Calculation Module Research Articles

Adaptive Dynamic Threshold Graph Neural Network: A Novel Deep Learning Framework for Cross-Condition Bearing Fault Diagnosis

Recently, bearing fault diagnosis methods based on deep learning have achieved significant success. However, in practical engineering applications, the limited labeled data and various working conditions severely constrain the widespread application of most deep-learning-based fault diagnosis methods. Additionally, many methods focus solely on the amplitude information of samples, neglecting the rich relational information between samples. To address these issues, this paper proposes a novel cross-condition few-shot fault diagnosis method based on an adaptive dynamic threshold graph neural network (ADTGNN). The aim of the proposed method is to rapidly identify fault types after they occur only a few times or even once. The adaptive threshold computation module (ATCM) in ADTGNN dynamically assigns thresholds to each edge based on edge confidence, optimizing the graph structure and effectively alleviating the over-smoothing issue. Furthermore, a dynamic threshold adjustment strategy (DTAS) is introduced to gradually increase the threshold with the training iterations, preventing the model from prematurely discarding crucial edges due to insufficient performance. The proposed model’s effectiveness is demonstrated using three bearing datasets. The experimental results indicate that the proposed approach significantly outperforms other comparison methods in cross-condition bearing fault diagnosis.

Lightweight 2D Human Pose Estimation Based on Joint Channel Coordinate Attention Mechanism

Traditional human pose estimation methods typically rely on complex models and algorithms. Lite-HRNet can achieve an excellent performance while reducing model complexity. However, its feature extraction scale is relatively single, which can lead to lower keypoints’ localization accuracy in crowded and complex scenes. To address this issue, we propose a lightweight human pose estimation model based on a joint channel coordinate attention mechanism. This model provides a powerful information interaction channel, enabling features of different resolutions to interact more effectively. This interaction can solve the problem of human pose estimation in complex scenes and improve the robustness and accuracy of the pose estimation model. The introduction of the joint channel coordinate attention mechanism enables the model to more effectively retain key information, thereby enhancing keypoints’ localization accuracy. We also redesign the lightweight basic module using the shuffle module and the joint channel coordinate attention mechanism to replace the spatial weight calculation module in the original Lite-HRNet model. By introducing this new module, we not only improve the network calculation speed and reduce the number of parameters of the entire model, but also ensure the accuracy of the model, thereby achieving a balance between performance and efficiency. We compare this model with current mainstream methods on the COCO and MPII dataset. The experimental results show that this model can effectively reduce the number of parameters and computational complexity while ensuring high model accuracy.

Application of CFD and nodal expansion method for simulation and analysis of boron dilution accident

Nuclear power, as a low-carbon energy source, provides the steady and reliable electricity. In this power source, safe operation plays a significant role from the point of view of design and economic cost. While nuclear power plants are designed to be safe in their operation and safe in the event of any malfunction or accident, some accidents with considerable occurrence frequency should be considered to provide related safety system and operation. The malfunction of boron injection system is one of this accident affecting the chemical and neutronic characteristics of cooling fluid. The aim of this study is the simulation and analysis of the boron dilution accident as one of Design Basis Accident, by considering the heterogeneous distribution of boron. For this purpose, CFD methods were used for exact evaluation of the boron distribution in each fuel assembly. Also, cross-section correction module and core calculation module are respectively developed by a cell calculation code and nodal expansion method. These modules are employed to apply absorber effects including boron and control banks and calculate 3D relative power. The results reveal the effect of heterogeneous distribution of boron in reactor pressure vessel and non-uniform chemical characteristics of coolant across fuel channels.

