Direction Of Local Gradient Research Articles

A Client Selection Method Based on Loss Function Optimization for Federated Learning

Federated learning is a distributed machine learning method that can solve the increasingly serious problem of data islands and user data privacy, as it allows training data to be kept locally and not shared with other users. It trains a global model by aggregating locally-computed models of clients rather than their raw data. However, the divergence of local models caused by data heterogeneity of different clients may lead to slow convergence of the global model. For this problem, we focus on the client selection with federated learning, which can affect the convergence performance of the global model with the selected local models. We propose FedChoice, a client selection method based on loss function optimization, to select appropriate local models to improve the convergence of the global model. It firstly sets selected probability for clients with the value of loss function, and the client with high loss will be set higher selected probability, which can make them more likely to participate in training. Then, it introduces a local control vector and a global control vector to predict the local gradient direction and global gradient direction, respectively, and calculates the gradient correction vector to correct the gradient direction to reduce the cumulative deviation of the local gradient caused by the Non-IID data. We make experiments to verify the validity of FedChoice on CIFAR-10, CINIC-10, MNIST, EMNITS, and FEMNIST datasets, and the results show that the convergence of FedChoice is significantly improved, compared with FedAvg, FedProx, and FedNova.

Facial Expression Recognition Using Convolution Neural Network Fusion and Texture Descriptors Representation

Facial expression recognition is an interesting research direction of pattern recognition and computer vision. It has been increasingly used in artificial intelligence, human–computer interaction and security monitoring. In recent years, Convolution Neural Network (CNN) as a deep learning technique and multiple classifier combination method has been applied to gain accurate results in classifying face expressions. In this paper, we propose a multimodal classification approach based on a local texture descriptor representation and a combination of CNN to recognize facial expression. Initially, in order to reduce the influence of redundant information, the preprocessing stage is performed using face detection, face image cropping and texture descriptors of Local Binary Pattern (LBP), Local Gradient Code (LGC), Local Directional Pattern (LDP) and Gradient Direction Pattern (GDP) calculation. Second, we construct a cascade CNN architecture using the multimodal data of each descriptor (CNNLBP, CNNLGC, CNNGDP and CNNLDP) to extract facial features and classify emotions. Finally, we apply aggregation techniques (sum and product rule) for each modality to combine the four multimodal outputs and thus obtain the final decision of our system. Experimental results using CK[Formula: see text] and JAFFE database show that the proposed multimodal classification system achieves superior recognition performance compared to the existing studies with classification accuracy of 97, 93% and 94, 45%, respectively.

Computer simulation of microstructure development in powder-bed additive manufacturing with crystallographic texture

Parts produced via laser powder-bed fusion additive manufacturing exhibit complex microstructures that depend on processing variables and often vary widely in crystallographic texture and grain morphology. The need to understand, predict, and control these microstructural variations motivates the development of modeling tools capable of accurately predicting LPBF microstructures. Monte Carlo (MC) Potts models have been employed to successfully model the formation of grain structures in additively manufactured parts but have lacked the ability to simulate crystallographic texture. We present an extension of the MC Potts model that assigns an orientation to each grain and penalizes growth of solid into the fusion zone based on proximity of the nearest 〈100〉 crystal direction to the local temperature gradient direction. This allows for crystallographically selective growth to drive texture formation during the development of the solidification microstructure in each melt track. LPBF builds of alloy 718 with a unidirectional scan pattern provided microstructures with substantial variations in grain size, grain morphology, and texture. These distinctive albeit atypical microstructures were used to validate the simulation method, i.e. good agreement was obtained between the simulated and experimental grain shapes and textures.

