arXiv subjects

Wuliang Yin

Publications and source records attributed to Wuliang Yin.

12 recordsLinked to original sources

arXiv2024

Enhanced Structured State Space Models via Grouped FIR Filtering and Attention Sink Mechanisms

Structured State Space Models (SSMs) have emerged as compelling alternatives to Transformer architectures, offering linear-time complexity and superior performance in various sequence modeling tasks. Despite their advantages, SSMs like the original Mamba-2 face training difficulties due to the sensitivities introduced by the extended series of recurrent matrix multiplications. In this paper, we propose an advanced architecture that mitigates these challenges by decomposing A-multiplications into multiple groups and optimizing positional encoding through Grouped Finite Impulse Response (FIR) filtering. This new structure, denoted as Grouped FIR-enhanced SSM (GFSSM), employs semiseparable matrices for efficient computation. Furthermore, inspired by the "attention sink" phenomenon identified in streaming language models, we incorporate a similar mechanism to enhance the stability and performance of our model over extended sequences. Our approach further bridges the gap between SSMs and Transformer architectures, offering a viable path forward for scalable and high-performing sequence modeling.

cs.CL↗

arXiv2024

Few-Shot Image Classification and Segmentation as Visual Question Answering Using Vision-Language Models

The task of few-shot image classification and segmentation (FS-CS) involves classifying and segmenting target objects in a query image, given only a few examples of the target classes. We introduce the Vision-Instructed Segmentation and Evaluation (VISE) method that transforms the FS-CS problem into the Visual Question Answering (VQA) problem, utilising Vision-Language Models (VLMs), and addresses it in a training-free manner. By enabling a VLM to interact with off-the-shelf vision models as tools, the proposed method is capable of classifying and segmenting target objects using only image-level labels. Specifically, chain-of-thought prompting and in-context learning guide the VLM to answer multiple-choice questions like a human; vision models such as YOLO and Segment Anything Model (SAM) assist the VLM in completing the task. The modular framework of the proposed method makes it easily extendable. Our approach achieves state-of-the-art performance on the Pascal-5i and COCO-20i datasets.

cs.CV↗

arXiv2023

Few-Shot Classification & Segmentation Using Large Language Models Agent

The task of few-shot image classification and segmentation (FS-CS) requires the classification and segmentation of target objects in a query image, given only a few examples of the target classes. We introduce a method that utilises large language models (LLM) as an agent to address the FS-CS problem in a training-free manner. By making the LLM the task planner and off-the-shelf vision models the tools, the proposed method is capable of classifying and segmenting target objects using only image-level labels. Specifically, chain-of-thought prompting and in-context learning guide the LLM to observe support images like human; vision models such as Segment Anything Model (SAM) and GPT-4Vision assist LLM understand spatial and semantic information at the same time. Ultimately, the LLM uses its summarizing and reasoning capabilities to classify and segment the query image. The proposed method's modular framework makes it easily extendable. Our approach achieves state-of-the-art performance on the Pascal-5i dataset.

cs.CV↗

arXiv2021

Depth Evaluation for Metal Surface Defects by Eddy Current Testing using Deep Residual Convolutional Neural Networks

Eddy current testing (ECT) is an effective technique in the evaluation of the depth of metal surface defects. However, in practice, the evaluation primarily relies on the experience of an operator and is often carried out by manual inspection. In this paper, we address the challenges of automatic depth evaluation of metal surface defects by virtual of state-of-the-art deep learning (DL) techniques. The main contributions are three-fold. Firstly, a highly-integrated portable ECT device is developed, which takes advantage of an advanced field programmable gate array (Zynq-7020 system on chip) and provides fast data acquisition and in-phase/quadrature demodulation. Secondly, a dataset, termed as MDDECT, is constructed using the ECT device by human operators and made openly available. It contains 48,000 scans from 18 defects of different depths and lift-offs. Thirdly, the depth evaluation problem is formulated as a time series classification problem, and various state-of-the-art 1-d residual convolutional neural networks are trained and evaluated on the MDDECT dataset. A 38-layer 1-d ResNeXt achieves an accuracy of 93.58% in discriminating the surface defects in a stainless steel sheet. The depths of the defects vary from 0.3 mm to 2.0 mm in a resolution of 0.1 mm. In addition, results show that the trained ResNeXt1D-38 model is immune to lift-off signals.

cs.LG↗

arXiv2020

A Class D Power Amplifier for Multi-Frequency Eddy Current Testing Based on Multi-Simultaneous-Frequency Selective Harmonic Elimination Pulse Width Modulation

Efficiency and multisimultaneous-frequency (MSF) output capability are two major criteria characterizing the performance of a power amplifier in the application of multifrequency eddy current testing (MECT). Switch-mode power amplifiers are known to have a very high efficiency, yet they have rarely been adopted in the instrumental development of MECT. In addition, switch-mode power amplifiers themselves are lacking in the research literature for MSF capability. In this article, a Class D power amplifier is designed so as to address the two issues. An MSF selective harmonic elimination pulsewidth modulation method is proposed to generate alternating magnetic fields, which are rich in selected harmonics. A field-programmable-gate-array-based experimental system has been developed to verify the design. Results show that the proposed methodology is capable of generating high MSF currents in the transmitting coil with a low distortion of signal.

eess.SY↗

arXiv2019

Custom Edge-Element FEM Solver and its Application to Eddy-Current Simulation of Realistic 2M-Element Human Brain Phantom