Effects of Ground Subsidence on Permafrost Simulation Related to Climate Warming

We develop a moving-mesh permafrost model that contains a ground subsidence computation module to estimate the effects of ground subsidence on permafrost simulation under different warming scenarios. Including the ground subsidence process in the permafrost simulation produces only a relatively small improvement in the simulation performance of the ground temperature field, as validated by observations from two sites on the Qinghai–Tibetan Plateau (QTP). It is shown that ignoring ground subsidence tends to achieve a larger active layer thickness (ALT) but a smaller original thickness of permafrost that has thawed when simulating permafrost changes in a warming climate. The heat consumed by permafrost changes will be underestimated in simulations that do not consider ground subsidence. The effects that ground subsidence exerts within the permafrost simulation are clearly demonstrated under a strong warming scenario, which will influence the global energy budget. Projections indicate that the permafrost in the continuous permafrost area of the QTP may be close to the phase transition temperature to become zero thermal gradients in 2030–2040 under the SSP5-8.5 scenario, and there will be a great risk of ground subsidence by that stage. For permafrost regions with rich ground ice, the downward propagating temperature signals caused by ground subsidence are more attenuated. However, the heat calculation error will be larger in a simulation that does not consider ground subsidence there. This study quantifies the effects of ground subsidence, which can provide a better understanding of the permafrost thaw and energy budget of the QTP.

Развитие CAE системы численного моделирования НДС SIGMA GT

The paper presents the directions and results of developing the Sigma GT software suite designed for numerical modeling of the stress-and-strain state of a rock mass using the finite element method. It describes the process of computational module optimization to ensure the possibility of refined simulation in subareas of the original model. The process of storing coarse grid matrices is upgraded, which optimizes the use of computational resources when calculating the stress-and-strain state of large-scale models. The trial-and-error procedure has been improved up to the system of hierarchical interrelation of different scale models, which allows to actualize the boundary conditions of individual block models when modifying the field model configuration. An extensible library of local models is presented, which makes it possible to quickly generate models of mining system elements with adequate boundary conditions. The expandable functionality of the software product is demonstrated, which allows to accelerate the process of finite element model creation. This tool allows the software user to generate local models of various configurations, minimizing time consumption and possible errors in preparation of the finite element mesh. Application and functionality of tools for post-processing of modeling results, both in 2D and 3D form, are described. Export of stress isolines allows saving the calculation data in the format used at the mining enterprise and combining them with the maps of mine working. 3D visualization of the calculated data contributes to a comprehensive assessment of the stress-and-strain state vector field and identification of the areas with critical stress and strain concentrations under a certain mining development option.

Estimation of photosynthetic parameters from hyperspectral images using optimal deep learning architecture

The maximum carboxylation rate (Vcmax) and maximum electron transport rate (Jmax) of leaves are crucial for comprehending carbon cycling in farmland. Nevertheless, estimating these photosynthetic parameters precisely and rapidly faces a considerable challenge. This study designed an optimal deep learning architecture that accurately extracts photosynthetic parameters from hyperspectral images and evaluated its stability across different crop species. Photosynthetic parameter models jointly driven by hyperspectral images and deep learning were compared with models jointly driven by one-dimensional reflectance and simple machine learning methods, as well as chlorophyll driven models. The proposed optimal deep learning architecture incorporated spatial attention and prior knowledge of spectral indices calculations. Our results demonstrated that the hyperspectral images and deep learning jointly driven models outperformed traditional models. Notably, incorporating the module of spatial attention and spectral indices calculation networks achieved the best estimations for Vcmax25 (R2: 0.86, RMSE: 10.18 μmol m−2 s−1) and Jmax25 (R2: 0.83, RMSE: 24.27 μmol m−2 s−1). In contrast, the performance of the one-dimension reflectance driven models deteriorated (R2: 0.43–0.58, RMSE: 19.71–39.99 μmol m−2 s−1). Moreover, the best architecture was interpreted. Weight analysis revealed that the hyperspectral information on the middle part of the leaf contributed most to photosynthetic parameters. Feature map analysis indicated that the spectral indices calculation module utilized the information of the visible light spectrum. Even though the migration of models across various crop species may lead to a slight degradation, the model performance remained satisfactory after fine-tuning. The proposed deep learning models, which used 3D hyperspectral images for estimation of photosynthetic parameters, outperformed the models jointly driven by 1D reflectance data and conventional machine learning algorithms. The results highlighted the significance of spatial information from hyperspectral images and prior knowledge through spectral indices calculations. Moreover, the stability and effectiveness of the proposed architecture remained excellent across different species. This study presents advanced and highly effective deep learning techniques for evaluating the photosynthetic capacity of crop leaves and modeling the carbon cycle.

