Showing papers on "Centroid published in 2018"

Associative Deep Clustering - Training a Classification Network with no Labels

Abstract: We propose a novel end-to-end clustering training schedule for neural networks that is direct, i.e. the output is a probability distribution over cluster memberships. A neural network maps images to embeddings. We introduce centroid variables that have the same shape as image embeddings. These variables are jointly optimized with the network’s parameters. This is achieved by a cost function that associates the centroid variables with embeddings of input images. Finally, an additional layer maps embeddings to logits, allowing for the direct estimation of the respective cluster membership. Unlike other methods, this does not require any additional classifier to be trained on the embeddings in a separate step. The proposed approach achieves state-of-the-art results in unsupervised classification and we provide an extensive ablation study to demonstrate its capabilities.

Know What Your Neighbors Do: 3D Semantic Segmentation of Point Clouds

Abstract: In this paper, we present a deep learning architecture which addresses the problem of 3D semantic segmentation of unstructured point clouds (Fig. 1). Compared to previous work, we introduce grouping techniques which define point neighborhoods in the initial world space and the learned feature space. Neighborhoods are important as they allow to compute local or global point features depending on the spatial extend of the neighborhood. Additionally, we incorporate dedicated loss functions to further structure the learned point feature space: the pairwise distance loss and the centroid loss. We show how to apply these mechanisms to the task of 3D semantic segmentation of point clouds and report state-of-the-art performance on indoor and outdoor datasets.

Know What Your Neighbors Do: 3D Semantic Segmentation of Point Clouds

Abstract: In this paper, we present a deep learning architecture which addresses the problem of 3D semantic segmentation of unstructured point clouds. Compared to previous work, we introduce grouping techniques which define point neighborhoods in the initial world space and the learned feature space. Neighborhoods are important as they allow to compute local or global point features depending on the spatial extend of the neighborhood. Additionally, we incorporate dedicated loss functions to further structure the learned point feature space: the pairwise distance loss and the centroid loss. We show how to apply these mechanisms to the task of 3D semantic segmentation of point clouds and report state-of-the-art performance on indoor and outdoor datasets.

A Robust Method to Measure Centroids of Spectral Lines

Abstract: Measuring the centroid of a spectral line is a common problem in astronomy. Many methods have been devised to overcome limitations due to either noise in the spectra or asymmetric profiles, the most common of which are the intensity weighted averages (first moment) or fits of analytical (typically Gaussian) profiles. If the spectral line can be considered a single component, we demonstrate that a simple quadratic fit to the pixel of maximum intensity and its two neighboring pixels provides a robust measure of the line centroid. This approach allows for a sub-velocity resolution precision on the line centroid, without be biases by noise or asymmetric features in the line profile and outperforming traditional methods in most situations.

Modulation-Constrained Clustering Approach to Blind Modulation Classification for MIMO Systems

Abstract: Blind modulation classification is a fundamental step before signal detection in cognitive radio networks where the knowledge of modulation scheme is not completely known. In this paper, a modulation-constrained (MC) clustering classifier is proposed for recognizing the modulation scheme with unknown channel matrix and noise variance for MIMO systems. By recognizing the fact that the received signals within an observation interval form into clusters and exploiting the intrinsic relationships between different digital modulation schemes, the modulation classification is transformed into a number of clustering problems, one for each modulation scheme, with the final classification decision based on the maximum likelihood criterion. To improve the learning efficiency, centroid reconstruction is proposed to recover all cluster centroids through a limited number of parameters by exploiting the structural relationships in constellation diagrams. Furthermore, a method to initialize the cluster centroids is also proposed. The proposed MC classifier together with centroid reconstruction and initialization methods not only reduce the number of parameters to be estimated, but also help to initialize the clustering algorithm for the enhanced convergence performance. Simulation results show that our algorithm can perform excellently even at low SNR and with very short observation interval length.

Positive and Unlabeled Learning via Loss Decomposition and Centroid Estimation

Abstract: Positive and Unlabeled learning (PU learning) aims to train a binary classifier based on only positive and unlabeled examples, where the unlabeled examples could be either positive or negative. The state-of-the-art algorithms usually cast PU learning as a cost-sensitive learning problem and impose distinct weights to different training examples via a manual or automatic way. However, such weight adjustment or estimation can be inaccurate and thus often lead to unsatisfactory performance. Therefore, this paper regards all unlabeled examples as negative, which means that some of the original positive data are mistakenly labeled as negative. By doing so, we convert PU learning into the risk minimization problem in the presence of false negative label noise, and propose a novel PU learning algorithm termed ?Loss Decomposition and Centroid Estimation? (LDCE). By decomposing the hinge loss function into two parts, we show that only the second part is influenced by label noise, of which the adverse effect can be reduced by estimating the centroid of negative examples. We intensively validate our approach on synthetic dataset, UCI benchmark datasets and real-world datasets, and the experimental results firmly demonstrate the effectiveness of our approach when compared with other state-of-the-art PU learning methodologies.

