Search Results (63)

In recent years, the research on human activity recognition based on channel state information (CSI) of Wi-Fi has gradually attracted much attention in order to avoid the deployment of additional devices and reduce the risk of personal privacy leakage. In this paper, we propose a hybrid network architecture, named VBCNet, that can effectively identify human activity postures. Firstly, we extract CSI sequences from each antenna of Wi-Fi signals, and the data are preprocessed and tokenised. Then, in the encoder part of the model, we introduce a layer of long short-term memory network to further extract the temporal features in the sequences and enhance the ability of the model to capture the temporal information. Meanwhile, VBCNet employs a convolutional feed-forward network instead of the traditional feed-forward network to enhance the model’s ability to process local and multi-scale features. Finally, the model classifies the extracted features into human behaviours through a classification layer. To validate the effectiveness of VBCNet, we conducted experimental evaluations on the classical human activity recognition datasets UT-HAR and Widar3.0 and achieved an accuracy of 98.65% and 77.92%. These results show that VBCNet exhibits extremely high effectiveness and robustness in human activity recognition tasks in complex scenarios. Full article

With the load center’s continuous expansion and development of the AC power grid’s scale and construction, the recipient grid under the multi-feed DC environment is facing severe challenges of DC commutation failure and bipolar blocking due to the high strength of AC-DC coupling and the low level of system inertia, which brings many complexities and uncertainties to economic scheduling. In addition, the large-scale grid integration of wind power, photovoltaic, and other intermittent energy sources makes the ultra-high-voltage (UHV) DC channel operation state randomized. The deterministic scenario-based timing power simulation is no longer suitable for the current complex and changeable grid operation state. In this paper, we first start with the description and analysis of the uncertainty in renewable energy (RE) sources, such as wind and solar, and reconstruct the time-sequence power model by using the stochastic differential equation model. Then, a carbon emission trading cost (CET) model is constructed based on the CET mechanism, and the two-stage locating and capacity optimization model for the UHV DC receiving end is proposed under the constraint of dispatch safety and stability. Among them, the first stage starts with the objective of maximizing the carrying capacity of the UHV DC receiving end grid; the second stage checks its dynamic safety under the basic and fault modes according to the results of the first stage and corrects the drop point and capacity of the UHV DC line with the objective of achieving safe and stable UHV DC operation at the lowest economic investment. In addition, the two-stage model innovatively proposes UHV DC relative inertia constraints, peak adjustment margin constraints, transient voltage support constraints under commutation failure conditions, and frequency support constraints under a DC blocking state. In addition, to address the problem that the probabilistic constraints of the scheduling model are difficult to solve, the discrete step-size transformation and convolution sequence operation methods are proposed to transform the chance-constrained planning into mixed-integer linear planning for solving. Finally, the proposed model is validated with a UHV DC channel in 2023, and the results confirm the feasibility and effectiveness of the model. Full article

The Czochralski method is the primary technique for single-crystal silicon production. However, anomalous states such as crystal loss, twisting, swinging, and squareness frequently occur during crystal growth, adversely affecting product quality and production efficiency. To address this challenge, we propose an enhanced multimodal fusion classification model for detecting and categorizing these four anomalous states. Our model initially transforms one-dimensional signals (diameter, temperature, and pulling speed) into time–frequency domain images via continuous wavelet transform. These images are then processed using a Dense-ECA-SwinTransformer network for feature extraction. Concurrently, meniscus images and inter-frame difference images are obtained from the growth system’s meniscus video feed. These visual inputs are fused at the channel level and subsequently processed through a ConvNeXt network for feature extraction. Finally, the time–frequency domain features are combined with the meniscus image features and fed into fully connected layers for multi-class classification. The experimental results show that the method can effectively detect various abnormal states, help the staff to make a more accurate judgment, and formulate a personalized treatment plan for the abnormal state, which can improve the production efficiency, save production resources, and protect the extraction equipment. Full article

Aiming to address the wear and replacement inconvenience of concrete mixer liners, this study utilizes a laser cladding system to clad Fe60 alloy powder on the liner. It investigates the influence of different process parameters on the forming quality of the Fe60 alloy powder cladding layer. The optimal process parameters were obtained by weighted comprehensive evaluation, and single-layer multi-pass cladding experiments were carried out under the optimal process parameters to investigate the effects of a 30%, 40%, and 50% lap rate on the surface flatness and forming quality of the cladding layer. Using a metallographic microscope, a scanning electron microscope analysis of the macro morphology and microstructure of the cladding layer was conducted, a DPT-5 penetration flaw detector was used to observe the cracks on the surface of the multi-channel cladding, a microhardness tester and friction and wear experimental machine were used for the hardness of the cladding layer, and an abrasive wear resistance test was conducted. The results show that under the process parameters of a laser power of 900 W, powder feeding speed of 7 g/min, scanning speed of 600 mm/min, and 50% lap rate, the average microhardness of the fused cladding layer reaches 742 HV, which is 1.8 times higher than that of the liner plate, and the coefficient of friction is 0.57, which improves the liner plate’s wear resistance performance and service life. Full article