VESPER: global and local cryo-EM map alignment using local density vectors

An increasing number of density maps of biological macromolecules have been determined by cryo-electron microscopy (cryo-EM) and stored in the public database, EMDB. To interpret the structural information contained in EM density maps, alignment of maps is an essential step for structure modeling, comparison of maps, and for database search. Here, we developed VESPER, which captures the similarity of underlying molecular structures embedded in density maps by taking local gradient directions into consideration. Compared to existing methods, VESPER achieved substantially more accurate global and local alignment of maps as well as database retrieval.

Keypoint-based Deep Learning Approach for Building Footprint Extraction Using Aerial Images

Building footprint extraction is an active topic in the domain of remote sensing, since buildings are a fundamental unit of urban areas. Deep convolutional neural networks successfully perform footprint extraction from optical satellite images. However, semantic segmentation produces coarse results in the output, such as blurred and rounded boundaries, which are caused by the use of convolutional layers with large receptive fields and pooling layers. The objective of this study is to generate visually enhanced building objects by directly extracting the vertices of individual buildings by combining instance segmentation and keypoint detection. The target keypoints in building extraction are defined as points of interest based on the local image gradient direction, that is, the vertices of a building polygon. The proposed framework follows a two-stage, top-down approach that is divided into object detection and keypoint estimation. Keypoints between instances are distinguished by merging the rough segmentation masks and the local features of regions of interest. A building polygon is created by grouping the predicted keypoints through a simple geometric method. Our model achieved an F1-score of 0.650 with an mIoU of 62.6 for building footprint extraction using the OpenCitesAI dataset. The results demonstrated that the proposed framework using keypoint estimation exhibited better segmentation performance when compared with Mask R-CNN in terms of both qualitative and quantitative results.

Agent-based modelling of juvenile eel migration via selective tidal stream transport

Recruitment of temperate eel species Anguilla anguilla, A. rostrata &A. japonica has declined over the last few decades due to human activities, such as overfishing and construction of migratory barriers (e.g. dams, weirs and sluices) and hazardous energy infrastructure (e.g. turbines, intakes and outfalls). Numerical models, substantiated with data from field and laboratory studies, can potentially predict and quantify the relative impacts of such activities, thereby assisting in the sustainable management of eel populations. Here, we present an agent-based model (ABM) of juvenile eel migration up estuaries. The model includes relevant eel behaviours and environmental conditions that, according to the literature, influence upstream migration. Crucially, by assessing the local salinity gradient and relative flow direction, the modelled eels (agents) self-determine whether the tide is flooding or ebbing and orientate themselves for navigation, with no top-down instructions. This allows the agents to decide which particular behaviour to undertake as part of Selective Tidal Stream Transport (STST). The developed ABM is coupled to a hydrodynamic model of the Thames Estuary and the results substantiated by comparison against eel trap data. Combinations of the various STST behaviours are systematically tested and the influence they have on up-estuary migration is assessed in terms of relative energy expenditure. The parameterised model is then used predictively at Milford Haven Waterway to investigate potential impacts on the juvenile eel population due to entrainment in a power plant cooling water intake and outfall. Results from the Thames model case study indicate that including bed anchoring behaviour is essential for achieving a good comparison with the eel trap data and the choice of salinity detection threshold is also important. If daylight avoidance (diel) behaviour is not included, the most energy efficient migration is achieved using just two STST behaviours (ebb tide bed anchoring and upward migration during flood). With diel behaviour included, energy expenditure is greater, but some efficiency is regained by including all of the STST behaviours. For the Milford Haven case study, the model predicted a juvenile eel intake and outfall entrainment rate of 2.0% and 4.7%, respectively. It is concluded that the ABM is a valuable tool for assessing potential impacts on the recruitment of eels (extendable to other species) and could be used to assist in site-selection and low impact design of energy infrastructure in tidal environments.