Development of supporting platform for the fine flow characteristics of reactor core

This study presents the Supporting platform for reactor fine flow characteristics calculation and analysis (Cilian platform), a user-friendly tool that supports the analysis and optimization of pressurized water reactor (PWR) cores with mixing vanes using computational fluid dynamics (CFD) computing. The Cilian platform allows for easy creation and optimization of PWR's main CFD calculation schemes and autonomously manages CFD calculation and analysis of PWR cores, reducing the need for human and computational resources. The platform's key features enable efficient simulation, rapid solution design, automatic calculation of core scheme options, and streamlined data extraction and processing techniques. The Cilian platform's capability to call external CFD software reduces the development time and cost while improving the accuracy and reliability of the results. In conclusion, the Cilian platform exemplifies an innovative solution for efficient computational fluid dynamics analysis of pressurized water reactor (PWR) cores. It holds great promise for driving advancements in nuclear power technology, enhancing the safety, efficiency, and cost-effectiveness of nuclear reactors. The platform adopts a modular design methodology, enabling the swift and accurate computation and analysis of diverse flow regions within core components. This design approach facilitates the seamless integration of multiple computational modules across various reactor types, providing a high degree of flexibility and reusability.

Investigation of power quality issues in 14-bus electrical network with high penetration of renewable generation

Power quality issues arise in electrical networks when variable renewable energy (VRE) is integrated into them due to their random and intermittent nature which depends on weather conditions and other factors. The variation of solar irradiance throughout the day affects the energy produced by solar panels and the integration of solar power into electrical networks will result in changes and fluctuations in the voltage profile of buses. Reactive power compensation is required to improve the bus voltage levels of the electrical network to be within the required limits and the optimal allocation of reactive power compensation devices in the network is a complex problem to be investigated for the optimum injection of reactive power to obtain better voltage profiles for the entire network. This research investigated the penetration of variable solar energy into an electrical network in terms of voltage and reactive power flow. A variety of literature was reviewed in the scope of reactive power management in power systems and a gap in addressing the optimal allocation of compensation devices in the IEEE-14 bus was addressed based on the proposed methods followed by discussions of the results in terms of voltage profiles and reactive power flow in the buses. The objective is to produce an output power of higher quality and reliability for the loads so that intermittent sources of renewable energy can be more competent with energy sources such as fossil fuels that do not depend on weather conditions. Integration of methods using compensation optimisation (optimal allocation of capacitors) and volt-var regulation (smart inverter) to improve the voltage profile that was dropped and the fluctuations after penetration of solar power were carried out. A solar bus with variable energy generation was connected to the IEEE-14 bus to study the voltage variations. This was executed by the power flow calculation module to determine the voltages and reactive power in the buses of the network. With the optimum allocation of the capacitors, the voltage levels in all weak buses of the IEEE-14 bus were increased to be between 0.95 p.u. and 1.05 p.u. which was the voltage specifications of the Malaysian Grid Code Requirements. The voltage for every weak bus in the IEEE-14 bus showed a rise of 5.7% from 7 a.m. to 12 p.m. With that, the volt-var function was used for reactive power regulation at the point of common coupling (PCC) and a reduction of voltage deviation of 2.828 to 1.3% in the IEEE-14 bus was observed. The average voltage profile of all buses managed to attain a value of 98.99% from 95.673% (with solar power) with the optimal allocation of capacitors and volt-var regulation. The beneficiaries of this project will be the Sustainable Energy Development Authority (SEDA) which administers the Net Energy Metering (NEM) scheme and Tenaga Nasional Berhad (TNB) which is the Malaysian multinational electrical company focused primarily on the generation, transmission, and distribution of electricity in Peninsular Malaysia. The Energy Commission and Ministry of Energy and Natural Resources are also beneficiaries as they carried out a competitive bidding programme for large-scale solar (LSS) known as the LSS@MEnTARI or LSSPV4 to attain bids for the development of around 1000 MW AC of LSS power plants to be operational in Malaysia by 2022. This work will also be beneficial in future research in planning reactive power compensation devices in networks of multiple VRE sources, communication, and coordinated control of smart inverters, and incorporation of these devices for smart grid applications.