Centroid Transformations of Intuitionistic Fuzzy Values Based on Aggregation Operators

Abstract: Atanassov’s intuitionistic fuzzy sets extend the notion of fuzzy sets. In addition to Zadeh’s membership function, a non-membership function is also considered. Intuitionistic fuzzy values play a crucial role in both theoretical and practical progress of intuitionistic fuzzy sets. This study introduces and explores various types of centroid transformations of intuitionistic fuzzy values. First, we present some new concepts for intuitionistic fuzzy values, including upper determinations, lower determinations, spectrum triangles, simple intuitionistic fuzzy averaging operators and simply weighted intuitionistic fuzzy averaging operators. With the aid of these notions, we construct centroid transformations, weighted centroid transformations, simple centroid transformations and simply weighted centroid transformations. We provide some basic characterizations regarding various types of centroid transformations, and show their difference using an illustrating example. Finally, we focus on simple centroid transformations and investigate the limit properties of simple centroid transformation sequences. Among other facts, we show that a simple centroid transformation sequence converges to the simple intuitionistic fuzzy average of the lower and upper determinations of the first intuitionistic fuzzy value in the sequence.

Centroid computation for Shack-Hartmann wavefront sensor in extreme situations based on artificial neural networks.

Abstract: This paper proposes a method used to calculate centroid for Shack-Hartmann wavefront sensor (SHWFS) in adaptive optics (AO) systems that suffer from strong environmental light and noise pollutions. In these extreme situations, traditional centroid calculation methods are invalid. The proposed method is based on the artificial neural networks that are designed for SHWFS, which is named SHWFS-Neural Network (SHNN). By transforming spot detection problem into a classification problem, SHNNs first find out the spot center, and then calculate centroid. In extreme low signal-noise ratio (SNR) situations with peak SNR (SNRp) of 3, False Rate of SHNN-50 (SHNN with 50 hidden layer neurons) is 6%, and that of SHNN-900 (SHNN with 900 hidden layer neurons) is 0%, while traditional methods’ best result is 26 percent. With the increase of environmental light interference’s power, the False Rate of SHNN-900 remains around 0%, while traditional methods’ performance decreases dramatically. In addition, experiment results of the wavefront reconstruction are presented. The proposed SHNNs achieve significantly improved performance, compared with the traditional method, the Root Mean Square (RMS) of residual decreases from 0.5349 um to 0.0383 um. This method can improve SHWFS’s robustness.

Functional Decomposition for Bundled Simplification of Trail Sets

Abstract: Bundling visually aggregates curves to reduce clutter and help finding important patterns in trail-sets or graph drawings. We propose a new approach to bundling based on functional decomposition of the underling dataset. We recover the functional nature of the curves by representing them as linear combinations of piecewise-polynomial basis functions with associated expansion coefficients. Next, we express all curves in a given cluster in terms of a centroid curve and a complementary term, via a set of so-called principal component functions. Based on the above, we propose a two-fold contribution: First, we use cluster centroids to design a new bundling method for 2D and 3D curve-sets. Secondly, we deform the cluster centroids and generate new curves along them, which enables us to modify the underlying data in a statistically-controlled way via its simplified (bundled) view. We demonstrate our method by applications on real-world 2D and 3D datasets for graph bundling, trajectory analysis, and vector field and tensor field visualization.

A New Nearest Centroid Neighbor Classifier Based on K Local Means Using Harmonic Mean Distance

Abstract: The K-nearest neighbour classifier is very effective and simple non-parametric technique in pattern classification; however, it only considers the distance closeness, but not the geometricalplacement of the k neighbors Also, its classification performance is highly influenced by the neighborhood size k and existing outliers In this paper, we propose a new local mean based k-harmonic nearest centroid neighbor (LMKHNCN) classifier in orderto consider both distance-based proximity, as well as spatial distribution of k neighbors In our method, firstly the k nearest centroid neighbors in each class are found which are used to find k different local mean vectors, and then employed to compute their harmonic mean distance to the query sample Lastly, the query sample is assigned to the class with minimum harmonic mean distance The experimental results based on twenty-six real-world datasets shows that the proposed LMKHNCN classifier achieves lower error rates, particularly in small sample-size situations, and that it is less sensitive to parameter k when compared to therelated four KNN-based classifiers

