Refinement Algorithm Research Articles

Weakly supervised power line detection algorithm using a recursive noisy label update with refined broken line segments

Detection of power lines in aerial images is an important problem to prevent accidents of unmanned aerial vehicles operating at low altitudes in the electrical industry. Recently, pixel-level power line detection using deep learning has been studied but production of the pixel-level annotations for massive dataset is difficult. In this study, we propose a power line detection algorithm using weakly supervised learning method to reduce the labeling cost for dataset generation. The algorithm is divided into two stages. First, an approximately localized mask was generated based on a convolutional neural network which was trained with only patch-level labels. Second, recursive training of segmentation network with refined broken line segments was executed. A refinement algorithm, line segment connecting (LSC) is a power-line-specialized refinement module that connects broken lines by approximating the segments as partially straight. In proposed algorithm, predicted image at each recursive step was updated as a label of the next training and the label was developed by itself with LSC. The comprehensive experimental results of our algorithm showed state-of-art F1-score of 94.3% in weakly supervised learning approaches on public dataset. This result suggests that the proposed algorithm is useful for low labeling cost with high performance in line detection application.

Adaptive Mesh Refinement Simulations of Gas Dynamic Flows on Hybrid Meshes

An adaptive mesh refinement algorithm for simulations of the Navier–Stokes equations on unstructured hybrid meshes is presented. Numerical results for compressible supersonic flow around a sphere are given.

Efficient analysis-suitable T-spline fitting for freeform surface reconstruction and intelligent sampling

Optical scanning instruments require sampling and reconstruction with high accuracy and low computational cost. T-splines have recently been developed that allow significant reductions in the number of control parameters by overcoming some of the topological constraints of B-splines and NURBS. As a subset of T-splines, analysis-suitable T-splines (ASTS) show promise due to the linear independence and partition of unity of their basis functions. In this paper, a computationally efficient ASTS fitting algorithm for freeform surface reconstruction is proposed. This algorithm starts with adaptive construction of an initial analysis-suitable T-mesh according to the distribution of high-curvature feature points. A local refinement and local optimisation algorithm of the analysis-suitable T-mesh is then iteratively performed until a preset accuracy condition is satisfied. Our experimental results show that the proposed ASTS fitting can produce over 50% root-mean-square reconstruction error reduction compared to NURBS fitting, with the same number of control parameters. The computing efficiency of the proposed algorithm is equivalent to or higher than that for simple T-spline fitting. Fast derivative analysis of the ASTS has also been carried out, where two automatic intelligent sampling design methods have been developed, namely element area sampling and element curvature sampling. Up to 50% reconstruction error reduction is observed when compared to uniform and statistically optimised sampling designs. With the novel reconstruction and compatible intelligent sampling design techniques, freeform surface measurement accuracy and efficiency could be effectively improved using coordinate measuring machines.

A framework for efficient isogeometric computations of phase-field brittle fracture in multipatch shell structures

We present a computational framework for applying the phase-field approach to brittle fracture efficiently to complex shell structures. The momentum and phase-field equations are solved in a staggered scheme using isogeometric Kirchhoff–Love shell analysis for the structural part and isogeometric second- and fourth-order phase-field formulations for the brittle fracture part. For the application to complex multipatch structures, we propose penalty formulations for imposing all the required interface constraints, i.e., displacement (C0) and rotational (C1) continuity for the structure as well as C0 and C1 continuity for the phase field, where the latter is required only in the case of the fourth-order phase-field model. All involved penalty terms are scaled with the corresponding problem parameters to ensure a consistent scaling of the penalty contributions to the global system of equations. As a consequence, all coupling terms are controlled by one global penalty parameter, which can be set to 103 independent of the problem parameters. Furthermore, we present a multistep predictor–corrector algorithm for adaptive local refinement with LR NURBS, which can accurately predict and refine the region around the crack even in cases where fracture fully develops in a single load step, such that rather coarse initial meshes can be used, which is essential especially for the application to large structures. Finally, we investigate and compare the numerical efficiency of loosely vs. strongly staggered solution schemes and of the second- vs. fourth-order phase-field models.