Real-Time Synchronous Acquisition and Processing of Signal in Coherent Doppler Wind Lidar Using FPGA

Coherent Doppler wind lidar has become a primary remote sensing technique for measuring atmospheric wind fields in recent years. However, the high bandwidth of the time domain echo signal has limited real-time data acquisition and processing. In this work, we propose a real-time data acquisition and preprocessing integrated solution. This approach is implemented using on-chip system field programmable gate array (FPGA) hardware while utilizing a 7-channel base-4 polyphase fast Fourier transform calculation module and a pipelined structure operation. Within a 1 s data collection time, the real-time rapid acquisition and spectral preprocessing of lidar echo signals at a range of 9.9 km can be achieved. We compare the observation data with the measurement data from Windcube100s and radiosondes, and the results indicate that the coherent wind lidar developed in this paper can detect heights above 4.0 km, with effective data acquisition rates of 82% and 61% in the height ranges of 1500–2000 m and 2000–2500 m, respectively. At the height range of 3.5 km, the correlations for horizontal wind speed and direction measured by the lidar were all above 0.99, with the wind speed and direction measurement deviations being better than 0.56 m/s and 8.40∘, respectively.

Automatic Detection and Counting of Stacked Eucalypt Timber Using the YOLOv8 Model

The objective of this project was to automate the detection and counting process of stacked eucalypt (hybrid Eucalyptus urophylla x Eucalyptus grandis) timber in the forestry industry using the YOLOv8 model. The dataset consists of 230 diverse images of eucalypt roundwood, including images of roundwood separated on a rail and stacked timber. The annotations were made using LabelImg, ensuring accurate delineation of target objects on the log surfaces. The YOLOv8 model is customized with a CSPDarknet53 backbone, C2f module, and SPPF layer for efficient computation. The model was trained using an AdamW optimizer and implemented using Ultralytics YOLOv8.0.137, Python-3.10.12, and torch-2.0.1 + cu118 with CUDA support on NVIDIA T1000 (4096MiB). For model evaluation, the precision, recall, and mean Average Precision at a 50% confidence threshold (mAP50) were calculated. The best results were achieved at epoch 261, with a precision of 0.814, recall of 0.812, and mAP50 of 0.844 on the training set and a precision of 0.778, recall of 0.798, and mAP50 of 0.839 on the validation set. The model’s generalization was tested on separate images, demonstrating robust detection and accurate counting. The model effectively identified roundwood that was widely spaced, scattered, and overlapping. However, when applied to stacked timber, the automatic counting was not very accurate, especially when using images. In contrast, when using video, the relative percentage error for automatic counting significantly decreased to −12.442%. In conclusion, video proved to be more effective than images for counting stacked timber, while photographs should be reserved for the counting of individual roundwood pieces.

Effect of Adding Global Geopotential Model Generated Data to Terrestrial Data on Geoid Model (case study at Bekasi, Indonesia)

