Filter Network Research Articles

Advanced measurement methods for high-power gallium nitride high electron mobility transistors.

The testing and modeling of semiconductor devices are the foundation of circuit design. The issue of high-power device testing urgently needs to be solved as the power level of the devices under test (DUTs) increases. This work proposes advanced measurement methods based on three aspects of "measuring capability, security, and stability" with a focus on the features of high output power, easy self-oscillation in mismatch tests, and safety risk in the measurement system of high-power transistors. In this paper, the wideband limiter and bias filter network are innovatively introduced to improve the stability and security of the measurement circuit. Meanwhile, the output signal of the DUT is suggested to be measured using a spectrum analyzer before the test to avoid damage to the circuit caused by the possible self-oscillation of the transistor. Moreover, an efficient test system of current parameters and S-parameters is developed, with coaxial fixtures offered to boost the test power capacity and cascade bridges adapted to satisfy the pulse operating conditions. Finally, based on the improved test methods, a gallium nitride high electron mobility transistor (GaN HEMT) with a gate width of 400 × 32μm and a power density of roughly 10 W/mm was tested. A relatively complete I-V curve and a S-parameter curve were obtained, demonstrating the effectiveness and applicability of the improved methods.

Integrated quantum communication network and vibration sensing in optical fibers

Communication and sensing technologies play crucial roles in various aspects of modern society. The seamless combination of communication and sensing systems has attracted significant interest in recent years. Without adding core devices, vibration-sensing functions can be integrated to build a quantum network with high efficiency and versatility. In this study, we propose and demonstrate a network architecture that integrates a downstream quantum access network (DQAN) and vibration sensing in optical fibers. By encoding the key information of eight users simultaneously on the sidemode quantum states of a single laser source and successively separating them using a specially designed narrow-bandwidth filter network, we achieved a secure and efficient DQAN with an average key rate of 1.94×104 bits per second over an 80 km single-mode fiber. Meanwhile, vibration locations with spatial resolutions of 131, 25, and 4 m at vibration frequencies of 100 Hz, 1 kHz, and 10 kHz, respectively, were implemented using the existing DQAN system infrastructure. The results indicate that the backward probe beam has a negligible effect on the DQAN system. Our integrated architecture provides a viable and cost-effective solution for building a quantum communication sensor network and paves the way for the functionality expansion of quantum communication networks.

EKFNet: edge-based Kalman filter network for real-time EEG signal denoising

Objective.Signal denoising methods based on deep learning have been extensively adopted on electroencephalogram devices. However, they are unable to deploy on edge-based portable or wearable (P/W) electronics due to the high computational complexity of the existed models. To overcome such issue, we propose an edge-based lightweight Kalman filter network (EKFNet) that does not require manual prior knowledge estimation.Approach.Specifically, we construct a multi-scale feature fusion module to capture multi-scale feature information and implicitly compute the prior knowledge. Meanwhile, we design an adaptive gain estimation module that incorporates long short-term memory and sequential channel attention module to dynamically predict the Kalman gain. Furthermore, we present an optimization strategy utilizing operator fusion and constant folding to reduce the model's computational overhead and memory footprint.Main results. Experimental results show that the EKFNet reduces the sum of the square of the distances by at least 12% and improves the cosine similarity by at least 2.2% over the state-of-the-art methods. Besides, the model optimization shortens the inference time by approximately 3.3×. The code of our EKFNet is available athttps://github.com/cathnat/EKFNet.Significance.By integrating Kalman filter with deep learning, the approach addresses the parameter-setting challenges in traditional algorithms while reducing computational overhead and memory consumption, which exhibits a good tradeoff between algorithm performance and computing power.

UAV Visual Object Tracking Based on Spatio-Temporal Context

To balance the real-time and robustness of UAV visual tracking on a single CPU, this paper proposes an object tracker based on spatio-temporal context (STCT). STCT integrates the correlation filter and Siamese network into a unified framework and introduces the target’s motion model, enabling the tracker to adapt to target scale variations and effectively address challenges posed by rapid target motion, etc. Furthermore, a spatio-temporal regularization term based on the dynamic attention mechanism is proposed, and it is introduced into the correlation filter to suppress the aberrance of the response map. The filter solution is provided through the alternating direction method of multipliers (ADMM). In addition, to ensure efficiency, this paper proposes the average maximum response value-related energy (AMRE) for adaptive tracking state evaluation, which considers the time context of the tracking process in STCT. Experimental results show that the proposed STCT tracker can achieve a favorable balance between tracking robustness and real-time performance for UAV object tracking while running at ∼38 frames/s on a low-cost CPU.