Adjoint-Based Accelerated Adaptive Refinement in Frequency Domain 3-D Finite Element Method Scattering Problems

We present the application of adjoint analysis to 3-D finite-element method scattering problems for a posteriori error estimation and adaptive refinement. Adjoint-based methodologies, though underutilized in computational electromagnetics (CEM), enable significant improvements for both efficiency and accuracy. We first formulate the adjoint problem of the 3-D double-curl wave equation and the error estimates for the construction of novel accelerated adaptive refinement algorithms. We demonstrate adaptive error control for a customizable quantity of interest (QoI) resulting in targeted refinement and improved resource allocation through the application of automatic global and local error tolerance heuristics that accelerate the refinement process. The proposed refinement algorithms rapidly refine even extremely coarse initial discretizations to high accuracy, eliminating or substantially reducing manual intervention in the generation of computationally efficient and accurate simulations. Moreover, comparisons with analytical results validate our approach to accelerating automatic refinement to fine tolerances.

Higher-order mesh generation using CISAMR: A case study on bias in presentation and interpretation of results

A modified version of the Conforming to Interface Structured Adaptive Mesh Refinement (CISAMR) algorithm is presented for the construction of higher-order Lagrangian and NURBS-enhanced (NE) finite element meshes. CISAMR non-iteratively transforms a structured grid into a conforming mesh with an upper bound of three on elements aspect ratio. In this work, we introduce new algorithmic aspects for generating higher-order Lagrangian and NE meshes using CISAMR. For each type of element, a comprehensive study is then provided on the performance of first, second, and third-order meshes for solving linear elastic problems with smooth and oscillatory curvilinear edges. Local gradient recovery error and computational cost are used as performance metrics in these examples. Outcomes are then used as a case study to elucidate common sources of bias in presenting and interpreting results. In particular, we show how bias, whether purposeful or not, could lead to misleading conclusions regarding the performance of a method.

An efficient Pseudo-spectral based phase field method for dendritic solidification

In the current study, Fourier Pseudo-spectral based dealiasing scheme Random Phase Shift Method (RPSM) using the RK4 temporal scheme is implemented for simulating some benchmark problems of Phase Field based pure metal solidification. This high-resolution scheme is capable of resolving the thin solid-liquid interface accurately without using Adaptive Mesh Refinement (AMR) algorithm even using just adequate number of grid points. Both the conservative and non-conservative forms of the Phase Field equations are studied. It is found that the conservative RPSM scheme gives better computational performances as indicated by the lower values of average CPU time and different error norms calculated for the Phase Field variable. The RPSM based Phase Field model is applied for simulating dendritic solidification and it is found that dendrite shape and dimensionless tip velocity agree well with the existing literature. The RK4 based RPSM scheme is also compared with a 10th order exponential Fourier smoothing Filter in the evolution of four arm dendritic structure and it is observed that the conservative Phase Field equations with filtering scheme display numerical instabilities at later simulation time, while both forms of the RPSM scheme show stable and accurate solutions. Subsequently, the capability of the RPSM based Phase Field model in simulating microstructure evolution consisting of multiple dendrites with random location and orientation is demonstrated. The present study shows that the RPSM based Phase Field model computes dendritic growth accurately and efficiently.

Three-dimensional direct current resistivity forward modeling based on the hp-adaptive finite element method

In recent years, the adaptive finite element method has been used for direct current resistivity forward modeling to improve the accuracy of numerical solutions. The accuracy of adaptive finite element solutions is mainly affected by two factors: the cell size (h) and the order of the shape function (p). To further improve the accuracy of the adaptive finite element solutions and keep low computational costs, we present a hp-adaptive finite element algorithm, combining h-adaptive and p-adaptive, for three-dimensional direct current forward modeling. The adaptive mesh refinement is guided by a posteriori error estimator and a smoothness estimator of the solution. In the adaptive process, for a cell with a large error, if the smoothness of the solution is low, the cell will be refined; otherwise, the order of the shape function on the cell will be increased by one order. To obtain high-precision finite element solutions for complex models with topographies the unstructured grids are adopted. The octree-based mesh refinement method is used to refine the cells, and the shape functions with arbitrary orders in three-dimensional space are generated by using tensor products of one-dimensional polynomials. A two-layer model is used to verify the correctness of the algorithm. The comparison between the convergence rate of global, h-adaptive, and hp-adaptive refinement algorithms indicates that our algorithm can provide the most accurate solution with the lowest computation costs. It shows exponentially convergence rate. Finally, a topography model with an abnormal body and a complex abnormal model are used to verify the robustness of our algorithm.