Real-time estimation of fish biomass plays a crucial role in real-world fishery production, as it helps formulate feeding strategies and other management decisions. In this paper, a dense fish counting network called Swin-CSRNet is proposed. Specifically, the VGG16 layer in the front-end is replaced with the Swin transformer to extract image features more efficiently. Additionally, a squeeze-and-excitation (SE) module is introduced to enhance feature representation by dynamically adjusting the importance of each channel through “squeeze” and “excitation”, making the extracted features more focused and effective. Finally, a multi-scale fusion (MSF) module is added after the back-end to fully utilize the multi-scale feature information, enhancing the model’s ability to capture multi-scale details. The experiment demonstrates that Swin-CSRNet achieved excellent results with MAE, RMSE, and MAPE and a correlation coefficient R² of 11.22, 15.32, 5.18%, and 0.954, respectively. Meanwhile, compared to the original network, the parameter size and computational complexity of Swin-CSRNet were reduced by 70.17% and 79.05%, respectively. Therefore, the proposed method not only counts the number of fish with higher speed and accuracy but also contributes to advancing the automation of aquaculture. Full article

In autonomous driving, environmental perception technology often encounters challenges such as false positives, missed detections, and low accuracy, particularly in detecting small objects and complex scenarios. Existing algorithms frequently suffer from issues like feature redundancy, insufficient contextual interaction, and inadequate information fusion, making it difficult to perform multi-task detection and segmentation efficiently. To address these challenges, this paper proposes an end-to-end multi-task environmental perception model named YOLO-Mg, designed to simultaneously perform traffic object detection, lane line detection, and drivable area segmentation. First, a multi-stage gated aggregation network (MogaNet) is employed during the feature extraction process to enhance contextual interaction by improving diversity in the channel dimension, thereby compensating for the limitations of feed-forward neural networks in contextual understanding. Second, to further improve the model’s accuracy in detecting objects of various scales, a restructured weighted bidirectional feature pyramid network (BiFPN) is introduced, optimizing cross-level information fusion and enabling the model to handle object detection at different scales more accurately. Finally, the model is equipped with one detection head and two segmentation heads to achieve efficient multi-task environmental perception, ensuring the simultaneous execution of multiple tasks. The experimental results on the BDD100K dataset demonstrate that the model achieves a mean average precision (mAP50) of 81.4% in object detection, an Intersection over Union (IoU) of 28.9% in lane detection, and a mean Intersection over Union (mIoU) of 92.6% in drivable area segmentation. The tests conducted in real-world scenarios show that the model performs effectively, significantly enhancing environmental perception in autonomous driving and laying a solid foundation for safer and more reliable autonomous driving systems. Full article

Recently, transformer-based face super-resolution (FSR) approaches have achieved promising success in restoring degraded facial details due to their high capability for capturing both local and global dependencies. However, while existing methods focus on introducing sophisticated structures, they neglect the potential feature map information, limiting FSR performance. To circumvent this problem, we carefully design a pair of guiding blocks to dig for possible feature map information to enhance features before feeding them to transformer blocks. Relying on the guiding blocks, we propose a spatial-channel mutual attention-guided transformer network for FSR, for which the backbone architecture is a multi-scale connected encoder–decoder. Specifically, we devise a novel Spatial-Channel Mutual Attention-guided Transformer Module (SCATM), which is composed of a Spatial-Channel Mutual Attention Guiding Block (SCAGB) and a Channel-wise Multi-head Transformer Block (CMTB). SCATM on the top layer (SCATM-T) aims to promote both local facial details and global facial structures, while SCATM on the bottom layer (SCATM-B) seeks to optimize the encoded features. Considering that different scale features are complementary, we further develop a Multi-scale Feature Fusion Module (MFFM), which fuses features from different scales for better restoration performance. Quantitative and qualitative experimental results on various datasets indicate that the proposed method outperforms other state-of-the-art FSR methods. Full article

