Accurate emotion recognition from physiological signals is critical for applications in healthcare, autonomous systems, and human-computer interaction. However, prevailing methods often fail to model long-term dependencies and overlook periodic patterns inherent in physiological data. To address these challenges, we propose Phy-FusionNet, a novel memory-augmented transformer architecture for multimodal emotion recognition. Phy-FusionNet introduces a Memory Stream Module with FIFO-queue and decay-based updates to preserve long-term contextual information. It further integrates Fourier-based positional encoding and frequency-aware attention, enabling robust detection of periodic emotional cues. An Adaptive Temporal Attention Module enhances computational efficiency and enables dynamic relevance in temporal feature extraction. For cross-modal fusion, we employ a transformer-based Multimodal Binding Learning framework that balances modality-specific and shared features. Extensive experiments on five public datasets-WESAD, CL-Drive, PPB-Emo, PhyMER, and EEG-VUI-demonstrate that Phy-FusionNet outperforms state-of-the-art models, achieving up to 16.3% improvement in accuracy and superior robustness across diverse emotional states and noisy environments. Notably, the model maintains low performance variance across emotion classes, with F1-Score differences under 2.5%, indicating stable recognition even for subtle or overlapping emotions. Our results underscore the importance of integrating memory, frequency, and adaptive attention for effective affective computing.

The construction of large-scale hydraulic projects has disrupted hydrological connectivity between rivers and lakes, directly affecting water quality and posing profound risks to water security. However, the quantitative understanding of how different modes of connectivity affect water quality in the sluice-controlled river network area remains limited. In this paper, we proposed a novel scenario-based framework to assess the water quality responses to connectivity from the dual perspectives of temporal frequency and flow magnitude, which is regulated by hydraulic facilities’ operation. Regional water quality was simulated under a series of connectivity frequency scenarios using connectivity–water quality response formulas. Meanwhile, the water quality of major rivers under varying discharge conditions was evaluated by the hydrodynamic–water quality model. Results indicate that continuous operation of recession and regulating sluices, together with extending the nonflood operation of pumping stations to 3–41 days, significantly reduces TP, TN, NH3-N, and CODMn concentrations by 1% to 4%. Besides, increasing inflow discharge results in moderate improvements in water quality, with a discharge of 50 m3/s reducing the mean TN concentration by approximately 1.4 mg/L compared with baseline conditions. Maintaining stable low-flow discharge at outflow sluices prevents prolonged stagnation and supports overall nutrient export. These findings demonstrate that water quality changes in regulated river–lake networks are jointly governed by connectivity timing and flow intensity, providing an analytical framework for adaptive water management in urbanized plain river networks.

Brain-computer interface systems can recognize users' emotions through electroencephalography (EEG). EEG-based human emotion recognition is an emerging field that is gaining significant traction within the realm of brain-computer interfaces. However, due to the complexity and diversity inherent in EEG signals, emotion recognition remains a challenge in pattern recognition. The critical task of selecting salient features from EEG and achieving high recognition accuracy warrants further exploration. In this paper, a hybrid emotion detection system is proposed by incorporating the reinforcement learning mechanism into a deep learning framework. Reinforcement learning is used to recursively select informative features, while a Long Short-Term Memory Network (LSTM) and a deep neural network are employed for enhanced feature selection and emotion recognition. Specifically, the LSTM, based on input features, determines and generates the current state, thereby aiding the policy model in making action decisions. This process successively retains or removes features to improve emotion recognition in the next state. The neural net-based policy model generates the policy actions based on the current state and the corresponding reward signal from the classification result, to control the feature selections for the subsequent states. A public EEG emotion dataset of SEED is used in the experiments. Results show that the proposed network model is effective in feature selections and emotion classifications, which reduces feature dimensions by 11.3% on average, and achieves a higher recognition accuracy of 92.65% compared to other approaches. The proposed system can use the current state info for prediction and adaptive feature selection, which can accommodate the data pattern differences of individual participants and leverage the model for a good performance.

