Nearest-neighbor Interpolation Research Articles

Diagnosis of Coronavirus Disease From Chest X-Ray Images Using DenseNet-169 Architecture.

The coronavirus disease (COVID-19) is a very contagious and dangerous disease that affects the human respiratory system. Early detection of this disease is very crucial to contain the further spread of the virus. In this paper, we proposed a methodology using DenseNet-169 architecture for diagnosing the disease from chest X-ray images of the patients. We used a pretrained neural network and then utilised the transfer learning method for training on our dataset. We also used Nearest-Neighbour interpolation technique for data preprocessing and Adam Optimizer at the end for optimization. Our methodology achieved 96.37 % accuracy which was better than that obtained using other deep learning models like AlexNet, ResNet-50, VGG-16, and VGG-19.

Ant weight-lifting algorithm for motion estimation

Every video coding standard includes and requires motion estimation and compensation. The full search algorithm, which provides the best motion estimation, has a very high computation cost. Researchers have developed several algorithms to reduce the cost of computation. However, most of these algorithms become trapped in local minima during the search. Population-based evolutionary algorithms are widely used to develop a computationally efficient and cost-effective motion estimation strategy. The most recent effort used the Jaya algorithm to develop a motion estimation process that outperformed the state-of-the-art test zone search algorithm. In this study, a motion estimation algorithm based on the ant weight-lifting approach is proposed. Previously, the ant weight-lifting algorithm was used to solve a variety of problems, such as image segmentation, signal compression, and so on. The ant weight-lifting algorithm's computation cost was reduced by adopting a fitness estimation method that uses nearest-neighbor interpolation and an early termination strategy. Compared to Jaya algorithm-based motion estimation, the proposed algorithm executes up to 3% more quickly and exhibits up to 1.2 dB less distortion.

Analysis and solution to aliasing artifacts in neural waveform generation models

In recent years, with the application of deep learning in speech synthesis, waveform generation models based on generative adversarial networks have achieved high quality comparable to natural speech. In most waveform generators, a neural upsampling unit plays an essential role as it is employed to upsample acoustic features to the sample point level. However, aliasing artifacts are observed in the generated speech regardless of whether transposed convolution, subpixel convolution, or nearest neighbor interpolation are used as temporary upsampling layers. Non-ideal upsampling filters produce aliasing, according to the Shannon-Nyquist sampling theorem. This paper aims to systematically analyze how aliasing artifacts are produced in non-ideal upsampling-based waveform generators. We investigate the HiFi-GAN and VITS generation processes and discover that high-frequency spectral details are generated based on low-frequency structures using the nonlinear transformation. However, the nonlinear transformation was unable to completely remove the low-frequency spectral imprint, which eventually manifested as spectral artifacts in generated waveforms. To suppress aliasing artifacts, a low-pass filter is applied after the upsampling layer, but this results in significant performance drops. The experimental results also show that aliasing speeds up the training process by filling high-frequency vacancies. In this regard, we propose to mix high-frequency components into low-pass filtered features, allowing models to converge faster while naturally avoiding artifacts. In addition, to assess the efficacy of our method, we created an artifact-detection algorithm based on structural similarity.

A precise geoid model of Bosnia and Herzegovina by the KTH method and its validation

We compute the first gravimetric geoid model for Bosnia and Herzegovina (BHG2022) based upon the KTH method and additive corrections. The BHG2022 is computed with the help of 34,820 terrestrial gravity points and a digital elevation model produced by the Shuttle Radar Topography Mission. The gravity data is gridded onto a 0.02 arc-degree resolution via the nearest neighbour interpolation method. ITU_GGC16 and ITG-Grace2010 models are utilised to provide long wavelengths of gravity field up to 280 and 180 maximum degree/order in the geoid computation, respectively. On the basis of an external evaluation by 609 Global Navigation Satellite System (GNSS)-levelling points homogeneously distributed over the country, the accuracy of the BHG2022 is estimated to be 5.65 cm absolutely. This is the most precise geoid model ever seen in this territory. The result highlights the significant effect of all gravity data obtained from various sources on the accuracy of the geoid model.