Maritime vessel emission monitoring by an UAV gas sensor system

Monitoring the gas emission of maritime vessels is very challenging as the fields are almost inaccessible and the emission is very susceptible to the surrounding environment. In ultra-large-scale maritime scenes, mobile robots capable of gas detection and tracking will most likely fall into atmospheric turbulence. In this work, an enhanced tracking algorithm using our unmanned aerial vehicle (UAV) gas sensor system is proposed for maritime vessel emission monitoring. A global prediction is acquired by modelling the emission of individual vessel, while another local gradient direction can be calculated from the real-time onboard sensor measurements. By a probabilistic framework, the global prediction and the gradient detection are fused to generate a more reliable navigation for the UAV. In order to verify the proposed algorithm, real monitoring experiments using the developed UAV gas sensor system are presented and SO2, NOx contents in vessel emissions are conducted. A comparison between the UAV-based detections and the off-line sampled data is preformed, and the results indicate that the proposed algorithm has the advantage in exactly guiding the UAV towards the vessel emission.

Phase-field simulations and microstructural analysis of epitaxial growth during rapid solidification of additively manufactured AlSi10Mg alloy

In metal additive manufacturing, the unique microstructure from extreme process conditions determines the properties of the build part. In this study, the microstructural evolution of the additively manufactured AlSi10Mg alloy was studied to elucidate the relationship between its growth morphology and the local thermal gradient during rapid solidification. The microstructure of a uniaxially built specimen, as analyzed by electron backscatter diffraction (EBSD), exhibiting the [100](001) texture along the building direction. To investigate the local growth morphology, phase-field simulations were performed to examine the effect of grain misorientation on the previously built layer with respect to the thermal gradient direction. Furthermore, the solidification of partially melted microstructures observed during the EBSD measurements was simulated under the rapid solidification conditions. The obtained results revealed that the growth morphology of alloy grains during solidification strongly depended on the misorientation angle of the previously built layer measured with respect to the local thermal gradient direction. The simulated microstructures were in good agreement with the experimentally observed alloy morphology, which indicated that the utilized phase-field method could be potentially used in quantitative microstructural modeling.

A new adaptive-weighted total variation sparse-view computed tomography image reconstruction with local improved gradient information.

Inspired by the compressed sensing (CS) theory, introducing priori information of sparse image into sparse-view reconstruction algorithm of computed tomography (CT) can improve image quality. In recent years, as a special case of CS, total variation (TV) reconstruction algorithm that uses both image sparsity and prior information of edge direction have attracted much research interest in sparse-view image reconstruction due to its ability to preserve image edges. In this paper, we propose a new adaptive-weighted total variation (NAWTV) algorithm for CT image reconstruction, which is derived by considering local gradient direction continuity and the anisotropic edge property. The anisotropic edge property is used to consolidate the image sparsity, where the associated weights are expressed as a combination of exponential and cosine function. The weights can also be adjusted adaptively according the local image intensity gradient. The NAWTV algorithm is numerically implemented with gradient descent method. The typical Shepp-Logan phantom and FORBILD head phantom are employed to perform image reconstruction simulation. To evaluate performance of NAWTV algorithm, we compared it with TV and AwTV reconstruction algorithms in experiments. Numerical experimental results verified the effectiveness and feasibility of the proposed algorithm. Comparison results also showed that the NAWTV algorithm achieved a satisfactory performance in suppressing artifacts and preserving the edge structure details information of the reconstructed image.

A novel WiFi indoor positioning strategy based on weighted squared Euclidean distance and local principal gradient direction