The magnetoencephalography (MEG) functional connectome has been shown to hold rich information on cognitive states and the health status of the brain. In this study, we investigate whether multiple recordings of MEG can be classified as to whether they occurred less than a week apart or a week or more apart. We use the difference between session functional connectomes as the primary feature, followed by multi-agglomerative clustering and logistic regression with elastic net regularization. We run this with 5-split stratified folds, repeating the runs with replacement 20 times, totaling 100 folds. We achieve a mean AUROC score of 0.72 ± 0.17SD in the theta band using phase-linearity-measurement (PLM) connectivity, indicating generalizable differentiable changes in the brain from both periods. In addition to longitudinal approaches to understanding MEG recordings, this finding could be applied to MEG fingerprinting work, where the re-identification of subjects largely depends on their longitudinal robustness.

Alzheimer’s disease is a neurodegenerative disease that seriously threatens the life and health of the elderly. This study used three-dimensional lightweight neural networks to classify the stages of Alzheimer’s disease and explore the relationship between the stages and the variations of brain tissue. The study used CAT12 to preprocess magnetic resonance images of the brain and got three kinds of preprocessed images: standardized images, segmented standardized gray matter images, and segmented standardized white matter images. The three kinds of images were used to train four kinds of three-dimensional lightweight neural networks respectively, and the evaluation metrics of the neural networks are calculated. The accuracies of the neural networks for classifying the stages of Alzheimer’s disease (cognitively normal, mild cognitive impairment, Alzheimer’s disease) in the study are above 96%, and the precisions and recalls of classifying the three stages are above 94%. The study found that for the classification of cognitively normal, the best classification results can be obtained by training with the segmented standardized gray matter images, and for mild cognitive impairment and Alzheimer’s disease, the best classification results can be obtained by training with the standardized images. The study analyzed that in the process of cognitively normal to mild cognitive impairment, variations in the segmented standardized gray matter images are more obvious at the beginning, while variations in the segmented standardized white matter images are not obvious. As the disease progresses, variations in the segmented standardized white matter images tend to become more significant, and variations in the segmented standardized gray matter images and white matter images are both significant in the development of Alzheimer’s disease.

Random Boolean Network (RBN) is a type of regulatory network in which the nodes have Boolean values representing their states. The robustness of RBNs against perturbations is a crucial characteristic, and there has been a growing interest in enhancing the network's robustness. In this study, a biologically inspired epigenetic regulation method is proposed to enhance the robustness of the RBNs. A frequency encoding method based on pulse counting is employed to encode the node states within a sliding time window, thereby improving the form of epigenetic regulation. To verify the performance of this method, an antifragility indicator is adopted to measure the robustness of RBNs and yeast cell networks at different scales. The experimental results demonstrate that the networks with epigenetic regulation exhibit excellent robustness, even in the presence of large-scale networks and severe perturbations. This approach provides a new perspective and idea for designing robust RBNs and discrete networks.

Hydrological connectivity remarkably affects the water quality of river-lake systems, particularly in densely urbanized plain river network areas, where its impact remains unclear. The growing urbanization and rapid changes in hydrological networks make it more challenging to manage water quality effectively. Understanding how hydrological connectivity changes and the influence on key water quality variables is crucial for improving management strategies. We quantified hydrological connectivity between lakes in the northern Taihu Lake Basin using a connectivity topological model based on graph theory and landscape ecology. XG-Boost models were developed to elucidate the potential threshold effect of hydrological connectivity on key water quality parameters. These models were accompanied by linear mixed-effect (LME) models, which included land use types as a random effect to evaluate the response relationship between hydrological connectivity and water quality. Results indicated that the spatiotemporal dynamics of hydrological connectivity decreased over the last 20 years. Furthermore, changes in hydrological connectivity considerably influenced environmental variables in river-lake network areas. The XG-Boost models identified a Pij value of 0.02 as a potential threshold, at which spatial hydrological connectivity begins to impact water quality as concentrations change steadily above this threshold. The LME models confirmed that enhanced spatial hydrological connectivity was generally associated with reduced concentrations of TN, TP, NH3-N, and CODMn, and increased DO levels. In addition, hydrological connectivity was influenced by factors such as the shortest river path between lakes and hydraulic facilities along the path. This finding suggests that hydrological connectivity can be restored to improve water quality by refining river network topology, optimizing existing sluice schedules, or removing unnecessary dikes. These results highlight the potential of hydrological connectivity optimization to support water quality improvement strategies in complex urban river networks.