Thermal Fault Diagnosis of Electrical Equipment in Substations Using Lightweight Convolutional Neural Network

Real-time equipment condition monitoring is crucial to ensure the regular operation of the electric system. However, the deployed heavy convolutional neural networks (CNNs) are defective for edge computation and offline diagnosis. The purpose of this article is to present a system for detecting overheating faults of substation equipment using infrared photos and U-Net deep learning techniques. First, a stepwise encoder employs a lightweight CNN (LCNN) based on inverted residuals with depthwise separable convolution. The fault location is then decoded in the decoder using stepwise upsampling and nearest-neighbor interpolation. We additionally incorporate low-level detail feature information from the encoder to include additional fault data. Finally, testing findings on our dataset demonstrated the proposed method’s superior reliability and efficiency under a variety of evaluation metrics. Ultimately, we reached a lighter structure and leading estimation with a small training dataset, so our method is well-suited for deployment on mobile devices.

Spatial-temporal interpolation of satellite geomagnetic data to study long-distance animal migration

IntroductionIncreased access to remote sensing datasets presents opportunities to model an animal's in-situ experience of the landscape to study behavior and test hypotheses such as geomagnetic map navigation. MagGeo is an open-source tool that combines high spatiotemporal resolution geomagnetic data with animal tracking data. Unlike gridded remote sensing data, satellite geomagnetic data are point-based measurements of the magnetic field at the location of each satellite. MagGeo converts these measurements into geomagnetic values at an animal's location and time. The objective of this paper is to evaluate different interpolation methods and data frameworks within the MagGeo software and quantify how accurately MagGeo can model geomagnetic values and patterns as experienced by animals. MethodWe tested MagGeo outputs against data from 109 terrestrial geomagnetic observatories across 7 years. Unlike satellite data, ground-based data are more likely to represent how animals near the Earth's surface experience geomagnetic field dynamics. Within the MagGeo framework, we compared an inverse-distance weighting interpolation with three different nearest-neighbour interpolation methods. We also compared model geomagnetic data with combined model and satellite data in their ability to capture geomagnetic fluctuations. Finally, we fit a linear mixed-effect model to understand how error is influenced by factors like geomagnetic activity and distance in space and time between satellite and point of interest. Results and conclusionsThe overall absolute difference between MagGeo outputs and observatory values was <1% of the total possible range of values for geomagnetic components. Satellite data measurements closest in time to the point of interest consistently had lowest error which likely reflects the ability of the nearest neighbour in time interpolation method to capture small continuous daily fluctuations and larger discrete events like geomagnetic storms. Combined model and satellite data also capture geomagnetic fluctuations better than model data alone across most geomagnetic activity levels. Our linear mixed-effect models suggest that most of the variation in error can be explained by location-specific effects originating largely from local crustal biases, and that high geomagnetic activity usually predicts higher error though ultimately remaining within the 1% error range. Our results indicate that MagGeo can help researchers explore how animals may use the geomagnetic field to navigate long distances by providing access to data and methods that accurately model how animals moving near the Earth's surface experience the geomagnetic field.

MDT-based Intelligent Route Selection for 5G-Enabled Connected Ambulances.

The fifth generation of cellular network (5G) can facilitate in-ambulance patient monitoring, diagnosis, and treatment by a remote specialist. However, 5G coverage and link quality can vary in time and location. The ambulance route selection can help meet the communication requirements of the in-ambulance applications. In this paper, we propose an innovative ambulance route selection framework which combines the communication requirements along with the network coverage and resources. The framework leverages the minimization of drive test (MDT) data to estimate the network coverage along the ambulance routes. To address the uneven distribution of location-based user-generated MDT data, we examine the performance and trustworthiness of several interpolation techniques to enrich the global MDT map for route selection. A simulated analysis shows that the proposed framework can dynamically adapt to varying application requirements as well as rapidly changing network conditions such as outages. Results also reveal that nearest neighbor and kriging interpolation techniques help complement the proposed framework by addressing the data sparsity problem.

Object Detection Algorithm Based on Improved Feature Pyramid

Feature pyramid network is widely used in advanced object detection. By simply changing the network connection, the performance of small object detection can be greatly improved without increasing the amount of calculation of the original model. However, the algorithm still has some shortcomings. Therefore, a new attention feature pyramid network (attention PFN) is proposed. Firstly, an improved receptive field module is added, which can make full use of global and local context information. Secondly, the connection mode of the pyramid is further optimized. Deconvolution is used to replace the nearest neighbor interpolation in top-down up-sampling and a channel attention module is added to the horizontal connection to highlight important context information. Finally, adaptive fusion and spatial feature balance are used for each feature pyramid, so that the network can learn the weights of different feature layers. Each pyramid layer contains more discrimination information. Attention PFN is tested on Pascal VOC and MS COCO datasets, respectively. The experiment results revealed that the proposed method has better performance than the original algorithm. Therefore, the attention PFN is an effective algorithm.