HGLA: Biomolecular Interaction Prediction based on Mixed High-Order Graph Convolution with Filter Network via LSTM and Channel Attention.

Predicting biomolecular interactions is significant for understanding biological systems. Most existing methods for link prediction are based on graph convolution. Although graph convolution methods are advantageous in extracting structure information of biomolecular interactions, two key challenges still remain. One is how to consider both the immediate and highorder neighbors. Another is how to reduce noise when aggregating high-order neighbors. To address these challenges, we propose a novel method, called mixed high-order graph convolution with filter network via LSTM and channel attention (HGLA), to predict biomolecular interactions. Firstly, the basic and high-order features are extracted respectively through the traditional graph convolutional network (GCN) and the two-layer Higher-Order Graph Convolutional Architectures via Sparsified Neighborhood Mixing (MixHop). Secondly, these features are mixed and input into the filter network composed of LayerNorm, SENet and LSTM to generate filtered features, which are concatenated and used for link prediction. The advantages of HGLA are: 1) HGLA processes high-order features separately, rather than simply concatenating them; 2) HGLA better balances the basic features and high-order features; 3) HGLA effectively filters the noise from high-order neighbors. It outperforms state-ofthe-art networks on four benchmark datasets. The codes are available at https://github.com/zznb123/HGLA.

Research on multi-submersible positioning prediction based on Kalman filter and LSTM neural network

Abstract Accurately predicting submarine positions is critical to ensure safe underwater navigation, especially in complex and dynamic marine environments. Traditional methods, such as the Kalman Filter, have been widely used in this area because of their ability to provide optimal state estimates by integrating dynamic models with sensor measurements. However, the Kalman Filter’s effectiveness diminishes in the face of the diverse and non-linear conditions present in underwater environments. This research proposes an enhanced prediction model that combines the Kalman Filter with Long Short-Term Memory (LSTM) neural networks to address these limitations. The proposed model leverages the strengths of both approaches: the recursive, real-time estimation capabilities of the Kalman Filter and the temporal dependency handling of LSTM networks. Through extensive simulations, the model performs better in predicting the three-dimensional trajectories of submersibles, accounting for uncertainties such as ocean currents and varying environmental conditions. The results indicate that this hybrid approach improves prediction accuracy and enhances adaptability and robustness in challenging marine scenarios. This study contributes to developing more reliable submersible positioning systems with potential applications in deep-sea exploration, underwater archaeology, and search and rescue operations.

Brain Tumor Detection through Image Fusion Using Cross Guided Filter and Convolutional Neural Network

This Data fusion has become a significant issue in diagnostic imaging, particularly in medical applications like radiation and guided image surgery. Medical image fusion aims to enhance the precision of tumor diagnosis, by preserving the salient information and characteristics of the original images in the fused image. It has been shown that guided filters are capable of maintaining edges well. In this paper, we propose a novel cross-guided filter-based fusion approach for multimodal medical images utilizing convolutional neural networks. The cross-guided filter is used in the proposed algorithm to extract the detailed features from the source images. Convolutional neural networks are used to generate the feature weights of source images derived from the detail layers. The weighted average rule is used to merge the source images based on these weights. We used thirty distinct types of medical images from diverse sources to compare the effectiveness of the proposed strategy to that of existing methods, both numerically and visually. The experimental findings demonstrated that, in terms of both objective evaluation and qualitative image quality, the suggested system performs better than other standard methods already in use. The quantitative results show that compared to existing methods under consideration for comparison, the proposed algorithm improves mutual information by 25%, image entropy by 9.5%, spatial frequency by 21%, standard deviation by 18.1%, structural similarity index by 30%, and edge strength of the fused image by 39%.