Learning TBox With a Cascaded Anchor-Free Network for Vehicle Detection

Vehicle detection, the process of identifying vehicles as axis-aligned bounding boxes in still images, is widely used to estimate the range, time-to-collision, and motion of autonomous vehicles (AVs). Bounding boxes, while convenient, are too coarse to adapt well to vehicle shape and pose variations. In this work, we present TBox (Trapezoid & Box), a novel fine-grained representation useful for both localization and recognition that extends the bounding box by restricting the spatial extent of a vehicle to a set of keypoints and indicating semantically significant local areas using subclasses. In contrast to the previous monolithic models, we propose a cascaded anchor-free architecture to estimate the bounding box and TBox. One subnetwork uses a stacked hourglass network to detect each vehicle as a pair of corners without using anchors. Specifically, it learns corner affinity fields, enabling it to perform robust corner grouping. The other subnetwork estimates a TBox as a set of keypoints. This subnetwork utilizes the bounding box results to avoid ambiguous keypoint associations and reuses existing features to reduce the number of parameters. We also propose a multitask learning strategy for training the cascaded model that implicitly integrates the global context with local details, introducing improvements for both tasks. During testing, a refinement algorithm explicitly uses robust local keypoints to correct possible global box errors, ensuring tight geometric representations for nearby critical vehicles. The experiments show that our method outperforms existing anchor-free detectors for vehicle detection and achieves better performance on the TBox task while using a small model.

OSS-DBS: Open-source simulation platform for deep brain stimulation with a comprehensive automated modeling.

In this study, we propose a new open-source simulation platform that comprises computer-aided design and computer-aided engineering tools for highly automated evaluation of electric field distribution and neural activation during Deep Brain Stimulation (DBS). It will be shown how a Volume Conductor Model (VCM) is constructed and examined using Python-controlled algorithms for generation, discretization and adaptive mesh refinement of the computational domain, as well as for incorporation of heterogeneous and anisotropic properties of the tissue and allocation of neuron models. The utilization of the platform is facilitated by a collection of predefined input setups and quick visualization routines. The accuracy of a VCM, created and optimized by the platform, was estimated by comparison with a commercial software. The results demonstrate no significant deviation between the models in the electric potential distribution. A qualitative estimation of different physics for the VCM shows an agreement with previous computational studies. The proposed computational platform is suitable for an accurate estimation of electric fields during DBS in scientific modeling studies. In future, we intend to acquire SDA and EMA approval. Successful incorporation of open-source software, controlled by in-house developed algorithms, provides a highly automated solution. The platform allows for optimization and uncertainty quantification (UQ) studies, while employment of the open-source software facilitates accessibility and reproducibility of simulations.

Machine Learning Accurately Predicts Next Season NHL Player Injury Before It Occurs: Validation of 10,449 Player-Years from 2007-17