Frame-rate up-conversion detection based on convolutional neural network for learning spatiotemporal features

With the advance in user-friendly and powerful video editing tools, anyone can easily manipulate videos without leaving prominent visual traces. Frame-rate up-conversion (FRUC), a representative temporal-domain operation, increases the motion continuity of videos with a lower frame-rate and is used by malicious counterfeiters in video tampering such as generating fake frame-rate video without improving the quality or mixing temporally spliced videos. FRUC is based on frame interpolation schemes and subtle artifacts that remain in interpolated frames are often difficult to distinguish. Hence, detecting such forgery traces is a critical issue in video forensics. This paper proposes a frame-rate conversion detection network (FCDNet) that learns forensic features caused by FRUC in an end-to-end fashion. The proposed network uses a stack of consecutive frames as the input and effectively learns interpolation artifacts using network blocks to learn spatiotemporal features. Moreover, it can cover the following three types of frame interpolation schemes: nearest neighbor interpolation, bilinear interpolation, and motion-compensated interpolation. In contrast to existing methods that exploit all frames to verify integrity, the proposed approach achieves a high detection speed because it observes only six frames to test its authenticity. Extensive experiments were conducted with conventional forensic methods and neural networks for video forensics to validate our research. The proposed work achieved an outstanding performance in terms of detecting the interpolated artifacts of FRUC. The experimental results also demonstrate that our model is robust against an unseen dataset, unlearned frame-rate, and unlearned quality factor. Furthermore, FCDNet can precisely localize the tampered region applied to manipulation along the time-domain through temporal localization.

Efficient algorithm for proper orthogonal decomposition of block-structured adaptively refined numerical simulations

Adaptive mesh refinement (AMR) is increasingly being used to simulate fluid flows that have vastly different resolution requirements throughout the computational domain. Proper orthogonal decomposition (POD) is a common tool to extract coherent structures from flow data and build reduced order models, but current POD algorithms do not take advantage of potential efficiency gains enabled by multi-resolution data from AMR simulations. Here, we explore a new method for performing POD on AMR data that eliminates repeated operations arising from nearest-neighbor interpolation of multi-resolution data onto uniform grids. We first outline our approach to reduce the number of computations with examples and provide the complete algorithms in the appendix. We examine the computational acceleration of the new algorithms compared to the standard POD method using synthetically generated AMR data and operation counting. We then use CPU times and operation counting to analyze data from an AMR simulation of an axisymmetric buoyant plume, finding that we are able to reduce the computational time by a factor of approximately 2−5 when using three levels of grid refinement. The new POD algorithm is the first to eliminate redundant operations for matrix multiplications with repeated values in each matrix, making it ideal for POD of data from AMR simulations.

An interpolation based steganographic technique with least-significant-bit and pixel value differencing in a pixel block

Over the &lt;span lang="EN-US"&gt;past few years, in order to improve the hiding capacity and the peak signal-to-noise ratio (PSNR) value, several steganographic techniques have been developed. Steganography has become a popular technique to transmit secret data through any medium. In image steganography, the human eye cannot easily identify the hidden data which is embedded into the image. Small changes are also not detected by the human eye. High hidden capacity along with high visual quality is provided by the pixel value differencing (PVD) method. This paper first proposes the method of interpolation between the pixel blocks and then applies the least-significant-bit (LSB) substitution technique with the PVD method. At the starting phase, the original image is fixedto a 2x2 block, then the nearest neighbor interpolation (NNI) technique is implemented. In the next phase, the upper left pixel isembedded by the k-bit LSB substitution method along with hidden data. The newly generated neighbouring pixel value is measured. Thus, data is hidden from three directions. Through this paper using two different range tables, the new algorithm is proposed. We observed that in both cases, PSNR and the hiding capacity are improved.&lt;/span&gt;

Machine Learning-Based Fast Integer and Fractional Vortex Modes Recognition of Partially Occluded Vortex Beams

In this work, a machine learning method is proposed to precisely classify partially occluded integer and fractional vortex modes for the first time in radio frequency (RF). Consequently, we introduce three training schemes, i.e., the direct recognition scheme with the phase data or the amplitude data (PD-DRS and AD-DRS), the phase data or amplitude data interpolated by nearest-neighbor interpolation algorithm (PD-NNI and AD-NNI), and the full data (FD) of the electric field with the NNI algorithm (FD-NNI), to recognize the topological charges. Based on the designed deep convolutional neural network (DCNN) models, the relationship between the test accuracy and the number of sampling points of the three schemes is presented. It is shown that <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$3\times3$ </tex-math></inline-formula> sampling points are enough for FD-NNI to achieve the classification accuracy of 98.2%. To validate the robustness of the proposed models, we evaluate them on the sample carrying up to 50% Gaussian noise, separately. Besides, the effects of propagation distance and the occlusion angle are also investigated. The numerical results present that the interpolated data performs better in terms of accuracy compared with the pure sampled data, among which FD-NNI possesses better generalization ability, suggesting great potential in the practical application of radio vorticity communication.