Measurement of Free-Form Curved Surfaces Using Laser Triangulation

Abstract: Laser triangulation (LT) is widely used in many fields due to its good stability, high resolution and fast speed. However, the accuracy in these applications suffers from severe constraints on the data acquisition accuracy of LT. To solve this problem, the optical triangulation principle, the object equation of the optical path relationship and the deviation of the laser spot centroid are applied to deduce a mathematical model. Therefore, the image sensor inclination errors can be quantitatively calculated, and the collected data are compensated in real time. Further, a threshold sub-pixel gray-gravity (GG) extraction algorithm is proposed; the gradient function and Gaussian fit algorithm are used to set thresholds to remove the impact of the spot edge noise area on the center location; and polynomial interpolation is employed to enhance the data density of the traditional GG method, thus improving the data acquisition accuracy of LT. Finally, the above methods are applied to on-machine measurement of the American Petroleum Institute (API) thread and the screw rotor, respectively. The experimental results prove that the proposed method can significantly improve the measurement accuracy of free-form curved surfaces using LT and that the improved laser spot center extraction algorithm is more suitable for free-form curved surfaces with smaller curvature and more uniform curvature changes.

A Direct Approach for Determining the Switch Points in the Karnik–Mendel Algorithm

Abstract: The Karnik–Mendel algorithm is used to compute the centroid of interval type-2 fuzzy sets, determining the switch points needed for the lower and upper bounds of the centroid, through an iterative process. It is commonly acknowledged that there is no closed-form solution for determining such switch points. Many enhanced algorithms have been proposed to improve the computational efficiency of the Karnik–Mendel algorithm. However, all of these algorithms are still based on iterative procedures. In this paper, a direct approach based on derivatives for determining the switch points without multiple iterations has been proposed, together with mathematical proof that these switch points are correctly determining the lower and upper bounds of the centroid. Experimental simulations show that the direct approach obtains the same switch points, but is more computationally efficient than any of the existing (iterative) algorithms. Thus, we propose that this algorithm should be used in any application of interval type-2 fuzzy sets in which the centroid is required.

An ultrasonic positioning algorithm based on maximum correntropy criterion extended Kalman filter weighted centroid

Abstract: Ultrasonic positioning technology is being used in a wide range of application areas. In an ultrasonic positioning system, the noise of an ultrasound wave may not follow a Gaussian distribution but has a strong impulse because of many factors. A traditional extended Kalman filter based on the minimum mean square error would produce a linear estimation error and cannot handle a non-Gaussian noise effectively. Therefore, we propose a novel maximum correntropy criterion extended Kalman filter weighted centroid positioning algorithm based on a new Kalman gain formula to determine the maximum correntropy criterion. The maximum correntropy criterion maps the signal to a high-dimensional space and effectively deals with the non-Gaussian noise in ultrasonic positioning. In addition, the weighted centroid uses the results of the extended Kalman filter as inputs and reduces the impact of the linear estimation error on the positioning results. Experimental results show that the maximum correntropy criterion extended Kalman filter weighted centroid algorithm can improve the positioning accuracy by 60.06% over the extended Kalman filter and 22.83% compared with the maximum correntropy criterion extended Kalman filter. Overall, the proposed algorithm is more robust and effective.

Application of a Variational Mode Decomposition-Based Instantaneous Centroid Estimation Method to a Carbonate Reservoir in China

Abstract: A novel hydrocarbon detection technique named the variational mode decomposition (VMD)-based instantaneous centroid method is proposed in this letter. It reveals frequency-dependent amplitude anomalies that may reflect some details deeply buried within the intrinsic mode functions (IMFs) in particular frequency bands. Instantaneous amplitude and instantaneous frequency information from each IMF are used to generate each IMF instantaneous centroid. A weighted correlation scheme is employed to generate the VMD-based instantaneous centroid volume for a seismic trace. Model testing and field data from a carbonate reservoir in China illustrate that the VMD-based instantaneous centroid method can provide a better hydrocarbon-prone interpretation with a higher resolution and accuracy. Comparisons between the VMD-based instantaneous centroid method and the short-time Fourier transform, and continuous wavelet transform and prestack wave impedance inversion technology indicate that the proposed method is more convenient and can effectively target gas reservoirs. This letter presents a complementary approach to current methods used to extract frequency-dependent amplitude anomaly information.