Objectives:With the accumulation of big data surrounding National Hockey League (NHL) and the advent of advanced computational processors, machine learning (ML) is ideally suited to develop a predictive algorithm capable of imbibing historical data to accurately project a future player’s availability to play based on prior injury and performance. To the end of leveraging available analytics to permit data-driven injury prevention strategies and informed decisions for NHL franchises beyond static logistic regression (LR) analysis, the objective of this study of NHL players was to (1) characterize the epidemiology of publicly reported NHL injuries from 2007-17, (2) determine the validity of a machine learning model in predicting next season injury risk for both goalies and non-goalies, and (3) compare the performance of modern ML algorithms versus LR analyses.Methods:Hockey player data was compiled for the years 2007 to 2017 from two publicly reported databases in the absence of an official NHL-approved database. Attributes acquired from each NHL player from each professional year included: age, 85 player metrics, and injury history. A total of 5 ML algorithms were created for both non-goalie and goalie data; Random Forest, K-Nearest Neighbors, Naive Bayes, XGBoost, and Top 3 Ensemble. Logistic regression was also performed for both non-goalie and goalie data. Area under the receiver operating characteristics curve (AUC) primarily determined validation.Results:Player data was generated from 2,109 non-goalies and 213 goalies with an average follow-up of 4.5 years. The results are shown below in Table 1.For models predicting following season injury risk for non-goalies, XGBoost performed the best with an AUC of 0.948, compared to an AUC of 0.937 for logistic regression. For models predicting following season injury risk for goalies, XGBoost had the highest AUC with 0.956, compared to an AUC of 0.947 for LR.Conclusion: Advanced ML models such as XGBoost outperformed LR and demonstrated good to excellent capability of predicting whether a publicly reportable injury is likely to occur the next season. As more player-specific data become available, algorithm refinement may be possible to strengthen predictive insights and allow ML to offer quantitative risk management for franchises, present opportunity for targeted preventative intervention by medical personnel, and replace regression analysis as the new gold standard for predictive modeling.Figure 1.

AIR: Iterative refinement acceleration using arbitrary dynamic precision

The increased degree of concurrent operations by lower precision arithmetic enables high performance for iterative refinement. Most of related work present statically defined mixed precision arithmetic approaches, while adapting a level of arithmetic precision dynamically in a loop with one-bit granularity can further improve the performance. This paper presents Arbitrary Dynamic Precision Iterative Refinement algorithm (AIR) that minimizes the total significand bit-width to solve iterative refinement. AIR detects the number of cancellation bits dynamically per iteration and uses the information to provide the least sufficient significand bit-width for the next iteration. We prove that AIR is a backward stable algorithm and can bring up to 2−3× speedups over a mixed precision iterative refinement depending on the characteristics of hardware. Our software demonstration shows that AIR requires only 83% of the significand bits required by mixed precision iterative refinement that solve linear systems for double precision accuracy for backward error with 32 × 32 standard normally distributed matrices.

Direct numerical simulation of multiscale flow physics of binary droplet collision

Direct numerical simulations are performed to investigate the multiscale flow physics of binary droplet collision over a wide range of Weber numbers and impact factors. All possible collision outcomes, including bouncing (both head-on and off-center), coalescence, reflexive separation, and stretching separation, are considered. The theoretical formulation is based on a complete set of conservation equations for both the liquid and gas phases. An improved volume-of-fluid technique, which is augmented by an adaptive mesh refinement algorithm, is used to track the liquid/gas interface. Several local refinement criteria are validated and employed to improve the computational accuracy and efficiency substantially. In particular, a thickness-based refinement technique is implemented for treating cases involving extremely thin gas films between droplets. The smallest numerical grid is ∼10 nm, which is on the order of 10−5 times the initial droplet diameter. A photorealistic visualization technique is employed to gain direct insights into the detailed collision dynamics, including both the shape evolution and mass relocation. The numerical framework allows us to systematically investigate the underlying mechanisms and processes, such as gas-film drainage and energy and mass transfer, at scales sufficient to resolve the near-field dynamics during droplet collision. The nonmonotonic transition of bouncing and merging outcomes for head-on collision is identified by varying the Weber number over two orders of magnitude. A geometric relation defining the droplet interactions is developed. Analytical models are also established to predict the mass transfer between colliding droplets.

Spacecraft close-range trajectory planning via convex optimization and multi-resolution technique

Discretization of the state and control is indispensable for solving a trajectory planning problem numerically via convex optimization, and uniform-grid discretization is the most common strategy in recent studies. However, when a higher-precision solution is desired, this kind of discretization will greatly increase the scale of optimization variables and then reduce the optimization efficiency. To remedy this weakness, a novel trajectory planning strategy for spacecraft relative motion is proposed in this paper by combining convex optimization and multi-resolution technique (MRT). In this optimization strategy, convex optimization works as an inner-layer algorithm for trajectory optimization, and the MRT works as an outer-layer algorithm for mesh refinement. Moreover, the no-fly zone constraints are considered in the trajectory optimization, and the affine approximations of the spherical and ellipsoidal no-fly zones are derived by the analytical formula of the tangent plane. Numerical simulations demonstrate the effectiveness of the proposed methods. Results show that the combined optimization method can adaptively adjust the local grid density according to the designed resolution level and performs better in computing efficiency than the traditional uniform discretization.