Image Stitching of Low-Resolution Retinography Using Fundus Blur Filter and Homography Convolutional Neural Network

Great advances in stitching high-quality retinal images have been made in recent years. On the other hand, very few studies have been carried out on low-resolution retinal imaging. This work investigates the challenges of low-resolution retinal images obtained by the D-EYE smartphone-based fundus camera. The proposed method uses homography estimation to register and stitch low-quality retinal images into a cohesive mosaic. First, a Siamese neural network extracts features from a pair of images, after which the correlation of their feature maps is computed. This correlation map is fed through four independent CNNs to estimate the homography parameters, each specializing in different corner coordinates. Our model was trained on a synthetic dataset generated from the Microsoft Common Objects in Context (MSCOCO) dataset; this work added an important data augmentation phase to improve the quality of the model. Then, the same is evaluated on the FIRE retina and D-EYE datasets for performance measurement using the Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM). The obtained results are promising: the average PSNR was 26.14 dB, with an SSIM of 0.96 on the D-EYE dataset. Compared to the method that uses a single neural network for homography calculations, our approach improves the PSNR by 7.96 dB and achieves a 7.86% higher SSIM score.

Design of Broadband Highly Efficient Power Amplifier Based on Low‐Pass Filtering Network With Periodic Structure

ABSTRACTIn this paper, a broadband high‐efficiency power amplifier (PA) is designed by using an easily implementable low‐pass filtering network with periodic structure as the output matching circuit to achieve bandwidth expansion of the PA. The proposed periodic structure is based on right‐handed T‐type basic cells, which can achieve wideband filtering responses. In addition, the load design space of the extended continuous Class‐F mode is combined as the target impedance to achieve an efficiency improvement in bandwidth PAs. For verification, a broadband highly efficient PA across 0.6–3.6 GHz with a relative bandwidth of 143% is designed, fabricated, and measured. The test results have shown a drain efficiency of 60.8%–70.4% and gain of 8.5–11.9 dB with a saturation output power under 3‐dB gain compression of 38.5–41.9 dBm.

Enhancing low‐light images with lightweight fused fixed‐directional filters network

AbstractDeep learning has made significant progress in the field of low‐light image enhancement. However, challenges remain, such as the substantial parameter consumption required for effective image enhancement. Inspired by multi‐scale geometric transformations in image detail enhancement, a novel model called the fixed‐directional filters network is proposed. Fixed‐directional filters network takes the original image as input and employs multiple branches for parallel processing. One branch uses conventional convolutional layers to extract features from the original image, while the other branches apply non‐linear mapping layers based on wavelet transforms. These wavelet transform branches capture the multi‐scale information of the image by combining different directions and convolutional kernels and utilize a trainable custom gamma mapping layer for non‐linear modulation to enhance specific regions of the image. The feature maps processed by each branch are merged through concatenation operations and then passed through convolutional layers to output the enhanced image. Using trainable mapping functions alone to enhance details significantly reduces the reliance on convolutional layers, effectively lowering the model's parameter count to only 13k parameters. Additionally, experiments demonstrate that fixed‐directional filters network significantly improves image quality, particularly in capturing image details and enhancing image contrast.

A visual measurement method for grinding surface roughness combining filter and branch convolution network

ABSTRACT The visual measurement method of grinding surface roughness is mainly predicted by designing image feature indicators. In contrast, the self-extraction method of grinding surface features based on deep learning has problems such as noise, low resolution and poor perception of details. This paper proposes a visual measurement method for grinding surface roughness to address these problems by combining filters and branching convolutional networks. The method adopts a two-branch convolutional neural network, inputs the images using filter denoising and the original grinding workpiece surface images simultaneously, self-extracts roughness correlation features and fuses them to achieve feature enhancement, effectively improves the recognition accuracy and generalisation ability of the model, and verifies the robustness of the model in the case of rusting on the workpiece surface. The experimental results show that the method can effectively solve the problem of the grinding process’s weak and difficult recognition of surface roughness feature information and provide a basis for automated visual online measurement of grinding surface roughness.

A hybrid model for state of charge estimation of lithium-ion batteries utilizing improved adaptive extended Kalman filter and long short-term memory neural network

With lithium-ion batteries being utilized in all aspects of life, accurately estimating the state of charge (SOC) of a battery has become a key issue in battery management systems. In this paper, an improved hybrid model based on adaptive extended Kalman filter (AEKF) and improved long short-term memory (ILSTM) neural network is proposed. The proposed model is based on a second-order RC equivalent circuit model, and the dynamic forgetting factor recursive least squares (DFFRLS) and AEKF algorithms are utilized to obtain the initial SOC estimates. And the estimation error in the AEKF algorithm due to neglecting the higher order terms of the Taylor expansion equations is compensated by the improvement of the LSTM network. The results under different working conditions indicate that the SOC estimation of the hybrid model has good convergence and high system robustness. The maximum error (MAX) of this algorithm is less than 2.3 %, especially the root mean square error (RMSE) and mean absolute error (MAE) are less than 0.84 % and 0.65 %, respectively.

