Search Results (31)

Optical metasurfaces enable compact, lightweight and planar optical devices. Their performances, however, are still limited by design approximations imposed by their macroscopic dimensions. To address this problem, we propose a neural network-based multi-stage gradient optimization method to efficiently modulate nonlocal interactions between meta-atoms, which is one of the major effects neglected by current design methods. Our strategy allows for the use of these interactions as an additional design dimension to enhance the performance of metasurfaces and can be used to optimize large-scale metasurfaces with multiple parameters. As an example of application, we design a meta-hologram with a zero-order energy suppressed to 26% (theoretically) and 57% (experimentally) of its original value. Our results suggest that neural networks can be used as a powerful design tool for the next generation of high-performance metasurfaces with complex functionalities. Full article

Deep learning greatly improves the detection efficiency of abnormal traffic through autonomous learning and effective extraction of data feature information. Among them, Graph Neural Networks (GNN) effectively fit the data features of abnormal traffic by aggregating the features and structural information of network nodes. However, the performance of GNN in the field of industrial Internet of Things (IIoT) is still insufficient. Since the asymmetry of GNN traffic data is greater than that of the traditional Internet, it is necessary to propose a detection method with high detection rate. At present, many algorithms overly emphasize the optimization of graph neural network models, while ignoring the heterogeneity of resources caused by the diversity of devices in IIoT networks, and the different traffic characteristics caused by multi type protocols. Therefore, universal GNN may not be fully applicable. Therefore, a novel intrusion detection framework incorporating graph neural networks is developed for Industrial Internet of Things systems. Design mini-batch sampling to support data parallelism and accelerate the training process in response to the distributed characteristics of the IIoT. Due to the strong real-time characteristics of the industrial IIoT, data packets in concentrated time periods contain a large number of feature attributes, and the high redundancy of features due to the correlation between features. This paper establishes a model temporal correlation and designs a new model. The performance of the proposed GIDS model is evaluated on several benchmark datasets such as BoT-IoT, ACI-IoT-2023 and OPCUA. The results marked that the method performs well on both binary classification task and multiclass classification task. The accuracy on binary classification task is 93.63%, 97.34% and 100% with F1 values of 94.34%, 97.68% and 100.00% respectively. The accuracy on multiclass classification task is 92.34%, 93.68% and 99.99% with F1 values of 94.55%, 94.12% and 99.99% respectively. Through experimental measurements, the model effectively utilizes the natural distribution characteristics of network traffic in both temporal and spatial dimensions, achieving better detection results. Full article

Background/Objectives: Structural magnetic resonance imaging (MRI) and 18-fluoro-deoxy-glucose positron emission tomography (PET) reveal the structural and functional information of the brain from different dimensions, demonstrating considerable clinical and practical value in the computer-aided diagnosis of Alzheimer’s disease (AD). However, the structure and semantics of different modal data are different, and the distribution between different datasets is prone to the problem of domain shift. Most of the existing methods start from the single-modal data and assume that different datasets meet the same distribution, but they fail to fully consider the complementary information between the multi-modal data and fail to effectively solve the problem of domain distribution difference. Methods: In this study, we propose a multi-modal deep domain adaptation (MM-DDA) model that integrates MRI and PET modal data, which aims to maximize the utilization of the complementarity of the multi-modal data and narrow the differences in domain distribution to boost the accuracy of AD classification. Specifically, MM-DDA comprises three primary modules: (1) the feature encoding module, which employs convolutional neural networks (CNNs) to capture detailed and abstract feature representations from MRI and PET images; (2) the multi-head attention feature fusion module, which is used to fuse MRI and PET features, that is, to capture rich semantic information between modes from multiple angles by dynamically adjusting weights, so as to achieve more flexible and efficient feature fusion; and (3) the domain transfer module, which reduces the distributional discrepancies between the source and target domains by employing adversarial learning training. Results: We selected 639 subjects from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and considered two transfer learning settings. In ADNI1→ADNI2, the accuracies of the four experimental groups, AD vs. CN, pMCI vs. sMCI, AD vs. MCI, and MCI vs. CN, reached 92.40%, 81.81%, 81.13%, and 85.45%, respectively. In ADNI2→ADNI1, the accuracies of the four experimental groups, AD vs. CN, pMCI vs. sMCI, AD vs. MCI, and MCI vs. CN, reached 94.73%, 81.48%, 85.48%, and 81.69%, respectively. Conclusions: MM-DDA is compared with other deep learning methods on two kinds of transfer learning, and the performance comparison results confirmed the superiority of the proposed method in AD prediction tasks. Full article