The COVID-19 outbreak has negatively impacted the income of many bank users. Many users without emergency funds had difficulty coping with this unexpected event and had to use credit or apply to the government for bailout funds. Therefore, it is necessary to develop spending plans and deposit plans based on transaction data of users to assist them in saving sufficient emergency funds to cope with unexpected events. In this paper, an emergency fund model is proposed, and two optimization algorithms are applied to solve the optimal solution of the model. Secondly, an early warning mechanism is proposed, i.e. an unexpected prevention index and a consumption index are proposed to measure the ability of users to cope with unexpected events and the reasonableness of their expenditure respectively, which provides early warning to users. Finally, the model is experimented with real bank users and the performance of the model is analysed. The experiments show that compared to the no-planning scenario, the model helps users to save more emergency funds to cope with unexpected events, furthermore, the proposed model is real-time and sensitive.

Humans have played a fundamental role in altering lake wetland ecosystems, necessitating the use of diverse data types to accurately quantify long-term changes, identify potential drivers, and establish a baseline status. We complied high-resolution historical topographic maps and Landsat imagery to assess the dynamics of the lake wetlands in the Yangtze Plain over the past century, with special attention to land use and hydrological connectivity changes. Results showed an overall loss of 45.6% (∼11859.5 km2) of the lake wetlands over the past century. The number of lakes larger than 10 km2 decreased from 149 to 100 due to lake dispersion, vanishing, and shrinkage. The extent of lake wetland loss was 3.8 times larger during the 1930s–1970s than that in the 1970s–1990s. Thereafter, the lake wetland area remained relatively stable, and a net increase was observed during the 2010s–2020s in the Yangtze Plain. The significant loss of lake wetland was predominately driven by agricultural activities and urban land expansion, accounting for 81.1% and 4.9% of the total losses, respectively. In addition, the changes in longitudinal and lateral hydrological connectivity further exacerbated the lake wetland changes across the Yangtze Plain through isolation between lakes and the Yangtze River and within the lakes. A total of 130 lakes have been isolated from the Yangtze River due to the construction of sluices and dykes throughout the Yangtze Plain, resulting in the decrease in the proportion of floodplain marsh from 28.3% in the 1930s to 8.0% in the 2020s. Furthermore, over 260 sub-lakes larger than 1 km2 (with a total area of 1276.4 km2) are experiencing a loss of connectivity with their parent lakes currently. This study could provide an improved historical baseline of lake wetland changes to guide the conservation planning to wetland protection and prioritization area in the Yangtze Plain.