Load prediction of bucket elevator for jigging sorting based on binocular vision and Global Filter Networks

ABSTRACT In jigging sorting, products are elevated by bucket elevators and transported via a shared belt. The product weight of each jig cannot be detected. The output of each product is the key parameter to calculate the jigging separation efficiency and measure the separation effect. At the same time, the bucket elevator overload will cause accidents, resulting in deformation and rupture of the bucket. To solve this problem, an intelligent monitoring scheme of bucket elevators based on binocular vision and deep learning was proposed. The YOLACT algorithm was utilized for coal and gangue image segmentation. Aiming at the problem of multi-layer accumulation of materials in the bucket elevator, a Global Filter Network (GFNet) based on the Pyramid Stereo Matching Network was used for more accurate stereo matching, and statistical filtering was used to denoise the generated 3D point cloud. The volume prediction model was established by integral. The Bootstrap method was used to obtain the best empirical density of coal and gangue. Finally, the volume parameters and density parameters were multiplied to get the mass of coal and gangue. The experimental results showed that the GFNet algorithm can obtain clear 3D point clouds. The average error of the volume calculation model based on these 3D point clouds was 4.19%. In the single-machine test of the coal preparation plant, the proposed system realized the real-time detection of the mass of materials carried by the bucket elevator. Compared with the electronic belt scale, the average error was 10.55%, which meets the industrial demand.

Life prediction of lithium battery based on particle filter and BP neural network

Abstract This paper focuses on using Particle Filter (PF) and Back Propagation (BP) neural networks for RUL prediction. This fusion strategy uses particle filtering (PF) for battery system state estimation and prediction while considering various types of noise and uncertainties to assess the system state comprehensively. Furthermore, the strategy employs a BP neural network to learn and combine historical data patterns with particle filtering estimation, improving prediction accuracy and reliability. This study validates the proposed fusion method by comparing it with the original PF and BP prediction methods using Toyota Motor Corporation and Stanford University datasets. The experimental results demonstrate its superior performance and higher prediction accuracy.

GFN: A novel joint entity and relation extraction model with redundancy and denoising strategies

Joint entity and relation extraction refers to the extraction of entities and their corresponding relationships in the given sentence, which has gained increasing attention in recent years. Some joint extraction models utilize a shared encoder to model the interactions between named entity recognition and relation extraction subtasks. Despite achieving decent performance, they inevitably face the issue of error propagation. One-step exhaustive methods can mitigate the error propagation problem to some extent, but they suffer from issues such as huge computation complexity and a proliferation of negative samples. Therefore, addressing the problems mentioned above, we propose a Greedy Filter Network that combines Greedy-NER and Filter-RE. GFN employs the Greedy-NER with a redundancy strategy to prioritize recall, thereby reducing error propagation between subtasks. To reduce the computational complexity, we design an innovative approach to represent and store spans in Greedy-NER. In Filter-RE, we traverse all pairwise combinations of candidate entities. To address the issue of widespread negative samples, we design a denoising strategy with two filters, effectively filtering out entity pairs without relations, which can eliminate noise and alleviate the issue of negative sample proliferation. Finally, to enable flexible control over the redundancy strategy, we design two misclassifying penalty parameters for each module. The experimental results indicate that GFN achieves the state-of-the-art F1-score on the CoNLL04 and NYT datasets, with a notable 2.0% improvement observed specifically on CoNLL04.