Concrete dams are prone to various hidden dangers after long-term operation and may lead to significant risk if failed to be detected in time. However, the existing hollowing detection techniques are few as well as inefficient when facing the demands of comprehensive coverage and intelligent management for regular inspections. Hence, we proposed an innovative, non-destructive infrared inspection method via constructed dataset and proposed deep learning algorithms. We first modeled the surface temperature field variation of concrete dams as a one-dimensional, non-stationary partial differential equation with Robin boundary. We also designed physics-informed neural networks (PINNs) with multi-subnets to compute the temperature value automatically. Secondly, we obtained the time-domain features in one-dimensional space and used the diffusion techniques to obtain the synthetic infrared images with dam hollowing by converting the one-dimensional temperatures into two-dimensional ones. Finally, we employed adaptive joint learning to obtain the spatio-temporal features. We designed the experiments on the dataset we constructed, and we demonstrated that the method proposed in this paper can handle the low-data (few shots real images) issue. Our method achieved 94.7% of recognition accuracy based on few shots real images, which is 17.9% and 5.8% higher than maximum entropy and classical OTSU methods, respectively. Furthermore, it attained a sub-10% cross-sectional calculation error for hollowing dimensions, outperforming maximum entropy (70.5% error reduction) and OTSU (7.4% error reduction) methods, which shows our method being one novel method for automated intelligent hollowing detection. Full article

To address the low estimation accuracy of deep learning-based crop yield image recognition methods under untrained shooting distances, this study proposes a shooting distance adaptive crop yield estimation method by fusing RGB and depth image information through multi-modal data fusion. Taking strawberry fruit fresh weight as an example, RGB and depth image data of 348 strawberries were collected at nine heights ranging from 70 to 115 cm. First, based on RGB images and shooting height information, a single-modal crop yield estimation model was developed by training a convolutional neural network (CNN) after cropping strawberry fruit images using the relative area conversion method. Second, the height information was expanded into a data matrix matching the RGB image dimensions, and multi-modal fusion models were investigated through input-layer and output-layer fusion strategies. Finally, two additional approaches were explored: direct fusion of RGB and depth images, and extraction of average shooting height from depth images for estimation. The models were tested at two untrained heights (80 cm and 100 cm). Results showed that when using only RGB images and height information, the relative area conversion method achieved the highest accuracy, with R² values of 0.9212 and 0.9304, normalized root mean square error (NRMSE) of 0.0866 and 0.0814, and mean absolute percentage error (MAPE) of 0.0696 and 0.0660 at the two untrained heights. By further incorporating depth data, the highest accuracy was achieved through input-layer fusion of RGB images with extracted average height from depth images, improving R² to 0.9475 and 0.9384, reducing NRMSE to 0.0707 and 0.0766, and lowering MAPE to 0.0591 and 0.0610. Validation using a developed shooting distance adaptive crop yield estimation platform at two random heights yielded MAPE values of 0.0813 and 0.0593. This model enables adaptive crop yield estimation across varying shooting distances, significantly enhancing accuracy under untrained conditions. Full article

This paper aims to address the challenges of significant data distribution differences and extreme data imbalances in the remaining useful life prediction modeling of rotating components of complex intelligent equipment under various working conditions. Grounded in deep learning modeling, it considers the multi-dimensional extraction method for degraded data features in the data feature extraction stage, proposes a network structure with multiple attention data extraction channels, and explores the extraction method for valuable data segments in the channel and time series dimensions. This paper also proposes a domain feature fusion network based on feature migration and examines methods that leverage abundant labeled data from the source domain to assist in target domain learning. Finally, in combination with a long short-term memory neural network (LSTM), this paper constructs an intelligent model to estimate the remaining lifespan of rotating components. Experiments demonstrate that, when integrating the foundational deep convolution network with the domain feature fusion network, the comprehensive loss error for life prediction on the target domain test set can be reduced by up to 6.63%. Furthermore, when adding the dual attention feature extraction network, the maximum reduction in the comprehensive loss error is 3.22%. This model can effectively enhance the precision of life prediction in various operating conditions; thus, it provides a certain theoretical basis and technical support for the operation and maintenance management of complex intelligent equipment. It has certain practical value and application prospects in the remaining life prediction of rotating components under multiple working conditions. Full article