The emotional health benefits of urban green space have been widely recognized. Flower borders, as a perennial plant landscape, have gradually become a current form of plant application in urban green spaces due to their rich color configurations. However, the related research primarily focuses on the impact of urban green spaces on public health, with relatively little attention given to how the colors of flower borders affect public emotional health. This study explored the relationship between the flower borders color characteristics and the public emotional health. In this study, 24 sample images were used as experimental materials, which selected based on their color richness and harmony. Additionally, face recognition technology and online random questionnaires were utilized to measure the public basic emotions and pleasure, respectively. The result shows that, based on the HSV color model and expert recommendations, 19 color characteristics were identified. The correlation analysis of the results from the public emotion with these color characteristics revealed that 13 color characteristics correlated with public emotional pleasure. Among them, blue, neutral purple, and low saturation were positively correlated. Through factor analysis, these thirteen color characteristics were summarized and categorized into four common factors (F1–F4), three of which are related to color. They are “low saturation of blue-violet percentage” (F1), “color configuration diversity” (F2), “bright red percentage” (F3), and “base green percentage” (F4), with F1 having the largest variance explained (27.88%). Finally, an evaluation model of color characteristics was constructed based on the variance explained by these four factors, which was demonstrated to effectively predict the level of public emotional pleasure when viewing flower borders. The results shed light on the effects of color characteristics on public emotions and provide new perspectives for subsequent flower border evaluations. Our results provide a valuable reference for future flower border color design, aiming to better improve public emotional health.

The proliferation of hydraulic facilities worldwide has greatly hindered the lake wetlands ecosystem services through impacting hydrological connectivity. However, developing a general method to evaluate hydrological connectivity in the plain river-lake wetlands networks heavily influenced by hydraulic facilities remains challenging. In this study, we present a novel framework for the hydrological connectivity evaluation of lake wetlands whilst considering different types of hydraulic facilities. By generalising lake wetlands into patches and rivers into edges, a river topological network connecting lake wetlands was constructed based on graph theory, and connectivity indices were calculated through three dimensions: individual patches - inter-patches - and study area, respectively. Using detailed historical topographic maps and high-resolution satellite data, this new framework, combining graph theory and connectivity indices, is applied to a river-lake wetlands district with dense sluices/pumping stations in eastern China. Results show that (1) Although the number of lake wetlands remained relatively stable over the past century, the average area decreased by 18.44%, and the density of hydraulic engineering boomed to 2.2/km2. (2) The increases in hydraulic facilities and the decrease in lake wetlands areas significantly reduced the degree of hydrological connectivity, resulting in the overall connectivity value decreasing from 0.97 in the 1910s to 0.45 in 2019. (3) The importance of the same lake wetlands patch has progressively heightened in the last hundred years. This study provides an effective method for quantifying hydrological connectivity in plain river-lake wetlands with dense hydraulic facilities and could shed light on lake wetlands protection and restoration.

Convolutional Neural Network-based Predictors (CNNP) have emerged as a viable solution for enhancing global optimization and prediction capabilities in the field of Reversible Data Hiding (RDH). However, they often encounter limitations in data embedding space due to scarcity of zero-valued points in predicted images. To address these challenges, in this paper, we propose an Optimized CNNP (OCNNP) technique that increases the zero-valued points in predicted images, significantly enhancing the embedding capacity. Additionally, we introduce a novel Lower Surround Background Complexity (LSBC)-based Prediction Error Expansion (PEE) method, which refines the sorting of prediction errors for data embedding, thereby reducing image distortion and improving overall embedding performance. Experimental results demonstrate that our approach markedly improves the Peak Signal-to-Noise Ratio (PSNR) across various embedding capacities compared to existing methods and shows high robustness against various noise types, establishing its superiority in enhancing both the capacity and quality of RDH in complex image distributions.

In brain-computer interface (BCI) systems, users' emotion can be recognized by using electroencephalography (EEG) data. Recent researches proposed different methods for feature extraction and EEG-based emotion classification. However, EEG data collected over time is not always stable and accuracies of emotion recognition are found not robust. In this paper, a novel EEG emotion recognition system is proposed by combining ReliefF for extracting features, and Long Short-Term Memory Network coupled with Support Vector Machine for classification. A public EEG emotion dataset of SEED which contains three different recordings of EEG data is used in the experiments. It is shown that the proposed network is effective for mitigating the phenomenon of unstable accuracies, and achieves higher emotion classification accuracies compared to other approaches using the same dataset.