CentroidNet: A Deep Neural Network for Joint Object Localization and Counting

Abstract: In precision agriculture, counting and precise localization of crops is important for optimizing crop yield. In this paper CentroidNet is introduced which is a Fully Convolutional Neural Network (FCNN) architecture specifically designed for object localization and counting. A field of vectors pointing to the nearest object centroid is trained and combined with a learned segmentation map to produce accurate object centroids by majority voting. This is tested on a crop dataset made using a UAV (drone) and on a cell-nuclei dataset which was provided by a Kaggle challenge. We define the mean Average F1 score (mAF1) for measuring the trade-off between precision and recall. CentroidNet is compared to the state-of-the-art networks YOLOv2 and RetinaNet, which share similar properties. The results show that CentroidNet obtains the best F1 score. We also explicitly show that CentroidNet can seamlessly switch between patches of images and full-resolution images without the need for retraining.

Blind Source Separation Method for Bearing Vibration Signals

Abstract: In underdetermined blind source separation (UBSS) of vibration signals, the estimation of the mixing matrix is often affected by noise and by the type of the used clustering algorithm. A novel UBSS method for the analysis of vibration signals, aiming to address the problem of the inaccurate estimation of the mixing matrix owing to noise and choice of the clustering method, is proposed here. The proposed algorithm is based on the modified k-means clustering algorithm and the Laplace potential function. First, the largest distance between data points is used to initialize the cluster centroid locations, and then the mean distance between clustering centroids average distance range of data points is used for updating the locations of cluster centroids. Next, the Laplace potential function that uses a global similarity criterion is applied to fine-tune the cluster centroid locations. Normalized mean squared error and deviation angle measures were used to assess the accuracy of the estimation of the mixing matrix. Bearing vibration data from Case Western Reserve University and our experimental platform were used to analyze the performance of the developed algorithm. Results of this analysis suggest that this proposed method can estimate the mixing matrix more effectively, compared with existing methods.

Automatic stacking‐velocity estimation using similarity‐weighted clustering

Abstract: Local seismic event slopes contain subsurface velocity information, and can be used to estimate seismic stacking velocity. In this paper, we propose a novel approach to estimate the stacking velocity automatically from seismic reflection data using similarity-weighted k-means clustering, in which the weights are local similarity between each trace in common midpoint gather and a reference trace. Local similarity reflects the local signal-to-noise ratio in common midpoint gather. We select the data points with high signal-to-noise ratio to be used in the velocity estimation with large weights in mapped traveltime and velocity domain by similarity-weighted k-means clustering with thresholding. By using weighted k-means clustering, we make clustering centroids closer to those data points with large weights which are more reliable and have higher signal-to-noise ratio. The interpolation is used to obtain the whole velocity volume after we have got velocity points calculated by weighted k-means clustering. Using the proposed method, one obtains a more accurate estimate of the stacking velocity because the similarity based weighting in clustering takes into account the signal-to-noise ratio and reliability of different data points in mapped traveltime and velocity domain. In order to demonstrate that, we apply the proposed method to synthetic and field data examples and the resulting images are of higher quality when compared to the ones obtained using existing methods. This article is protected by copyright. All rights reserved

An Accurate Centroiding Algorithm for PSF Reconstruction

Abstract: In this work, we present a novel centroiding method based on Fourier space Phase Fitting (FPF) for Point Spread Function (PSF) reconstruction. We generate two sets of simulations to test our method. The first set is generated by GalSim with an elliptical Moffat profile and strong anisotropy that shifts the center of the PSF. The second set of simulations is drawn from CFHT i band stellar imaging data. We find non-negligible anisotropy from CFHT stellar images, which leads to ～0.08 scatter in units of pixels using a polynomial fitting method (Vakili & Hogg). When we apply the FPF method to estimate the centroid in real space, the scatter reduces to ～0.04 in S/N?=?200 CFHT-like sample. In low signal-to-noise ratio (S/N; 50 and 100) CFHT-like samples, the background noise dominates the shifting of the centroid; therefore, the scatter estimated from different methods is similar. We compare polynomial fitting and FPF using GalSim simulation with optical anisotropy. We find that in all S/N (50, 100, and 200) samples, FPF performs better than polynomial fitting by a factor of ～3. In general, we suggest that in real observations there exists anisotropy that shifts the centroid, and thus, the FPF method provides a better way to accurately locate it.