Since multi-view learning leverages complementary information from multiple feature sets to improve model performance, a tensor-based data fusion layer for neural networks, called Multi-View Data Tensor Fusion (MV-DTF), is used. It fuses M feature spaces ${\mathcal{X}}_1,\cdots ,{\mathcal{X}}_M$, referred to as views, in a new latent tensor space, $\mathcal{S}$, of order P and dimension $J_1\times \cdots \times J_P$, defined in the space of affine mappings composed of a multilinear map $\mathcal{T}:{\mathcal{X}}_1\times \cdots \times {\mathcal{X}}_M\to \mathcal{S}$—represented as the Einstein product between a $(P+M)$-order tensor $\mathcal{A}$ anda rank-one tensor, $\mathcal{X}={\mathbf{x}}^{(1)}\otimes \cdots \otimes {\mathbf{x}}^{(M)}$, where ${\mathbf{x}}^{(m)}\in {\mathcal{X}}_m$ is the m-th view—and a translation. Unfortunately, as the number of views increases, the number of parameters that determine the MV-DTF layer grows exponentially, and consequently, so does its computational complexity. To address this issue, we enforce low-rank constraints on certain subtensors of tensor $\mathcal{A}$ using canonical polyadic decomposition, from which M other tensors ${\mathcal{U}}^{(1)},\cdots ,{\mathcal{U}}^{(M)}$, called here Hadamard factor tensors, are obtained. We found that the Einstein product $\mathcal{A}{⊛}_M\mathcal{X}$ can be approximated using a sum of R Hadamard products of M Einstein products encoded as ${\mathcal{U}}^{(m)}{⊛}_1{\mathbf{x}}^{(m)}$, where R is related to the decomposition rank of subtensors of $\mathcal{A}$. For this relationship, the lower the rank values, the more computationally efficient the approximation. To the best of our knowledge, this relationship has not previously been reported in the literature. As a case study, we present a multitask model of vehicle traffic surveillance for occlusion detection and vehicle-size classification tasks, with a low-rank MV-DTF layer, achieving up to $92.81\%$ and $95.10\%$ in the normalized weighted Matthews correlation coefficient metric in individual tasks, representing a significant $6\%$ and $7\%$ improvement compared to the single-task single-view models. Full article

Wild mushrooms are popular for their taste and nutritional value; however, non-experts often struggle to distinguish between toxic and non-toxic species when foraging in the wild, potentially leading to poisoning incidents. To address this issue, this study proposes a compact bilinear neural network method based on Transformer and multi-scale feature fusion. The method utilizes a dual-stream structure that integrates multiple feature extractors, enhancing the comprehensiveness of image information capture. Additionally, bottleneck attention and efficient multi-scale attention modules are embedded to effectively capture multi-scale features while maintaining low computational costs. By employing a compact bilinear pooling module, the model achieves high-order feature interactions, reducing the number of parameters without compromising performance. Experimental results demonstrate that the proposed method achieves an accuracy of 98.03%, outperforming existing comparative methods. This proves the superior recognition performance of the model, making it more reliable in distinguishing wild mushrooms while capturing key information from multiple dimensions, enabling it to better handle complex scenarios. Furthermore, the development of public-facing identification tools based on this method could help reduce the risk of poisoning incidents. Building on these findings, the study suggests strengthening the research and development of digital agricultural technologies, promoting the application of intelligent recognition technologies in agriculture, and providing technical support for agricultural production and resource management through digital platforms. This would provide a theoretical foundation for the innovation of digital agriculture and promote its sustainable development. Full article