PurposeThis paper aims to introduce the weighted squared Euclidean distance between points in signal space, to improve the performance of the Wi-Fi indoor positioning. Nowadays, the received signal strength-based Wi-Fi indoor positioning, a low-cost indoor positioning approach, has attracted a significant attention from both academia and industry.Design/methodology/approachThe local principal gradient direction is introduced and used to define the weighting function and an average algorithm based on k-means algorithm is used to estimate the local principal gradient direction of each access point. Then, correlation distance is used in the new method to find the k nearest calibration points. The weighted squared Euclidean distance between the nearest calibration point and target point is calculated and used to estimate the position of target point.FindingsExperiments are conducted and the results indicate that the proposed Wi-Fi indoor positioning approach considerably outperforms the weighted k nearest neighbor method. The new method also outperforms support vector regression and extreme learning machine algorithms in the absence of sufficient fingerprints.Research limitations/implicationsWeighted k nearest neighbor approach, support vector regression algorithm and extreme learning machine algorithm are the three classic strategies for location determination using Wi-Fi fingerprinting. However, weighted k nearest neighbor suffers from dramatic performance degradation in the presence of multipath signal attenuation and environmental changes. More fingerprints are required for support vector regression algorithm to ensure the desirable performance; and labeling Wi-Fi fingerprints is labor-intensive. The performance of extreme learning machine algorithm may not be stable.Practical implicationsThe new weighted squared Euclidean distance-based Wi-Fi indoor positioning strategy can improve the performance of Wi-Fi indoor positioning system.Social implicationsThe received signal strength-based effective Wi-Fi indoor positioning system can substitute for global positioning system that does not work indoors. This effective and low-cost positioning approach would be promising for many indoor-based location services.Originality/valueA novel Wi-Fi indoor positioning strategy based on the weighted squared Euclidean distance is proposed in this paper to improve the performance of the Wi-Fi indoor positioning, and the local principal gradient direction is introduced and used to define the weighting function.

Shack-Hartmann optimum threshold estimation for the point source

The Shack-Hartmann wavefront sensor (SHWFS) is an optical detection device based on the measurements of wavefront slopes. It is widely used in an adaptive optics system due to its simple structure and strong environment adaptability. The measuring accuracy of the SHWFS depends mainly on the accuracy of the spot image centroid in each sub-aperture. There are many centroid algorithms including the center of gravity algorithm, Gauss fitting algorithm, and correlation algorithm. As to the simplicity, robustness, high accuracy and stability, the center of gravity algorithm is more widely used. However, the accuracy of gravity algorithm is sensitive to the noise including discretization, aliasing, photon noise, readout noise, stray light, and direct current bias. To improve the accuracy of centroid, the output signals of SHWFS must be pre-processed to suppress the noise effect by using the method of thresholding in general. Many threshold methods have been presented to reduce the error of centroid and there theoretically exists an optimum threshold which causes the minimum error of centroid based on the characteristics of SHWFS and noise. However, it is difficult to separate the signals from the noises, and the optimum threshold cannot be estimated accurately in real time in the SHWFS systems. In this paper aiming at noises in SHWFS, which vary with time and space rapidly, a method based on the noise weighted function of the mean value of pixels and the local gradient direction of image signals in the moving windows is presented according to the characteristics of the Gaussian spot and noise distributions. Moreover, the theory and parameters determination of the method are analyzed. The method utilizes the probability that the pixels in the moving windows belong to the noise, and the probability is inversely proportional to the mean value of pixels and the local gradient direction of image signals, and so the monotonically reducing probability function of pixels is constructed. Finally, the standard deviation and mean value of noise can be obtained, and the estimation value of optimum threshold is equal to the mean value of noise plus three times the standard deviation of noise. To investigate the effects of the optimum threshold estimation with the different spot sizes, spot strengths and noise levels, the proposed algorithm is compared with traditional methods. The simulation and experimental results show that the proposed method could achieve higher accuracy, and the error between the threshold obtained by the method presented in this paper and theoretical optimum threshold is less than 10%, which is less than those from the traditional methods.