Peak-locking centroid bias in Shack–Hartmann wavefront sensing

Abstract: Shack-Hartmann wavefront sensing relies on accurate spot centre measurement. Several algorithms were developed with this aim, mostly focused on precision, i.e. minimizing random errors. In the solar and extended scene community, the importance of the accuracy (bias error due to peak-locking, quantisation or sampling) of the centroid determination was identified and solutions proposed. But these solutions only allow partial bias corrections. To date, no systematic study of the bias error was conducted. This article bridges the gap by quantifying the bias error for different correlation peak-finding algorithms and types of sub-aperture images and by proposing a practical solution to minimize its effects. Four classes of sub-aperture images (point source, elongated laser guide star, crowded field and solar extended scene) together with five types of peak-finding algorithms (1D parabola, the centre of gravity, Gaussian, 2D quadratic polynomial and pyramid) are considered, in a variety of signal-to-noise conditions. The best performing peak-finding algorithm depends on the sub-aperture image type, but none is satisfactory to both bias and random errors. A practical solution is proposed that relies on the anti-symmetric response of the bias to the sub-pixel position of the true centre. The solution decreases the bias by a factor of ~7 to values of < 0.02 pix. The computational cost is typically twice of current cross-correlation algorithms.

Single- and two-person action recognition based on silhouette shape and optical point descriptors

Abstract: In this paper, we present action descriptors that are capable of performing single- and two-person simultaneous action recognition. In order to exploit the shape information of action silhouettes, we detect junction points and geometric patterns at the silhouette boundary. The motion information is exploited by using optical flow points. We compute centroid distance signatures to construct the junction points and optical flow-based action descriptors. By taking advantage of the distinct poses, we extract key frames and construct geometric pattern action descriptor, which is based on histograms of the geometric patterns classes obtained by a distance-based classification method. In order to exploit the shape and motion information simultaneously, we follow the information fusion strategy and construct a joint action descriptor by combining geometric patterns and optical flow descriptors. We evaluate the performance of these descriptors on the two widely used action datasets, i.e., Weizmann dataset (single-person actions) and SBU Kinect interaction dataset, clean and noisy versions (two-person actions). The experimental outcomes demonstrate the ability of the individual descriptors to give satisfactory performance on average. It is found that the joint action descriptor shows the best performance among the proposed descriptors due to its high discriminative power and also outperforms state-of-the-art approaches.

MIMO Over-the-Air Computation: Beamforming Optimization on the Grassmann Manifold

Abstract: To support future IoT networks with dense sensor connectivity, a technique called over-the-air computation (Air-Comp) was recently developed to enable a data-fusion center to receive a desired function (e.g., mean value) of sensing data from concurrent sensor transmissions. This is made possible by exploiting the superposition property of a multi-access channel. This work aims at further developing AirComp for next-generation multi-antenna multi-modal sensor networks where a multi-modal sensor monitors multiple environmental parameters such as temperature, pollution and humidity. To be specific, we design beamforming techniques for AirComp of multiple functions, each corresponding to a particular sensing-data type. Given the objective of minimizing sum mean-squared error of computed functions, the optimization of receive beamforming for multi-function AirComp is a NP-hard problem. The approximate problem based on tightening transmission-power constraints, however, is shown to be solvable using differential geometry. The solution is proved to be the weighted centroid of points on a Grassmann manifold, where each point represents the subspace spanned by the channel matrix of a sensor. Simulation results demonstrate the effectiveness of the proposed solution.

Enhancing keyword correlation for event detection in social networks using SVD and k -means: Twitter case study

Abstract: Extracting textual features from tweets is a challenging task due to the noisy nature of the content and the weak signal of most of the words used. In this paper, we propose using singular value decomposition (SVD) with clustering to group related words as enhanced signals for textual features in tweets in order to improve the correlation with events. The proposed method applies SVD to the time series vector for each feature to factorize the matrix of feature/day counts, to ensure the independence of the feature vectors. Then, k-means clustering is applied to build a look-up table that maps members of each cluster to the cluster centroid. The look-up table is used to map each feature in the original data to the centroid of its cluster. Then, we calculate the sum of the term-frequency vectors of all features in each cluster to the term-frequency vector of the cluster centroid. To evaluate the method, we calculated the correlations of the cluster centroids with the golden standard record vector before and after summing the vectors of the cluster members to the centroid vector. The proposed method is applied to multiple correlation techniques including the Pearson, Spearman, distance correlation, and Kendal Tao. The experiments also considered the different word forms and lengths of the features including keywords, n grams, skip grams, and bags-of-words. The correlation results are enhanced significantly as the highest correlation scores have increased from 0.22 to 0.70, and the average correlation scores have increased from 0.22 to 0.60.