With the rapid development of industry, the risks factories face are increasing. Therefore, the anomaly detection algorithms deployed in factories need to have high accuracy, and they need to be able to promptly discover and locate the specific equipment causing the anomaly to restore the regular operation of the abnormal equipment. However, the neural network models currently deployed in factories cannot effectively capture both temporal features within dimensions and relationship features between dimensions; some algorithms that consider both types of features lack interpretability. Therefore, we propose a high-precision, interpretable anomaly detection algorithm based on variational autoencoder (VAE). We use a multi-scale local weight-sharing convolutional neural network structure to fully extract the temporal features within each dimension of the multi-dimensional time series. Then, we model the features from various aspects through multiple attention heads, extracting the relationship features between dimensions. We map the attention output results to the latent space distribution of the VAE and propose an optimization method to improve the reconstruction performance of the VAE, detecting anomalies through reconstruction errors. Regarding anomaly interpretability, we utilize the VAE probability distribution characteristics, decompose the obtained joint probability density into conditional probabilities on each dimension, and calculate the anomaly score, which provides helpful value for technicians. Experimental results show that our algorithm performed best in terms of F1 score and AUC value. The AUC value for anomaly detection is 0.982, and the F1 score is 0.905, which is 4% higher than the best-performing baseline algorithm, Transformer with a Discriminator for Anomaly Detection (TDAD). It also provides accurate anomaly interpretation capability. Full article

The recently introduced complex-valued convolutional neural network (CV-CNN) has shown considerable advancements for polarimetric synthetic aperture radar (PolSAR) image classification by effectively incorporating both magnitude and phase information. However, a solitary 2D or 3D CNN encounters challenges such as insufficiently extracting scattering channel dimension features or excessive computational parameters. Moreover, these networks’ default is that all information is equally important, consuming vast resources for processing useless information. To address these issues, this study presents a new hybrid CV-CNN with the attention mechanism (CV-2D/3D-CNN-AM) to classify PolSAR ground objects, possessing both excellent computational efficiency and feature extraction capability. In the proposed framework, multi-level discriminative features are extracted from preprocessed data through hybrid networks in the complex domain, along with a special attention block to filter the feature importance from both spatial and channel dimensions. Experimental results performed on three PolSAR datasets demonstrate our present approach’s superiority over other existing ones. Furthermore, ablation experiments confirm the validity of each module, highlighting our model’s robustness and effectiveness. Full article

In response to the limitations of existing evaluation methods for gas well types in tight sandstone gas reservoirs, characterized by low indicator dimensions and a reliance on traditional methods with low prediction accuracy, therefore, a novel approach based on a two-dimensional convolutional neural network (2D-CNN) is proposed for predicting gas well types. First, gas well features are hierarchically selected using variance filtering, correlation coefficients, and the XGBoost algorithm. Then, gas well types are determined via spectral clustering, with each gas well labeled accordingly. Finally, the selected features are inputted, and classification labels are outputted into the 2D-CNN, where convolutional layers extract features of gas well indicators, and the pooling layer, which, trained by the backpropagation of CNN, performs secondary dimensionality reduction. A 2D-CNN gas well classification prediction model is constructed, and the softmax function is employed to determine well classifications. This methodology is applied to a specific tight gas reservoir. The study findings indicate the following: (1) Via two rounds of feature selection using the new algorithm, the number of gas well indicator dimensions is reduced from 29 to 15, thereby reducing the computational complexity of the model. (2) Gas wells are categorized into high, medium, and low types, addressing a deep learning multi-class prediction problem. (3) The new method achieves an accuracy of 0.99 and a loss value of 0.03, outperforming BP neural networks, XGBoost, LightGBM, long short-term memory networks (LSTMs), and one-dimensional convolutional neural networks (1D-CNNs). Overall, this innovative approach demonstrates superior efficacy in predicting gas well types, which is particularly valuable for tight sandstone gas reservoirs. Full article

Emotion recognition using the electroencephalogram (EEG) has garnered significant attention within the realm of human–computer interaction due to the wealth of genuine emotional data stored in EEG signals. However, traditional emotion recognition methods are deficient in mining the connection between multi-domain features and fitting their advantages. In this paper, we propose a novel capsule Transformer network based on a multi-domain feature for EEG-based emotion recognition, referred to as MES-CTNet. The model’s core consists of a multichannel capsule neural network(CapsNet) embedded with ECA (Efficient Channel Attention) and SE (Squeeze and Excitation) blocks and a Transformer-based temporal coding layer. Firstly, a multi-domain feature map is constructed by combining the space–frequency–time characteristics of the multi-domain features as inputs to the model. Then, the local emotion features are extracted from the multi-domain feature maps by the improved CapsNet. Finally, the Transformer-based temporal coding layer is utilized to globally perceive the emotion feature information of the continuous time slices to obtain a final emotion state. The paper fully experimented on two standard datasets with different emotion labels, the DEAP and SEED datasets. On the DEAP dataset, MES-CTNet achieved an average accuracy of 98.31% in the valence dimension and 98.28% in the arousal dimension; it achieved 94.91% for the cross-session task on the SEED dataset, demonstrating superior performance compared to traditional EEG emotion recognition methods. The MES-CTNet method, utilizing a multi-domain feature map as proposed herein, offers a broader observation perspective for EEG-based emotion recognition. It significantly enhances the classification recognition rate, thereby holding considerable theoretical and practical value in the EEG emotion recognition domain. Full article