Pixel sampling by clustering

In this paper, we describe Pixel Sampling Clustering Technique (PSCT), a data-driven sampling procedure used to reduce pixel sets. We view the pixels in an image as a high redundancy 3D space. We also refer to this space as our color model. Our method aims to retain a relevant sample of the data so it can act as a new smaller, hence more efficient, color model. PSCT uses a pair of fast density-based clustering algorithms in tandem. First, it applies Birch and then DBSCAN to keep the most densely represented colors. We cluster the resulting color model and use the labels to segment images. We also complement the sampling method with a refinement algorithm intended to improve color representation. In our paper, we show how to reconstruct images using our reduced color model. We also show that reconstructed images have enough information to perform image related learning tasks with almost the same accuracy than the original images but with only a small fraction of the data. We test our sampling method in three image related supervised and unsupervised tasks and compare them with state-of-the-art methods. For our experiments, we use two image datasets: MIT’s Vision Texture Dataset and Berkeley’s BSD500.

Modulator refinement algorithm for coherent modulation imaging

Coherent modulation imaging (CMI) has been shown to be an effective lensless diffraction approach to imaging general extended samples with fast algorithmic convergence and high robustness to data imperfection. In the reported phasing algorithms of CMI, an exact knowledge of modulator is used as a priori. Extra characterization of the modulator is thus required before the CMI experiments are conducted and this can be cumbersome. Here we propose a modulator refinement algorithm that allows for modulator refinement in the same iterative process of image reconstruction. We demonstrate the method for both near-field and far-field geometries in simulations and for a far-field experiment. A relaxed requirement on exactly knowing the modulator would turn CMI into a standalone technique and make it much easier to implement, thus open up its wider applications in biology and materials science.

An Improved Algorithm for Refinement of Overlapping Images of Marine Organisms

Li, K.; Jia, L., and Shi, X., 2020. An improved algorithm for refinement of overlapping images of marine organisms. In: Malvarez, G. and Navas, F. (eds.), Global Coastal Issues of 2020. Journal of Coastal Research, Special Issue No. 95, pp. 67–71. Coconut Creek (Florida), ISSN 0749-0208.Aiming at the shortcomings of the current ocean Ocean image edge detection algorithm, such as poor real-time performance, slow Ocean image processing and limited storage space, an improved ocean Ocean image overlay Ocean image refinement algorithm is designed. Collect overlapping Ocean images of marine organisms and store large amounts of Ocean images in SDRAM memory; use traditional SOBEL operators to detect overlapping portions of ocean Ocean images, use non-maximum suppression to refine Ocean image edges, and clearly and accurately display ocean edges through VGA interface Ocean image. The experimental results show that the method can locate the edge of the overlapping Ocean images of marine organisms more accurately, and the refinement process has better real-time performance, which can be applied to more complex edge detection of overlapping Ocean images of marine organisms.

A locally refined adaptive isogeometric analysis for steady-state heat conduction problems

This paper presents an isogeometric analysis with adaptivity using locally refined B-splines (LR B-splines) for steady-state heat conduction simulations in solids. Within this framework, the LR B-splines, which have an efficient and simple local refinement algorithm, are used to represent the geometry, and are also employed for spatial discretization, thus providing a seamless interaction between the CAD models and the numerical analysis. A Zienkiewicz-Zhu a posteriori error estimator in terms of the temperature gradient recovery is used to identify the regions for local mesh refinement. The accuracy and convergence properties of the proposed framework are demonstrated through several two-dimensional isotropic examples. Numerical results indicate good performance of the present method as the adaptive refinement technique yields more accurate solution compared to the uniform global refinement with an improved convergence rate.

Is Deep Learning On Par with Human Observers for Detection of Radiographically Visible and Occult Fractures of the Scaphoid?