To achieve the anticipated performance of massive multiple input multiple output (MIMO) systems in wireless communication, it is imperative that the user equipment (UE) accurately feeds the channel state information (CSI) back to the base station (BS) along the uplink. To reduce the feedback overhead, an increasing number of deep learning (DL)-based networks have emerged, aimed at compressing and subsequently recovering CSI. Various novel structures are introduced, among which Transformer architecture has enabled a new level of precision in CSI feedback. In this paper, we propose a new method named TransNet+ built upon the Transformer-based TransNet by updating the multi-head attention layer and implementing an improved training scheme. The simulation results demonstrate that TransNet+ outperforms existing methods in terms of recovery accuracy and achieves state-of-the-art. Full article

Maize is the crop which is most commonly exposed to toxigenic fungi that produce many toxins that are harmful to humans and animals alike. Preharvest grain yield loss, preharvest toxin contamination (at harvest), and storage loss are estimated to be between 220 and 265 million metric tons. In the past ten years, the preharvest mycotoxin damage was stable or increased mainly in aflatoxin and fumonisins. The presence of multiple toxins is characteristic. The few breeding programs concentrate on one of the three main toxigenic fungi. About 90% of the experiments except AFB1 rarely test toxin contamination. As disease resistance and resistance to toxin contamination often differ in regard to F. graminearum, F. verticillioides, and A. flavus and their toxins, it is not possible to make a food safety evaluation according to symptom severity alone. The inheritance of the resistance is polygenic, often mixed with epistatic and additive effects, but only a minor part of their phenotypic variation can be explained. All tests are made by a single inoculum (pure isolate or mixture). Genotype ranking differs between isolates and according to aggressiveness level; therefore, the reliability of such resistance data is often problematic. Silk channel inoculation often causes lower ear rot severity than we find in kernel resistance tests. These explain the slow progress and raise skepticism towards resistance breeding. On the other hand, during genetic research, several effective putative resistance genes were identified, and some overlapped with known QTLs. QTLs were identified as securing specific or general resistance to different toxicogenic species. Hybrids were identified with good disease and toxin resistance to the three toxigenic species. Resistance and toxin differences were often tenfold or higher, allowing for the introduction of the resistance and resistance to toxin accumulation tests in the variety testing and the evaluation of the food safety risks of the hybrids within 2–3 years. Beyond this, resistance breeding programs and genetic investigations (QTL-analyses, GWAM tests, etc.) can be improved. All other research may use it with success, where artificial inoculation is necessary. The multi-toxin data reveal more toxins than we can treat now. Their control is not solved. As limits for nonregulated toxins can be introduced, or the existing regulations can be made to be stricter, the research should start. We should mention that a higher resistance to F. verticillioides and A. flavus can be very useful to balance the detrimental effect of hotter and dryer seasons on aflatoxin and fumonisin contamination. This is a new aspect to secure food and feed safety under otherwise damaging climatic conditions. The more resistant hybrids are to the three main agents, the more likely we are to reduce the toxin losses mentioned by about 50% or higher. Full article

Wide-band laser cladding technology has emerged as a solution to the limitations of traditional cladding techniques, which are small single-path dimensions and low processing efficiency. The existing wide-band cladding technology presents challenges related to the high precision required for the laser–powder coupling and the significant powder-divergence phenomenon. Based on the inside-beam powder-feeding technology, a wide-band powder-feeding nozzle was designed using the multi-channel powder flow shaping method. The size of the powder spot obtained at the processing location can reach 40 mm × 3 mm. A computational fluid dynamics analysis using the FLUENT software was conducted to investigate the impact of the nozzle’s structural parameters on the powder distribution. It was determined that the optimal configuration was achieved when the powder-feeding channel was 8, and the transverse and longitudinal dimensions for the collimating gas outlet were 0.5 mm and 1 mm, respectively. Among the process parameters, an increase in the carrier gas flow rate was found to effectively enhance the stability of powder transportation. However, the powder feed rate had minimal impact on the powder concentration distribution, and the collimating gas flow rate appeared to have a minimal effect on the divergence angle of the powder stream. Wide-band laser cladding experiments were conducted using the designed powder-feeding nozzle, and a single-path cladding with a width of 39.96 mm was finally obtained. Full article

Although the most widely studied datasets in fraud-detection systems belong to the banking sector, the aviation industry is susceptible to fraud activities that seriously harm airline companies. Therefore, big airline companies have started to purchase or develop their own fraud-detection systems in order to prevent their financial loss and prestige decline. Chronological order and temporal flow are intrinsically of high importance in fraud detection in the banking sector as well as in airline sale channels. Therefore, the transactions in the datasets used in fraud-detection systems should be evaluated not only according to the information they contain but also according to the past transactions they are linked to. One of the best ways to raise awareness about the connected past transactions to the fraud-detection system is to profile the data fields whose historical data is important and dynamically place these profiles on each transaction. In this study, we first draw the baseline, i.e., the first touch in this field, for fraud detection in aviation and then introduce a novel multi-modal profiling mechanism based on deep learning for the detection of fraudulent airline ticket activities. We achieved great success by feeding the new features obtained from those profiles into a deep neural network that is fine-tuned by adjusting the well-known hyperparameters regarding the aviation data. Thanks to the combination of profiling and deep learning, the F1 score of the proposed system reaches up to 89.3% and 93.2% in terms of quantity-based success and cost-based success, respectively. Full article