Tampered images with false information can mislead viewers and pose security issues. Tampering traces in images are difficult to detect. To locate tampering traces effectively, a dual-domain deep-learning-based image tampering localization method based on RGB and frequency stream branches is proposed in this work. The former branch learns and extracts tampered features on the image and content features of the tampered region. The latter branch extracts tampered features from the frequency domain to complement the RGB stream branch. In addition, an attention mechanism is used to integrate the features from both branches at the fusion stage. In the experiments, the F1 score of the proposed method outperformed those of the baselines on the NIST16 dataset (with a 15.3% improvement), and the AUC score outperformed those of the baselines on the NIST16 and COVERAGE datasets (improvements of 3.9% and 4.7%, respectively). This study provides a beneficial alternative to image tampering localization techniques.

Recent studies have shown that the Magnetoen-cephalography (MEG) functional connectome is person-differentiable in a same-day recording with as little as 20 latent components, showing variability across synchrony measures and spectral bands. Here, we succeed with ![Graphic][1] components of the functional connectome on a multi-day dataset of 43 subjects and link it to related clinical applications. By optimizing sub-networks of ![Graphic][2] regions with 30 seconds of broadband signal, we find robust fingerprinting performance, showing several patterns of region re-occurrence. From a search space of 5.72 trillion, we find 46,071 of many more acceptable solutions, with minimal duplicates found in our optimization. Finally, we show that each of these sub-networks can identify 30 Parkinson’s patient sub-networks from 30 healthy subjects with a mean F1 score of 0.716 ± 0.090SD. MEG fingerprints have previously been shown on multiple occasions to hold patterns on the rating scales of progressive neurodegenerative diseases using much coarser features. Furthermore, these sub-networks may similarly be useful for identifying patterns across characteristics for age, genetics, and cognition. ### Competing Interest Statement The authors have declared no competing interest. [1]: /embed/inline-graphic-1.gif [2]: /embed/inline-graphic-2.gif

This study presents a robust framework that leverages advanced deep-learning techniques for ear-based human recognition. Faced with the challenge of dataset sizes, our approach is developed based on a generative adversarial network (GAN) method namely Pix2Pix to augment the dataset. It is demonstrated that this approach offers the ability to produce complementary images for ear recognition. To be more specific, Pix2Pix GAN is employed to generate missing sides in ear image pairs (i.e., creating corresponding left ear images for right ear images and vice versa). As such, this augmentation could substantially increase the dataset size, making it more diverse and of significantly greater use for training purposes. The employed dataset consisted of several images of the right ear and only one left ear for each individual. A series of corresponding synthetic left-ear images is generated using Pix2Pix GAN as a tool for augmenting the available data and mitigate the dataset’s lack of left ear images. The experiment framework used the EarNet model and conducted comparative evaluations before and after Pix2Pix GAN augmentation using the AMI Ear dataset. By employing the Pix2Pix GAN, the proposed approach can effectively double the size of a dataset and, in the process, provide significantly greater utility regarding how that data can be utilised in real-world applications scenarios. The resulting accuracy reaches 98% on the AMI dataset, demonstrating that this technique can improve model performance for ear-based human recognition.

A novel instance-based algorithm for pattern classification is presented and evaluated in this paper. This new method is motivated by the challenge of pattern classifications where only limited and/or noisy training data are available. For every classification, the proposed system transforms the query data and the training templates based on their distributions in the feature space. One of the major novelties of the proposed method is the concept of template reconstruction enabling improved performance with limited training data. The technique is compared with similar algorithms and evaluated using both the image and time-series modalities to demonstrate its effectiveness and versatility. Two public image databases, FASHION-MNIST and CIFAR-10, were used to test its effectiveness for the classification of images using small amounts of training samples. An average classification improvement of 2~3% was observed while using a small subset of the training database, compared to the performances achieved by state-of-the-art techniques using the full datasets. To further explore its capability in solving more challenging classification problems such as non-stationary time-series electroencephalography (EEG) signals, a clinical grade 64-electrode EEG database, as well as a low-quality (high-noise level) EEG database, obtained using a low-cost system equipped with a single dry sensor, have also been used to test the algorithm. Adaptive reconstruction of the feature instances has been seen to have substantially improved class separation and matching performance for both still images and time-series signals. In particular, the method is found to be effective for the classification of noisy non-stationary data with limited training data volumes, indicating its potential suitability for a wide range of applications.