To meet the demand for rapid bacterial detection in clinical practice, this study proposed a joint determination model based on spectral database matching combined with a deep learning model for the determination of positive–negative bacterial infection in directly smeared urine samples. Based on a dataset of 8124 urine samples, a standard hyperspectral database of common bacteria and impurities was established. This database, combined with an automated single-target extraction, was used to perform spectral matching for single bacterial targets in directly smeared data. To address the multi-scale features and the need for the rapid analysis of directly smeared data, a multi-scale buffered convolutional neural network, MBNet, was introduced, which included three convolutional combination units and four buffer units to extract the spectral features of directly smeared data from different dimensions. The focus was on studying the differences in spectral features between positive and negative bacterial infection, as well as the temporal correlation between positive–negative determination and short-term cultivation. The experimental results demonstrate that the joint determination model achieved an accuracy of 97.29%, a Positive Predictive Value (PPV) of 97.17%, and a Negative Predictive Value (NPV) of 97.60% in the directly smeared urine dataset. This result outperformed the single MBNet model, indicating the effectiveness of the multi-scale buffered architecture for global and large-scale features of directly smeared data, as well as the high sensitivity of spectral database matching for single bacterial targets. The rapid determination solution of the whole process, which combines directly smeared sample preparation, joint determination model, and software analysis integration, can provide a preliminary report of bacterial infection within 10 min, and it is expected to become a powerful supplement to the existing technologies of rapid bacterial detection. Full article

Yield prediction is a crucial activity in scheduling agronomic operations and in informing the management and financial decisions of a wide range of stakeholders of the organic durum wheat supply chain. This research aims to develop a yield forecasting system by combining vegetation index time-series data from Sentinel-2 L2A time-series data, field-measured yields, and deep learning techniques. Remotely sensed data over a season could be, in general, noisy and characterized by a variable density due to weather conditions. This problem was mitigated using Functional Principal Component Analysis (FPCA). We obtained a functional representation of acquired data, and starting from this, we tried to apply deep learning to predict the crop yield. We used a Convolutional Neural Network (CNN) approach, starting from images that embed temporal and spectral dimensions. This representation does not require one to a priori select a vegetation index that, typically, is task-dependent. The results have been also compared with classical approaches as Partial Least Squares (PLS) on the main reference vegetation indexes such as the Normalized Difference Vegetation Index (NDVI) and Normalized Difference Red Edge index (NDRE), considering both in-season and end-season scenarios. The obtained results show that the image-based representation of multi-spectral time series could be an effective method to estimate the yield, also, in the middle stage of cropping with R² values greater than 0.83. The developed model could be used to estimate yield the neighbor fields characterized by similar setups in terms of the crop, variety, soil, and, of course, management. Full article

Efficient stock status analysis and forecasting are important for stock market participants to be able to improve returns and reduce associated risks. However, stock market data are replete with noise and randomness, rendering the task of attaining precise price predictions arduous. Moreover, the lagging phenomenon of price prediction makes it hard for the corresponding trading strategy to capture the turning points, resulting in lower investment returns. To address this issue, we propose a framework for Important Trading Point (ITP) prediction based on Return-Adaptive Piecewise Linear Representation (RA-PLR) and a Batch Attention Multi-Scale Convolution Recurrent Neural Network (Batch-MCRNN) with the starting point of improving stock investment returns. Firstly, a novel RA-PLR method is adopted to detect historical ITPs in the stock market. Then, we apply the Batch-MCRNN model to integrate the information of the data across space, time, and sample dimensions for predicting future ITPs. Finally, we design a trading strategy that combines the Relative Strength Index (RSI) and the Double Check (DC) method to match ITP predictions. We conducted a comprehensive and systematic comparison with several state-of-the-art benchmark models on real-world datasets regarding prediction accuracy, risk, return, and other indicators. Our proposed method significantly outperformed the comparative methods on all indicators and has significant reference value for stock investment. Full article