Extensive research papers of three-dimensional computational techniques are widely used for the investigation of human brain pathophysiology. Eddy current analyzing could provide an indication of conductivity change within a biological body. A significant obstacle to current trend analyses is the development of a numerically stable and efficiency-finite element scheme that performs well at low frequency and does not require a large number of degrees of freedom. Here, a custom finite element method (FEM) solver based on edge elements is proposed using the weakly coupled theory, which separates the solution into two steps. First, the background field (the magnetic vector potential on each edge) is calculated and stored. Then, the electric scalar potential on each node is obtained by FEM based on Galerkin formulations. Consequently, the electric field and eddy current distribution in the object can be obtained. This solver is more efficient than typical commercial solvers since it reduces the vector eddy current equation to a scalar one, and reduces the meshing domain to just the eddy current region. It can therefore tackle complex eddy current calculations for models with much larger numbers of elements, such as those encountered in eddy current computation in biological tissues. An example is presented with a realistic human brain mesh of 2 million elements. In addition, with this solver, the equivalent magnetic field induced from the excitation coil is applied, and therefore there is no need to mesh the excitation coil. In combination, these significantly increase the efficiency of the solver.

physics.med-ph↗

arXiv2019

An equivalent-effect phenomenon in eddy current non-destructive testing of thin structures

The inductance/impedance due to thin metallic structures in non-destructive testing (NDT) is difficult to evaluate. In particular, in Finite Element Method (FEM) eddy current simulation, an extremely fine mesh is required to accurately simulate skin effects especially at high frequencies, and this could cause an extremely large total mesh for the whole problem, i.e. including, for example, other surrounding structures and excitation sources like coils. Consequently, intensive computation requirements are needed. In this paper, an equivalent-effect phenomenon is found, which has revealed that alternative structures can produce the same effect on the sensor response, i.e. mutual impedance/inductance of coupled coils if a relationship (reciprocal relationship) between the electrical conductivity and the thickness of the structure is observed. By using this relationship, the mutual inductance/impedance can be calculated from the equivalent structures with much fewer mesh elements, which can significantly save the computation time. In eddy current NDT, coils inductance/impedance is normally used as a critical parameter for various industrial applications, such as flaw detection, coating and microstructure sensing. Theoretical derivation, measurements and simulations have been presented to verify the feasibility of the proposed phenomenon.

eess.SP↗

arXiv2019

A Novel Dual Modality Sensor With Sensitivities to Permittivity, Conductivity, and Permeability

In this paper, an electromagnetic sensor which can operate simultaneously in capacitive and inductive modalities with sensitivities to permittivity, conductivity, and permeability is developed, and a novel measurement strategy is proposed accordingly. The sensor is composed of two planar spiral coils with a track width of 4 mm, which promotes its capacitive mode. The capacitive coupling is measured in common mode, while the inductive coupling is measured in differential mode. In capacitive mode, the sensor is sensitive to changes in permittivity, i.e., the dielectric material distribution; while in inductive mode, it is sensitive to magnetically permeable material and electrically conductive material. Furthermore, it is demonstrated that the sensor can simultaneously measure dielectric and conductive materials. This novel sensing element has been designed and implemented. Experimental results verified its effectiveness in dual modality measurement.

eess.SP↗

arXiv2019

Measurement of permeability for ferrous metallic plates using a novel lift-off compensation technique on phase signature

Lift-off of sensor affects the prediction of electromagnetic properties for both ferrous and non-ferrous steel plates. In this paper, we developed a strategy to address this issue for ferrous plates. With increased lift-off, the phase of the measured impedance for steel plates reduces. Meanwhile, the magnitude of the impedance signal decreases. Based on these facts, a phase compensation algorithm is developed which corrects the phase change due to lift-off considering the magnitude of the impedance signal. Further, a new magnetic permeability prediction technique is presented, which has been validated by analytical and measured results. With this new technique, the error in permeability prediction is less than 2% within the range of lift-offs tested.

eess.SP↗

arXiv2019

Determination of the magnetic permeability, electrical conductivity, and thickness of ferrite metallic plates using a multi-frequency electromagnetic sensing system

In this paper, an inverse method was developed which can, in principle, reconstruct arbitrary permeability, conductivity, thickness, and lift-off with a multi-frequency electromagnetic sensor from inductance spectroscopic measurements. The system fives the error of "Bad character(s) in field Abstract" for no reason. Please refer to manuscript for the full abstract.

eess.SP↗

arXiv2019

A Novel Compensation Algorithm for Thickness Measurement Immune to Lift-Off Variations Using Eddy Current Method

Lift-off variation causes errors in the eddy current thickness measurements of metallic plates. In this paper, we have developed an algorithm that can compensate for this variation and produce an index that is linked to the thickness, but is virtually independent of lift-off. This index, termed as the compensated peak frequency, can be obtained from the measured multifrequency inductance spectral data using the algorithm we developed in this paper. This method has been derived through mathematical manipulation and verified by both the simulation and experimental data. Accuracy in the thickness measurements at different lift-offs proved to be within 2%.

eess.SP↗

arXiv2019

Reducing the Lift-Off Effect on Permeability Measurement for Magnetic Plates From Multifrequency Induction Data

Liftoff variation causes errors in eddy current measurement of nonmagnetic plates as well as magnetic plates. For nonmagnetic plates, previous work has been carried out to address the issue. In this paper, we follow a similar strategy, but try to reduce the lift-off effect on another index, zerocrossing frequency for magnetic plates. This modified index, termed as the compensated zero-crossing frequency, can be obtained from the measured multifrequency inductance spectral data using the algorithm we developed in this paper. Since the zero-crossing frequency can be compensated, the permeability of magnetic plates can finally be predicted by deriving the relation between the permeability and zero-crossing frequency from Dodd and Deeds method. We have derived the method through mathematical manipulation and verified it by both simulation and experimental data. The permeability error caused by liftoff can be reduced within 7.5%.

eess.SP↗