Adaptive CAPTCHA: A CRNN-Based Text CAPTCHA Solver with Adaptive Fusion Filter Networks

Text-based CAPTCHAs remain the most widely adopted security scheme, which is the first barrier to securing websites. Deep learning methods, especially Convolutional Neural Networks (CNNs), are the mainstream approach for text CAPTCHA recognition and are widely used in CAPTCHA vulnerability assessment and data collection. However, verification code recognizers are mostly deployed on the CPU platform as part of a web crawler and security assessment; they are required to have both low complexity and high recognition accuracy. Due to the specifically designed anti-attack mechanisms like noise, interference, geometric deformation, twisting, rotation, and character adhesion in text CAPTCHAs, some characters are difficult to efficiently identify with high accuracy in these complex CAPTCHA images. This paper proposed a recognition model named Adaptive CAPTCHA with a CNN combined with an RNN (CRNN) module and trainable Adaptive Fusion Filtering Networks (AFFN), which effectively handle the interference and learn the correlation between characters in CAPTCHAs to enhance recognition accuracy. Experimental results on two datasets of different complexities show that, compared with the baseline model Deep CAPTCHA, the number of parameters of our proposed model is reduced by about 70%, and the recognition accuracy is improved by more than 10 percentage points in the two datasets. In addition, the proposed model has a faster training convergence speed. Compared with several of the latest models, the model proposed by the study also has better comprehensive performance.

A GIS-based approach to determining optimal location for decentralized inner city smart filters: Toward net zero cities

Climate change has already begun to take visible effect globally in recent years. Given the climate change paradox and urbanization trends, cities’ success would not only depend on smartness and sustainability, but also resilience to all forthcoming economic, environmental, or behavioral changes. Numerous technologies have surfaced and proved effective in CO2 removal from the local environment. However, the optimal placement of these smart filters is a complex task and require logical and strategic decision-making. Determining the optimal location is one of the key factors for establishing a network of smart air filters. This study used a GIS-based suitability analysis for identifying optimal locations for smart filters based on pollution hotspots (population and spatial proximity to industry, commercial centers, roads, high-traffic areas, and intersections). The spatial analysis involves the determination and preparation of input layers, ranking layers, assigning weights to each criterion, and generation of a suitability map. The sites with a higher suitability score (7 or above) are optimum sites for air filters. The sites are spatially distributed over different regions. The findings revealed that GIS-based suitability analysis can be an effective technique for placing smart filters within an urban environment. These findings can help decision-makers to prioritize the location considering environmental constraints. The proposed solution aims to pave the way for fostering resilient, smart, and sustainable cities through a community sensing platform targeting hotspots within spatial variations.

HorSR: High-order spatial interactions and residual global filter for efficient image super-resolution

Recent advances in efficient image super-resolution (EISR) include convolutional neural networks, which exploit distillation and aggregation strategies with copious channel split and concatenation operations to fully exploit limited hierarchical features. In contrast, the Transformer network presents a challenge for EISR because multiheaded self-attention is a computationally demanding process. To respond to this challenge, this paper proposes replacing multiheaded self-attention in the Transformer network with global filtering and recursive gated convolution. This strategy allows us to design a high-order spatial interaction and residual global filter network for efficient image super-resolution (HorSR), which comprises three components: a shallow feature extraction module, a deep feature extraction module, and a high-quality image-reconstruction module. In particular, the deep feature extraction module comprises residual global filtering and recursive gated convolution blocks. The experimental results show that the HorSR network provides state-of-the-art performance with the lowest FLOPs of existing EISR methods.

Output-mask-based adaptive NN control for stochastic time-delayed multi-agent systems with a unified event-triggered approach

This paper investigates a class of output-mask-based adaptive neural network (NN) tracking control for nonlinear stochastic time-delayed multi-agent systems (STMASs) based on a unified event-triggered approach. The output signal relies on an output mapping acted as a mask, defined as a privacy-protection-like method, so that the internal state of one agent cannot be identified by other distrustful eavesdroppers or attackers. Moreover, the construction of a unified event-triggered control scheme retains the advantages of the saturation threshold triggering strategy, incorporates distributed errors, and increases the flexibility of thresholds. Furthermore, for stochastic time-delay multi-agent systems, the initial value limitation of the conventional first-order filter is removed by a first-order Levant differentiator, and a new estimation term in the fuzzy observer is established to solve the nonlinear fault. The unknown function in pure-feedback form is addressed via combining Butterworth low-pass filter and radial basis function neural networks (RBF NNs). Finally, the boundedness of all signals in the closed-loop systems is demonstrated, and the effectiveness of the proposed algorithm is verified by some simulation results.

The syntheses design of an S-band 32-channel broadband filter and monitoring network

The syntheses design of an S-band 32-channel broadband filter and monitoring network