Implementing multi-directional molecular diffusion terms into Reaction Diffusion Manifolds (REDIMs)

The Reaction Diffusion Manifold (REDIM) method requires specifying an estimate of the system’s gradients as an input to characterize the molecular transport effects. So far, applications used a gradient estimate taken from simple combustion scenarios, like one-dimensional flame simulations. However, because the reduced model is typically used in turbulent combustion simulations, gradient estimates from a representative turbulent combustion scenario appear more appropriate. In this work, we study how gradients from Direct Numerical Simulations (DNS) of a turbulent flame can be incorporated into the REDIM method. A method for analyzing scalar gradients from a DNS simulation, and for representing them in a form that is usable in the REDIM method, is presented and applied. The analysis reveals a hierarchical structure in the local gradient directions of different scalars, which enables construction of gradient estimates on different levels of complexity (and fidelity). It is shown how these different levels appear in the dissipation terms of the REDIM equation. The influence of varying the detail of the DNS-based gradient estimate is studied, and it is found that the addition of the more complex levels has minor influence only on the REDIM. It can have an influence, however, onto the dynamics of the combustion system on the REDIM, i.e., onto the trajectory that the system takes on the REDIM.

Scale Invariant Feature Transform

Scale Invariant Feature Transform (SIFT) is an image descriptor for image-based matching developed by David Lowe (1999,2004). This descriptor as well as related image descriptors are used for a large number of purposes in computer vision related to point matching between different views of a 3-D scene and view-based object recognition. The SIFT descriptor is invariant to translations, rotations and scaling transformations in the image domain and robust to moderate perspective transformations and illumination variations. Experimentally, the SIFT descriptor has been proven to be very useful in practice for robust image matching and object recognition under real-world conditions.In its original formulation, the SIFT descriptor comprised a method for detecting interest points from a grey-level image at which statistics of local gradient directions of image intensities were accumulated to give a summarizing description of the local image structures in a local neighbourhood around each interest point, with the intention that this descriptor should be used for matching corresponding interest points between different images. Later, the SIFT descriptor has also been applied at dense grids (dense SIFT) which have been shown to lead to better performance for tasks such as object categorization and texture classification. The SIFT descriptor has also been extended from grey-level to colour images and from 2-D spatial images to 2+1-D spatio-temporal video.

Chatter detection algorithm based on machine vision

Production of precise high-value mechanical elements requires a hundred percent on-site control. Chatter may occur due to random events. Although an unaided human eye can also easily identify the presence of chatter marks, it is economically ineffective. Therefore, an algorithm based on machine vision signals was proposed for surface inspection. The algorithm was designed to build an error map of the examined surface and highlight the regions of probable imperfections. The algorithm is based on local gradient estimation, applied in respect to milling parameters. Estimated local gradient directions are used to calculate the ridge/valley orientations of the machined surface. The local ridge orientation is used for the purpose of computing the surface error map. An experiment was made to test the algorithm. Milled surfaces in both chatter-free and chatter-rich conditions have been analysed with the presented method.

A strong magnetic anomaly affects pigeon navigation

Pigeons were released in a strong magnetic anomaly with fast changes in intensity and gradients directions, about 60 km from their loft, and, for comparison, at the border of the anomaly and at a control site. The vanishing bearings were found to be closely related to the home direction, but unrelated to the local gradient directions. The vector lengths and the vanishing intervals, however, were significantly correlated with the maximum difference in intensity within a 2.5 km radius around the release site. This correlation was negative for the vector lengths and positive for the vanishing intervals, indicating that steep local gradients increase scatter between pigeons and delay their departure. These findings suggest that an irregular, fast changing magnetic field as found in the anomaly leads to confusion during the navigational processes. This, in turn, implies that pigeons can sense the respective changes in magnetic intensity. Magnetic cues seem to be included in the normal navigational processes that determine the departure direction.

Adaptive diffusion using Hesse matrix for detecting nonstationary signals

Adaptive diffusion, using a gradient‐square tensor, was proposed to improve the readability of time frequency representations (TFRs) of the Cohen class and the affine class in 2005. It works with a signal‐dependent kernel and can adapt to a wide range of nonstationary signals. Here we substitute the Hesse matrix for the gradient‐square tensor after making an analysis of local diffusion behavior in an adaptive diffusion process, in close conjunction with the local signature of the auto terms and the interference terms. The eigenvectors of the Hesse matrix not only can give the local average gradient direction and thus the local diffusion direction, but also can provide an explicit mathematical explanation for local diffusion behavior. The Hesse method is capable of providing TFRs of better readability than the gradient square tensor. The validity of the new method is verified by computational simulations.

