Multimodal Sentiment Analysis (MSA), a key area in affective computing, involves developing models to process multimodal inputs and predict sentiment values. However, existing MSA methods often struggle with performance degradation due to random modality feature loss, which is a common occurrence in real-world scenarios. This paper introduces ATD-AMSMamba, a novel Mamba-based architecture for MSA. It integrates two key components—Adversarial Token Dropout (ATD) and Adaptive Multimodal Scanning Mamba (AMSM)—to enhance robustness through token transformation and state-space transition. Specifically, ATD mitigates over-dependence on specific uni-modal tokens during training, thereby improving robustness to missing modality scenarios. Furthermore, we introduce AMSM, which employs adaptive state weighting to ensure robust intra-propagation of multimodal state transitions, effectively mitigating corrupted transitions caused by missing modalities. This paper explores the potential of Mamba-based models for enhancing robustness in multimodal sentiment analysis. Experiments on CMU-MOSI, CMU-MOSEI, and CH-SIMS validate that ATD-AMSMamba significantly outperforms previous state-of-the-art methods.

With the widespread adoption of wearable devices, human activity recognition (HAR), which is an essential branch of multivariate time series classification (MTSC), has been broadly applied in areas such as intelligent health monitoring, smart homes, and sports training. Traditional time series processing methods focus on time-domain features and often neglect frequency-domain characteristics; however, much of the critical information related to human activities is usually reflected in the frequency components of signals. To address this issue, we propose a Fourier transform convolution fusion module for extracting frequency-domain features and integrating them with time-domain features. Based on this module, we further propose an innovative Fourier transform convolution network (FTCNet). FTC-Net efficiently integrates time-domain and frequencydomain features by combining the Fourier transform convolution fusion module with an efficient channel attention block (ECA block), which enhances the model's ability to extract key features. The experimental results show that FTC-Net achieved excellent classification performance across 11 HAR-related datasets, with an average accuracy of $87.85 \%$, significantly outperforming current mainstream models. The results of ablation studies also demonstrate the positive impact of frequency-domain features on model performance. FTC-Net offers a new solution for complex human activity recognition tasks and opens new research directions for integrating time-domain and frequency-domain features.

This study aimed to detect stressful hazard perception scenarios subjectively and objectively when using intelligent driving systems. We used electrooculography (EOG), electroencephalography (EEG), subjective ratings, and interviews to identify potential stressful hazard perceptions and record improvements in an intelligent navigation-guided pilot (NGP) system. Moreover, we analyzed electrophysiological data. Our study contributes to the use of engagement, concentration, and phase locking value connectivity based on EEG to support previous research methodologies using beta power, pupil size, fixation ratio, fixation duration, and subjective evaluations for investigating hazard perception. Our analyses showed that stressful hazard perception scenarios occurred mainly when encountering broken and solid lines, frequent lane changes, cars approaching suddenly, several cars driving in parallel, and the decision to change lanes but immediately pulling back upon using the NPG system. Our findings shed light on obtaining accurate results based on subjective and objective evaluations for developing intelligent driving systems.

With the rise of sensor-equipped mobile devices, Mobile Crowd Sensing (MCS) has emerged as an efficient method for information gathering. In smart city environmental sensing, workers can acquire data by merely being within the sensing area. Currently, most studies select opportunistic workers based on the workers’ prior preferences and ignore the effect of movement trajectories on potential opportunistic workers. This may result in the selected opportunistic workers being less-than-ideal, or even ignoring the failure of some tasks to be accomplished, thus resulting in a waste of resources. Therefore, this paper proposes a Recruitment Framework for judging Opportunistic Workers based on Movement Trajectories (RFOW-MT), a two-phase framework for worker recruitment. In the offline phase, combining the neural network model Long Short-Term Memory (LSTM) and Geohash algorithm, an algorithm to detect the set of candidate opportunistic workers is proposed, solving the problems of location privacy and search efficiency. In the online phase, in order to maximize the task spatial coverage under the task budget constraint, a task allocation algorithm based on geographic location packed grouping is proposed. Finally, RFOW-MT outperforms other methods in terms of task spatial coverage and runtime as verified by experiments on real datasets.

Current diagnostic approaches for macular degeneration in optical coherence tomography (OCT) images often lead to misdiagnosis due to their limited ability to capture the disease’s multiscale and irregular features. We present PDC-DETR, a Parallel Dilated Convolutional Detection Transformer, which introduces three major innovations for accurate and efficient macular degeneration analysis. First, a Parallel Feature-Optimized Attention Pyramid Network (PFOAPN) enables simultaneous modeling of global context and local details through multi-scale feature pathways. Second, a novel Wise-MPDIoU loss dynamically adjusts to variations in image quality while improving lesion localization accuracy. Third, the lightweight design ensures real-time clinical applicability, requiring only 38.2 MB parameters and 58.5 GFLOPs, while achieving 94.1% accuracy across five macular degeneration categories at 71 FPS. This study establishes a new benchmark for automated OCT-based retinal disease detection, providing both high diagnostic accuracy and practical clinical deployment potential.

Abstract In recent years, secure communication and privacy‐preserving are gaining popularity because of the advent of wireless network. Certificateless cryptography (CLC) and identity‐based cryptography (IBC) have already been widely applied in the area of wireless network. Hence, to achieve secure communication from CLC to IBC, we present a heterogeneous signcryption scheme that enables a sender in the CLC to transmit a message to multi‐receiver in the IBC. Our scheme has no restrictions on the system parameters and master keys; private key generator and key generating center generate their own system parameters and master key for the IBC users and the CLC user, separately. Considering privacy‐preserving in heterogeneous communication, we show that our scheme can realize the anonymity of all receivers. Compared with other schemes, our scheme has more advantages, such as public verification and decryption fairness, which are attractive for wireless network applications. Moreover, our scheme has the indistinguishability against adaptive chosen ciphertext attacks, the existential unforgeability against adaptive chosen messages attacks, and anonymous indistinguishability of encryptions under selective‐ID, chosen ciphertext attack in the random oracle model. Copyright © 2016 John Wiley & Sons, Ltd.