Geoid modeling requires accurate and evenly distributed gravity data throughout the study area. Although airborne gravity measurements are getting more massive, some areas are politically restricted areas. Therefore, terrestrial measurements are still needed, such as in Bekasi City and its surroundings. In terrestrial geoid modeling, ideal conditions are often difficult to achieve, because terrestrial measurements can only be made in accessible, safe, and permitted places. These non-ideal conditions result in an uneven distribution of data which can decrease in the accuracy of the resulting geoid. One way to overcome this data acquisition problem is to fill in the data using synthetic data calculated from the global geopotential model data. This study aimed to analyze the effect of using EGM data on geoid accuracy. The data needed for this research were terrestrial gravity data, EGM2008, and SRTM30 Plus, and GEBCO. The geoid was calculated using several calculation modules from Gravsoft. Specific modifications were made using Simply Fortran to facilitate geoid modelling with integral of Hotine. This research was conducted with two scenarios, namely geoid calculations using only terrestrial gravity data and geoid calculations using a combination of terrestrial data and EGM2008. The accuracy of the geoid was analyzed based on the magnitude of the deviation between the local Bekasi geoid and the national geoid represented by INAGEOID2020 2020 version 2. Free Air gravity anomaly provide by Badan Informasi Geospasial were converted to be gravity disturbance data on the earth’s surface. The secondary data gravity disturbance was calculated from EGM2008 with n=360 and n=2190. The geoid calculated from terrestrial geoid data alone produced the largest offset to the national geoid. Combination of terrestrial gravity data and generated data from EGM2008 n=360 performed higher precision geoid.

Herramienta educativa: software didáctico en la enseñanza de ingeniería en intercambiadores de calor

The Information and Communication Technologies (ICT) have a great potential to transform the field of engineering, and their use is increasingly widespread. In this context, the software (simulator) consists of three calculation modules: two of them focus on the simulation and evaluation of a shell-and-tube heat exchanger, and the third one centers on thermal-hydraulic design. The choice of the module to be used within the program depends on the user's needs. The freely accessible software allows teachers and students to adjust parameters, perform parametric sensitivity analyses, enabling a better understanding of the effects of variables involved in the equipment's performance. The software incorporates equations, graphs, and theoretical support for in-depth analysis of variables, as well as potential adjustments that can be made to meet the thermal-hydraulic requirements of the heat exchange unit. This makes it an educational tool that facilitates comprehensive analysis of the different variables involved. It aids in designing potential configurations for the required service of the shell-and-tube heat exchanger.

An efficient numerical simulation of coupled thermo-hydro-mechanical processes in deep shale gas reservoirs

Shale gas is an important unconventional natural gas resource, and it is an important target of exploration and development in recent years. Deep shale gas reservoirs are high-temperature, high-pressure, and high-stress with complex gas–water relationships, and its occurrence and flow mechanisms are still unclear, making the simulation of deep shale gas reservoirs still a challenging problem. Aiming at the key problems faced in the development of deep shale gas reservoirs, this paper establishes a flow-stress-temperature field coupled mathematical model and numerical model, studies the thermo-hydro-mechanical sequential decoupling method to solve the coupled mathematical model, and forms a multi-field coupled simulation technology which provides theoretical support for the development of deep shale gas. Based on the thermo-hydro-mechanical coupling calculation module, this paper conducts research on the influence of thermo-hydro-mechanical multi-field coupling parameters on the production and development performance, and it further verifies the practicability of our proposed model in real reservoirs.

Sustained Culture of Safety After Select Agent Exposure

Abstract Introduction/Objective In January of 2021 our laboratory experienced the occupational exposure of 3 technologists to the select agent Burkholderia pseudomallei due to improper recognition and manipulation outside a biosafety cabinet. Subsequently, an interactive computer module was created to train technologists on timely recognition, followed by a “fake” patient test to evaluate comprehension. In that initial period, 100% compliance and successful recognition was demonstrated. To evaluate sustained compliance and recognition, another wave of evaluation was performed Jan- Dec 2022. Methods/Case Report During the study period several new staff members were hired and demonstrated compliance with completing the computer module. Clinical isolates suspicious for a select agent were reported to the Medical Director on rounds daily and reviewed for potential exposure, which was defined as manipulation of the suspicious isolate outside of the biosafety cabinet. Additional challenge “fake” patients were circulated. Results (if a Case Study enter NA) During the study period, 16 isolates were deemed suspicious for select agents based on macroscopic and microscopic morphology and biochemical reactions. Of these, 0 resulted in laboratory exposures. Of the challenge isolates, 1 resulted in a near-miss potential for exposure due to the lack of recognition of macroscopic characteristics. Conclusion The use of an interactive computer module focused around timely recognition and safe handling of potential select agents resulted in sustained laboratory safety and avoidance of potential occupational exposures.