Preliminary experience suggests that deep learning algorithms are nearly as good as humans in detecting common, displaced, and relatively obvious fractures (such as, distal radius or hip fractures). However, it is not known whether this also is true for subtle or relatively nondisplaced fractures that are often difficult to see on radiographs, such as scaphoid fractures. (1) What is the diagnostic accuracy, sensitivity, and specificity of a deep learning algorithm in detecting radiographically visible and occult scaphoid fractures using four radiographic imaging views? (2) Does adding patient demographic (age and sex) information improve the diagnostic performance of the deep learning algorithm? (3) Are orthopaedic surgeons better at diagnostic accuracy, sensitivity, and specificity compared with deep learning? (4) What is the interobserver reliability among five human observers and between human consensus and deep learning algorithm? We retrospectively searched the picture archiving and communication system (PACS) to identify 300 patients with a radiographic scaphoid series, until we had 150 fractures (127 visible on radiographs and 23 only visible on MRI) and 150 non-fractures with a corresponding CT or MRI as the reference standard for fracture diagnosis. At our institution, MRIs are usually ordered for patients with scaphoid tenderness and normal radiographs, and a CT with radiographically visible scaphoid fracture. We used a deep learning algorithm (a convolutional neural network [CNN]) for automated fracture detection on radiographs. Deep learning, an advanced subset of artificial intelligence, combines artificial neuronal layers to resemble a neuron cell. CNNs-essentially deep learning algorithms resembling interconnected neurons in the human brain-are most commonly used for image analysis. Area under the receiver operating characteristic curve (AUC) was used to evaluate the algorithm's diagnostic performance. An AUC of 1.0 would indicate perfect prediction, whereas 0.5 would indicate that a prediction is no better than a flip of a coin. The probability of a scaphoid fracture generated by the CNN, sex, and age were included in a multivariable logistic regression to determine whether this would improve the algorithm's diagnostic performance. Diagnostic performance characteristics (accuracy, sensitivity, and specificity) and reliability (kappa statistic) were calculated for the CNN and for the five orthopaedic surgeon observers in our study. The algorithm had an AUC of 0.77 (95% CI 0.66 to 0.85), 72% accuracy (95% CI 60% to 84%), 84% sensitivity (95% CI 0.74 to 0.94), and 60% specificity (95% CI 0.46 to 0.74). Adding age and sex did not improve diagnostic performance (AUC 0.81 [95% CI 0.73 to 0.89]). Orthopaedic surgeons had better specificity (0.93 [95% CI 0.93 to 0.99]; p < 0.01), while accuracy (84% [95% CI 81% to 88%]) and sensitivity (0.76 [95% CI 0.70 to 0.82]; p = 0.29) did not differ between the algorithm and human observers. Although the CNN was less specific in diagnosing relatively obvious fractures, it detected five of six occult scaphoid fractures that were missed by all human observers. The interobserver reliability among the five surgeons was substantial (Fleiss' kappa = 0.74 [95% CI 0.66 to 0.83]), but the reliability between the algorithm and human observers was only fair (Cohen's kappa = 0.34 [95% CI 0.17 to 0.50]). Initial experience with our deep learning algorithm suggests that it has trouble identifying scaphoid fractures that are obvious to human observers. Thirteen false positive suggestions were made by the CNN, which were correctly detected by the five surgeons. Research with larger datasets-preferably also including information from physical examination-or further algorithm refinement is merited. Level III, diagnostic study.

Adaptive procedures for meshfree RBF unsymmetric and symmetric collocation methods

Abstract In this article we present an adaptive scheme for solving radial basis function collocation problems, which involve elliptic partial differential equations. The proposed algorithm is applied to two usual numerical methods, known as the nonsymmetric Kansa’s method and the symmetric Hermite-based approach. Basically, the refinement algorithm is firstly characterized by the use of an adaptive superimposition scheme, in which an error estimate compares two approximate solutions computed on a coarser and a finer set of collocation points, and then on a modified adaptive residual subsampling scheme. Blending these computational techniques we detect the areas that need to be refined, also having the chance to further add and/or remove adaptively collocation points. Our study is supported by several numerical results, which illustrate the performance of our iterative algorithm.