Medical image segmentation primarily utilizes a hybrid model consisting of a Convolutional Neural Network and sequential Transformers. The latter leverage multi-head self-attention mechanisms to achieve comprehensive global context modelling. However, despite their success in semantic segmentation, the feature extraction process is inefficient and demands more computational resources, which hinders the network’s robustness. To address this issue, this study presents two innovative methods: PTransUNet (PT model) and C-PTransUNet (C-PT model). The C-PT module refines the Vision Transformer by substituting a sequential design with a parallel one. This boosts the feature extraction capabilities of Multi-Head Self-Attention via self-correlated feature attention and channel feature interaction, while also streamlining the Feed-Forward Network to lower computational demands. On the Synapse public dataset, the PT and C-PT models demonstrate improvements in DSC accuracy by 0.87% and 3.25%, respectively, in comparison with the baseline model. As for the parameter count and FLOPs, the PT model aligns with the baseline model. In contrast, the C-PT model shows a decrease in parameter count by 29% and FLOPs by 21.4% relative to the baseline model. The proposed segmentation models in this study exhibit benefits in both accuracy and efficiency. Full article

In order to minimize the gaps between the direct laser deposition channels and improve the quality and performance of the formed parts, the process of direct laser deposition is utilized in laser additive manufacturing to create sequential, single- and double-channel deposition layers on 304 stainless steel plates. Under the premise of keeping the layer rate and defocusing amount unchanged, this study investigates the effects of laser power, scanning speed, and powder feeding rate on the morphology and inclusions of single- and double-channel deposited layers. The aim is to determine the optimal process parameter values for direct laser deposition of single-layer, single-channel Inconel 738. The effects of the three process parameters on the response values were investigated using a multi-factor, multi-level experimental design. The evaluation indexes for the analysis included the deposited layer wetting angle and aspect ratio. The analysis involved one-way extreme analysis and ANOVA analysis. The optimal process parameters are a laser power of 550~750 W, a scanning speed of 7~13 mm/s, and the powder feeding rate was 2.1~4.33 g/min. At the same time, the relationship between surface tension and gravity was integrated with the spherical coronal model and Young’s equation to develop a mathematical model of the direct laser deposition process at a theoretical level. The mathematical model of the direct laser deposition process was utilized to analyze the correlation between the geometric parameters of the cross-section of the deposited layer. This analysis provides a valuable data reference for future Inconel 738 direct laser deposition. Full article

Images captured during marine engineering operations suffer from color distortion and low contrast. Underwater image enhancement helps to alleviate these problems. Many deep learning models can infer multi-source data, where images with different perspectives exist from multiple sources. To this end, we propose a multichannel deep convolutional neural network (MDCNN) linked to a VGG that can target multi-source (multi-domain) underwater image enhancement. The designed MDCNN feeds data from different domains into separate channels and implements parameters by linking VGGs, which improves the domain adaptation of the model. In addition, to optimize performance, multi-domain image perception loss functions, multilabel soft edge loss for specific image enhancement tasks, pixel-level loss, and external monitoring loss for edge sharpness preprocessing are proposed. These loss functions are set to effectively enhance the structural and textural similarity of underwater images. A series of qualitative and quantitative experiments demonstrate that our model is superior to the state-of-the-art Shallow UWnet in terms of UIQM, and the performance evaluation conducted on different datasets increased by 0.11 on average. Full article

The larvae of certain wood-boring beetles typically inhabit the interior of trees and feed on the wood, leaving almost no external traces during the early stages of infestation. Acoustic techniques are commonly employed to detect the vibrations produced by these larvae while they feed on wood, significantly increasing detection efficiency compared to traditional methods. However, this method’s accuracy is greatly affected by environmental noise interference. To address the impact of environmental noise, this paper introduces a signal separation system based on a multi-channel attention mechanism. The system utilizes multiple sensors to receive wood-boring vibration signals and employs the attention mechanism to adjust the weights of relevant channels. By utilizing beamforming techniques, the system successfully removes noise from the wood-boring vibration signals and separates the clean wood-boring vibration signals from the noisy ones. The data used in this study were collected from both field and laboratory environments, ensuring the authenticity of the dataset. Experimental results demonstrate that this system can efficiently separate the wood-boring vibration signals from the mixed noisy signals. Full article