Low-Resolution Steel Surface Defects Classification Network Based on Autocorrelation Semantic Enhancement

Aiming at the problems of low-resolution steel surface defects imaging, such as defect type confusion, feature blurring, and low classification accuracy, this paper proposes an autocorrelation semantic enhancement network (ASENet) for the classification of steel surface defects. It mainly consists of a backbone network and an autocorrelation semantic enhancement module (ASE), in which the autocorrelation semantic enhancement module consists of three main learnable modules: the CS attention module, the autocorrelation computation module, and the contextual feature awareness module. Specifically, we first use the backbone network to extract the basic features of the image and then use the designed CS attention module to enhance the basic features. In addition, to capture different aspects of semantic objects, we use the autocorrelation module to compute the correlation between neighborhoods and contextualize the basic and augmented features to enhance the recognizability of the features. Experimental results show that our method produces significant results, and the classification accuracy reaches 96.24% on the NEU-CLS-64 dataset. Compared with ViT-B/16, Swin_t, ResNet50, Mobilenet_v3_small, Densenet121, Efficientnet_b2, and baseline, the accuracy is 9.43%, 5.15%, 4.87%, 3.34%, 3.28%, 3.01%, and 2.72% higher, respectively.

Fast Calculation on Atmospheric Rayleigh Scattering Extinction Optical Depth for Day/Night Band Radiative Transfer Model With Periodic Lunar Illumination

AbstractRayleigh scattering effect inspired by tiny atmospheric molecules generates significant extinction at relatively short wavelength, such as Day/Night Band (DNB) covering visible (VIS, 0.4–0.76 μm) and near‐infrared (NIR, 0.76–1.0 μm) wavelengths. This salient extinction effect coupled with periodic changes in lunar illumination and wide DNB will further lead to increased complexity in nocturnal fast radiative transfer model (RTM) calculation of DNB. For these reasons, this study is aimed at developing a precise and robust fast calculation method on Rayleigh scattering extinction optical depth (RayOD) for pertinent fast RTM calculation of DNB. Firstly, we coupled the lunar spectral irradiance model and the parameterized Rayleigh scattering extinction calculation module into a RTM for forward simulations. Moreover, we find that the errors in DNB‐averaged RayOD calculated from the classical fast algorithm from RTTOV (Radiative Transfer for TOVS) of training the DNB instrument coefficients (about 20% of relative error in reflectance) and the parameterization method using the central wavelength of DNB (about 5%) are too large to be tolerable in fast RTM. Finally, to address this issue, we skillfully conceptualize a new equivalent optimal central wavelength algorithm to retrieve a dynamically wavelength in DNB using observation scattering and lunar phase angles for rapidly calculating DNB‐averaged RayOD. The relative errors in simulated reflectance from coupled RTM aforementioned are subtle nuances, which are about 10−5∼10−6 order of magnitude. Besides, this robust algorithm can also correct the relative error of about 1% at dark surface caused by periodic changes in lunar illumination.

CSformer: Enhancing deep learning efficiency for intelligent IoT

The rapid development of deep learning technology has led to increasing demand for more intelligent, automated, and humanized Internet of Things (IoT) devices. Deep learning models, while endowing IoT devices with the capability to learn higher-level features, concurrently impose more demanding computational and storage prerequisites on the hardware. To tackle the challenge and enable the practical application of deep learning models in IoT devices, we propose a novel efficient Transformer called CSformer, which incorporates intra-layer cluster and inter-layer selection. Intra-layer cluster is performed using a k-means++ based generation algorithm to improve cluster accuracy. To address the issues of information loss caused by clustering, we propose cluster center information enhancement and clustering loss calculation modules. The inter-layer selection strategy selects tokens according to their contribution, consistently diminishes redundancy, and prioritizes the retention of crucial information. By consistently reducing the sequence length, the inter-layer selection significantly improves training speed and reduces the memory occupation of the model. The experimental results indicate that in two common scenarios for intelligent IoT, namely text classification and sequence labeling, CSformer significantly outperforms the baseline models. Specifically, in the text classification task, our model achieves an average 22.66% reduction in memory consumption, a 37.98% decrease in time consumption, and a superior 9.58% performance improvement compared to baseline models across six datasets. Additional experiments substantiate the efficacy of the intra-layer cluster and inter-layer selection modules, as demonstrated through ablation experiments, overall performance, and visualization. The intra-layer cluster module enhances the performance of existing models by achieving more precise clustering and mitigating information loss, leading to significant performance improvements. The inter-layer selection module enhances the efficiency of existing studies by reducing model memory consumption and improving computational efficiency through the selective retention of essential tokens. This can effectively facilitate future research in applying advanced deep learning models to intelligent IoT, expanding the range of application scenarios and tasks within the field of intelligent IoT.