Orientation behavior of homing pigeons at the Gernsheim anomaly

Pigeons were released at four release sites within the Gernsheim anomaly, a magnetic 'hill' with a peak 199 nT above the regional reference field and gentle 'slopes' to all sides, situated 44 km south of the Frankfurt loft. Local magnetic conditions at the sites differed in total intensity and in direction and steepness of the intensity gradient. At all sites, the pigeons were well oriented, showing counterclockwise deviations from the home directions that were most pronounced in the western part of the anomaly. There was no systematic difference in orientation behavior or homing performance between the sites within the anomaly and a control site outside. No effect of the local gradient direction was found, nor did the difference in intensity between home loft and the release site affect behavior. This argues against the use of magnetic navigational factors. However, pigeons released for the first time within the anomaly tended to have longer mean vectors with increasingly steeper gradients, which could mean that the birds might somehow have realized the anomalous nature of the local magnetic conditions and ignored them, relying on non-magnetic cues instead.

Detection and enhancement of line structures in an image by anisotropic diffusion

This paper describes a method to enhance line structures in a gray-level image. For this purpose, we blur the image using anisotropic Gaussian filters along the directions of each line structure. In a line structure region the gradients of image gray levels have a uniform direction. To find such line structures, we evaluate the uniformity of the directions of the local gradients. Before this evaluation, we need to smooth out small structures to obtain line directions. We, first, blur the given image by a set of Gaussian filters. The variance of the filter which maximizes the uniformity of the local gradient directions is detected position by position. Then, the line directions in the image are obtained from this blurred image. Finally, we blur the image using an anisotropic filter again along the directions, and enhance every line structure.

Optimal settings of residential oil burners

Residential oil burners are capable of almost complete burning of the fuel oil, without visible smoke, when they are operated to deliver approximately 12% CO 2 in the flue gases. The positions of the air damper and of the combustion nozzle are adjusted at start-up and during operation in order to maximize the combustion efficiency. In practice, one factor at a time is varied, starting with the air damper. However, this method fails to detect the interaction between air excess and nozzle position and results in non-optimal settings. Optimal designed experiments allow obtaining local regression models and statistical analyses indicate if experiment augmentation is required. The air damper and combustion nozzle settings are changed in the direction of local gradient until a second order model that contains the optimal point in its experimental region is obtained. The gain in combustion efficiency thus obtained may be up to 5% as compared with the classical approach.

A second‐order upwind finite‐volume method for the Euler solution on unstructured triangular meshes

AbstractA scheme for the numerical solution of the two‐dimensional (2D) Euler equations on unstructured triangular meshes has been developed. The basic first‐order scheme is a cell‐centred upwind finite‐volume scheme utilizing Roe's approximate Riemann solver. To obtain second‐order accuracy, a new gradient based on the weighted average of Barth and Jespersen's three‐point support gradient model is used to reconstruct the cell interface values. Characteristic variables in the direction of local pressure gradient are used in the limiter to minimize the numerical oscillation around solution discontinuities. An Approximate LU (ALU) factorization scheme originally developed for structured grid methods is adopted for implicit time integration and shows good convergence characterisitics in the test. To eliminate the data dependency which prohibits vectorization in the inversion process, a black‐gray‐white colouring and numbering technique on unstructured triangular meshes is developed for the ALU factorization scheme. This results in a high degree of vectorization of the final code. Numerical experiments on transonic Ringleb flow, transonic channel flow with circular bump, supersonic shock reflection flow and subsonic flow over multielement aerofoils are calculated to validate the methodology.