfactory, especially in complex environments involving multiple floors and buildings, where signal interference and structural complexity significantly degrade localization performance. To address these challenges, this paper proposes a deep neural network (DNN) framework that integrates a Stacked Autoencoder (SAE) for feature extraction and a multi-label classification strategy for accurate localization using received signal strength (RSS) data. To optimize the model parameters effectively in high-dimensional spaces, we introduce a Dual-Swarm Particle Swarm Optimization (DSPSO) algorithm. Unlike conventional PSO, which is prone to premature convergence, DSPSO partitions the particle population into two distinct sub-swarms with adaptive update mechanisms to enhance global exploration and avoid local optima. Experimental evaluations on seven 100-dimensional benchmark functions demonstrate that DSPSO outperforms traditional PSO and recent algorithms like the Sparrow Search Algorithm (SSA), achieving the global optimum in four cases. When applied to the WiFi localization task, the DSPSO-optimized SAE-DNN model achieves an average positioning error of 9.42 m—representing 13.74% improvement over the Support Vector Regression (SVR) model and 24.03% improvement over the non-optimized version. Furthermore, the model achieves 100% accuracy in building identification and 92.97% accuracy in floor prediction, proving its effectiveness in multi-building, multi-floor indoor localization scenarios.

Random numbers play a particularly critical role in cryptographic applications, especially true random numbers (TRNs). However, generating TRNs typically incurs high costs. This paper proposes a novel True Random Number Generator (TRNG) method based on multimodal feature fusion from smartphone video. The proposed approach innovatively leverages video frames, audio signals, and simulated triaxial accelerometer data as heterogeneous entropy sources. By applying grayscale conversion and generalized cat map scrambling to image frames, followed by vectorization, and combining with normalized audio waveforms and perturbed inertial sensor data, a high-entropy composite source is constructed. A post-processing stage then employs a Chaotic Hash function based on a piecewise linear chaotic map (PWLCM) to produce highly unpredictable and cryptographically strong random bit sequences. The generated sequences are rigorously evaluated using the NIST SP 800-22 statistical test suite. Experimental results demonstrate that the proposed method outperforms traditional TRNG schemes based solely on image entropy in both randomness quality and computational efficiency. Furthermore, the system operates without requiring external hardware, indicating strong practical viability and scalability.

With the increasing demand for secure multimedia data transmission, chaotic systems have attracted widespread attention due to their sensitivity to initial conditions and pseudo-randomness. However, traditional low-dimensional chaotic systems often suffer from limited complexity and predictability, making them insufficient for high-security image encryption applications. To enhance the practicality and security of chaotic systems in the field of image encryption, this paper proposes a novel four-dimensional multi-scroll multi-wing hyperchaotic system and a corresponding color image encryption algorithm incorporating a DNA mutation mechanism. First, A nonlinear coupling term and an external periodic modulation factor are introduced into the Lorenz system to construct a four-dimensional multi-scroll multi-wing hyperchaotic system. This system features 16-wing symmetric attractors, time-varying equilibrium points, and complex dynamical behavior. Its hyperchaotic properties and nonlinear evolution characteristics are validated using Lyapunov exponent spectra, bifurcation diagrams, phase portraits, and Poincaré sections. Subsequently, based on this system, a DNA-based image encryption framework is proposed by combining DNA computing with gene mutation mechanisms. The encryption scheme includes two-level scrambling, self-feedback, inter-channel cross diffusion, and a key-controlled DNA mutation module. Experimental results show that the algorithm achieves excellent performance in information entropy, NPCR/UACI, and key sensitivity, demonstrating its effectiveness and robustness for image information protection.

With the rapid development of IoT technology, the Fundamentals of IoT Hardware course—one of the core subjects in university IoT programs—urgently requires innovation and improvement in both its content and teaching methodology. This paper, based on an educational reform project funded by the Hainan Provincial University Education Program, explores how to effectively implement general education on Very Large-Scale Integration (VLSI) design and manufacturing within the Fundamentals of IoT Hardware course. The study conducts practical teaching experiments through innovative instructional models, visualized presentation of semiconductor device structures and processes, integration of industrial-grade simulation tools, and the application of cutting-edge technologies. The objective is to stimulate students’ innovative thinking and enhance their hands-on abilities. Finally, this paper summarizes the outcomes of implementing VLSI general education in the course and offers relevant suggestions for further educational reform.

Quantum secret sharing (QSS) is a fundamental primitive in quantum cryptography, enabling secure distribution of sensitive information among multiple parties. However, existing protocols often suffer from vulnerabilities such as collusion attacks, uneven share distribution, and protocol failure due to random basis selection. This paper proposes a novel QSS protocol that integrates hash-based measurement basis derivation with decoy photon-enhanced eavesdropping detection. The scheme ensures fair and deterministic secret reconstruction while eliminating the need for entangled states or complex quantum operations, thereby significantly reducing the quantum operational complexity and hardware requirements. The protocol's correctness, security against both internal and external adversaries, and efficiency in terms of qubit usage, transmission delay, and computational overhead are rigorously analyzed. Notably, the protocol supports dynamic participant management, allowing agents to join or leave securely without re-executing the entire protocol. Comparative analysis demonstrates that the proposed approach outperforms existing QSS protocols across multiple performance metrics. By minimizing quantum resource requirements-such as the elimination of entanglement generation and complex quantum operations-our protocol enhances compatibility with near-term quantum networks, thus facilitating the transition from theoretical QSS models to experimentally feasible implementations.