Analyzing the Spread of Misinformation on Social Networks: A Process and Software Architecture for Detection and Analysis

The rapid dissemination of misinformation on social networks, particularly during public health crises like the COVID-19 pandemic, has become a significant concern. This study investigates the spread of misinformation on social network data using social network analysis (SNA) metrics, and more generally by using well known network science metrics. Moreover, we propose a process design that utilizes social network data from Twitter, to analyze the involvement of non-trusted accounts in spreading misinformation supported by a proof-of-concept prototype. The proposed prototype includes modules for data collection, data preprocessing, network creation, centrality calculation, community detection, and misinformation spreading analysis. We conducted an experimental study on a COVID-19-related Twitter dataset using the modules. The results demonstrate the effectiveness of our approach and process steps, and provides valuable insight into the application of network science metrics on social network data for analysing various influence-parameters in misinformation spreading.

HilbertSCNet: Self-attention networks for small target segmentation of aerial drone images

In recent years, many advanced semantic segmentation algorithms have made great progress in common datasets. However, in the drones aerial photography dataset, since some of the segmentation targets are very small in such high-resolution images, these current algorithms are not designed specifically for them resulting in their mediocre performance in the drones aerial photography dataset. To address the problem of difficult segmentation of these small targets, this paper proposes a new semantic segmentation network model-HilbertSCNet: (1) Combine the image dimensionality reduction algorithm of Hilbert curve traversal and the idea of dual pathway to design a new spatial computation module to solve the problem of small target information is easily lost in the downsampling; (2) Add a windowing algorithm in the image dimensionality reduction algorithm of Hilbert curve traversal to solve the problem that the computational complexity of the module is too high, which makes the module cannot be applied to high-resolution feature maps. Experiments show that the proposed network is very effective for segmentation of small targets under high-resolution maps such as drone aerial photography, with certain superiority and generalization. The overall segmentation performance is improved compared to the current new network, including 1.76% better than OCNet in MIoU and 4.55% better than the dedicated drone algorithm RCCT-ASPPNet.

Functional diversity of visual cortex improves constraint-free natural image reconstruction from human brain activity

Previous brain decoding studies using functional magnetic resonance imaging (fMRI) have greatly advanced our understanding of human visual coding and non-invasive brain-machine interfaces. However, most of these studies focus on classifying a limited number of image categories or reconstructing visual images with additional information, e.g., semantic categories and textual cues. Constraint-free visual reconstruction remains scarce. Here, we propose a generative network based on the functional diversity of the human visual cortex (FDGen) that takes multivariate brain activity as input and directly reconstructs natural images perceived by observers without any additional cues (semantic categories or textual description). Our FDGen is augmented by two bio-inspired computational modules. Based on the functional specializations of the human visual cortex, we propose a new function-based input module (FIM) that projects responses from different brain regions into separate feature spaces. Second, inspired by human attention, we construct a computational module to derive attentive feature weights at the function level to refine the feature map. These function-selection modules (FSMs) allow the network to dynamically select multiscale visual information during the generation process. We test FDGen on the popular fMRI datasets of natural images and achieve highly robust performance. Our work represents an important step forward in the development of fMRI-based brain decoding algorithms and highlights the utility of neuroscience theories in the design of deep learning models.