Interpolation Methods with Phase Control for Backprojection of Complex-Valued SAR Data.

Time-domain backprojection algorithms are widely used in state-of-the-art synthetic aperture radar (SAR) imaging systems that are designed for applications where motion error compensation is required. These algorithms include an interpolation procedure, under which an unknown SAR range-compressed data parameter is estimated based on complex-valued SAR data samples and backprojected into a defined image plane. However, the phase of complex-valued SAR parameters estimated based on existing interpolators does not contain correct information about the range distance between the SAR imaging system and the given point of space in a defined image plane, which affects the quality of reconstructed SAR scenes. Thus, a phase-control procedure is required. This paper introduces extensions of existing linear, cubic, and sinc interpolation algorithms to interpolate complex-valued SAR data, where the phase of the interpolated SAR data value is controlled through the assigned a priori known range time that is needed for a signal to reach the given point of the defined image plane and return back. The efficiency of the extended algorithms is tested at the Nyquist rate on simulated and real data at THz frequencies and compared with existing algorithms. In comparison to the widely used nearest-neighbor interpolation algorithm, the proposed extended algorithms are beneficial from the lower computational complexity perspective, which is directly related to the offering of smaller memory requirements for SAR image reconstruction at THz frequencies.

Joint gravity and magnetic inversion with trans-dimensional alpha shapes and autoregressive noise models

Typical geophysical inverse problems are ill-posed and non-unique which causes challenges for interpretation. To address these issues, deterministic inversion methods often apply constraints to parameter values, which control the effective number of parameters. However, such approaches can inhibit inference on complex structural boundaries. Bayesian trans-dimensional (trans-D) parametrizations for Earth structure partition space based on data information with the ability to adapt the parametrization locally to data information. Therefore, trans-D approaches can avoid under- or over-parametrizing regions of the model. Nonetheless, these parametrizations depend on the choice of partitioning types, such as Voronoi nodes or wavelet decomposition. In addition, trade-offs exist between spatial resolution and correlated data errors. We present a hierarchical model that treats both spatial and data noise parametrizations as trans-D to better incorporate trade-offs between noise and structure into uncertainty quantification. This includes a hierarchical spatial partitioning based on linear and nearest-neighbor interpolations and alpha shapes. The alpha shapes provide advantages for the inversion of potential field data by permitting flexibility in the shapes of structures of interest. The trans-D autoregressive noise model quantifies the impact of correlated noise on geophysical parameter estimates. We compare these methods with nested Voronoi partitioning and show differences in uncertainties, data fit, and parsimony of the parametrizations. Studies on simulated data show well-resolved structures and successful decorrelation of data residuals while requiring few parameters. The inversion of field data infers basement and salt broadly consistent with previous studies, but results show additional details that are consistent with independent geological knowledge.

Pixel Resolution Imaging in Parallel Phase-Shifting Digital Holography

Parallel phase-shifting digital holography (PPSDH) employing a polarization image sensor can suppress zero-order and twin-image noise through a single exposure, achieve instantaneous measurement of complex-valued dynamic objects, and have broad applications in the areas of biomedicine, etc. To improve the imaging resolution of PPSDH, we propose an oversampled super-pixel image reconstruction method, which can be expressed as the implementation of nearest-neighbor interpolation to replace blank pixels in sparse sub-phase-shift holograms. We found experimentally that the maximum spatial lateral resolution of the reconstructed image based on the existing super-pixel method, B-spline, bicubic, bilinear, and the proposed nearest-neighbor interpolation was 12.4 µm, 11.4 µm, 9.8 µm, 8.8 µm, and 7.8 µm, respectively. The main reason for not reaching the ideal value of 6.9 µm was the inherent residual zero-order and twin-image noise, which needs to be removed in the future.