Introduction: Fruit diseases have a serious impact on fruit production, causing a significant drop in economic returns from agricultural products. Due to its excellent performance, deep learning is widely used for disease identification and severity diagnosis of crops. This paper focuses on leveraging the high-latitude feature extraction capability of deep convolutional neural networks to improve classification performance. Methods: The proposed neural network is formed by combining the Inception module with the current state-of-the-art EfficientNetV2 for better multi-scale feature extraction and disease identification of citrus fruits. The VGG is used to replace the U-Net backbone to enhance the segmentation performance of the network. Results: Compared to existing networks, the proposed method achieved recognition accuracy of over 95%. In addition, the accuracies of the segmentation models were compared. VGG-U-Net, a network generated by replacing the backbone of U-Net with VGG, is found to have the best segmentation performance with an accuracy of 87.66%. This method is most suitable for diagnosing the severity level of citrus fruit diseases. In the meantime, transfer learning is applied to improve the training cycle of the network model, both in the detection and severity diagnosis phases of the disease. Discussion: The results of the comparison experiments reveal that the proposed method is effective in identifying and diagnosing the severity of citrus fruit diseases identification.

Both gradual and abrupt changes in lake surface area in permafrost regions are crucial for understanding the water cycles in cold regions under climate change. However, seasonal changes in lake area in permafrost regions are not available, and their occurrence conditions are still unclear. Based on remotely sensed water body products at a 30 m resolution, this study provides a detailed comparison of lake area changes across seven basins characterized by clear gradients in climatic, topographic and permafrost conditions in the Arctic and Tibetan Plateau between 1987 and 2017. The results show that the maximum surface area of all lakes net increased by 13.45 %. Among them, the seasonal lake area net increased by 28.66 %, but there was also a 2.48 % loss. The permanent lake area net increased by 6.39 %, and the area loss was approximately 3.22 %. The total permanent lake area generally decreased in the Arctic but increased in the Tibetan Plateau. At lake region scale (0.1° grid), the changes in permanent area of contained lakes were divided into four types including no change, homogeneous changes (only expansion or only shrinkage), heterogeneous changes (expansion neighboring shrinkage) and abrupt changes (newforming or vanishing). The lake regions with heterogeneous changes accounted for over one-quarter of all lake regions. All types of changes in lake regions, especially the heterogeneous changes and abrupt changes (e.g., vanishing), occurred more extensively and intensely on low and flat terrain, in high-density lake regions and in warm permafrost regions. These findings indicate that, considering the increase in surface water balance in these river basins, surface water balance alone cannot fully explain changes in permanent lake area in the permafrost region, and the thawing or disappearance of permafrost plays a tipping point effect on the lake changes.

Reinforcement Learning (RL) is an effective method for adaptive traffic signals control. As one type of RL, the teacher-student framework has been found helpful in improving the model performance for different application fields (such as robot control, game, hybrid intelligence), but it is rarely applied for traffic control due to that the hyper-parameters and the number of state-action pairs experienced are difficult to determine. In this work, the teacher-student framework is used for traffic signal control, where only a single reward function is designed to guide the student agent and by using this method the number of hyper-parameters and the model complexity are reduced. Specifically, the teacher agent uses an importance function to evaluate and guide the student, where the importance function combines with environment reward to form a synthetic reward for the student agent. Experimental results under different traffic environments show that the proposed method achieves the expected performance enhancement and is better than most of the state-of-the-art RL-based traffic signal control methods.