Spatio-Temporal Spectrum Load Prediction Using Convolutional Neural Network and ResNet

Radio spectrum is a limited and increasingly scarce resource, which motivates alternative usage methods such as dynamic spectrum allocation (DSA). However, DSA requires an accurate prediction of spectrum usage in both time and spatial domains with minimal sensing cost. In this paper, we propose NN-ResNet prediction model to address this challenge in two steps. First, in order to make the best use of the sensors in the region, we deploy a deep learning prediction model based on convolutional neural networks (CNNs) and residual networks (ResNets), to predict spatio-temporal spectrum usage of the region. Second, to reduce sensing cost, the nearest neighbor (NN) interpolation is applied to recover spectrum usage data in the unsensed areas. In this case, fewer sensors are needed for prediction with the help of the reconstruction procedure. The model is verified through groups of comparison simulations in terms of the sensors’ sparsity and the number of transmitters involved. In addition, the proposed model is compared with CNN and ConvLSTM prediction model. The results show that the proposed NN-ResNet model maintains a lower error rate under various sparse sensor circumstances.

Back Projection Algorithm for Multi-Receiver Synthetic Aperture Sonar Based on Two Interpolators

The back projection (BP) algorithm is characterized by its high performance for multi-receiver synthetic aperture sonar (SAS). For this reason, it is usually used to evaluate the imaging performance of Fourier-domain methods. However, this algorithm suffers from a large computation load, and the imaging efficiency is seriously lowered. In order to improve the imaging performance, this paper proposes focusing the multi-receiver SAS data using the BP algorithm based on two interpolators, including the linear interpolation and nearest-neighbor interpolation. The former interpolation is used to decrease the interpolation error based on adjacent sampled data; the latter estimates the data at the desired moment by assigning the data value of the nearest sample as estimated data. Then, the imaging performance of the presented method is discussed in detail based on simulations and real-data processing. With the presented method, the imaging performance can be improved without a loss of efficiency compared to nearest-neighbor interpolation without an upsampling operation. In comparison with the traditional BP algorithm, the presented method can be used to improve the imaging efficiency without any loss of performance.

Gaussian process regression for ultrasound scanline interpolation.

Purpose: In ultrasound imaging, interpolation is a key step in converting scanline data to brightness-mode (B-mode) images. Conventional methods, such as bilinear interpolation, do not fully capture the spatial dependence between data points, which leads to deviations from the underlying probability distribution at the interpolation points. Approach: We propose Gaussian process ( ) regression as an improved method for ultrasound scanline interpolation. Using ultrasound scanlines acquired from two different ultrasound scanners during in vivo trials, we compare the scanline conversion accuracy of three standard interpolation methods with that of regression, measuring the peak signal-to-noise ratio (PSNR) and mean absolute error (MAE) for each method. Results: The PSNR and MAE scores show that regression leads to more accurate scanline conversion compared to the nearest neighbor, bilinear, and cubic spline interpolation methods, for both datasets. Furthermore, limiting the interpolation window size of regression to 15 reduces computation time with minimal to no reduction in PSNR. Conclusions: regression quantitatively leads to more accurate scanline conversion and provides uncertainty estimates at each of the interpolation points. Our windowing method reduces the computational cost of using regression for scanline conversion.

A Machine Learning Framework to Predict the Tensile Stress of Natural Rubber: Based on Molecular Dynamics Simulation Data.

Natural rubber (NR), with its excellent mechanical properties, has been attracting considerable scientific and technological attention. Through molecular dynamics (MD) simulations, the effects of key structural factors on tensile stress at the molecular level can be examined. However, this high-precision method is computationally inefficient and time-consuming, which limits its application. The combination of machine learning and MD is one of the most promising directions to speed up simulations and ensure the accuracy of results. In this work, a surrogate machine learning method trained with MD data is developed to predict not only the tensile stress of NR but also other mechanical behaviors. We propose a novel idea based on feature processing by combining our previous experience in performing predictions of small samples. The proposed ML method consists of (i) an extreme gradient boosting (XGB) model to predict the tensile stress of NR, and (ii) a data augmentation algorithm based on nearest-neighbor interpolation (NNI) and the synthetic minority oversampling technique (SMOTE) to maximize the use of limited training data. Among the data enhancement algorithms that we design, the NNI algorithm finally achieves the effect of approaching the original data sample distribution by interpolating at the neighborhood of the original sample, and the SMOTE algorithm is used to solve the problem of sample imbalance by interpolating at the clustering boundaries of minority samples. The augmented samples are used to establish the XGB prediction model. Finally, the robustness of the proposed models and their predictive ability are guaranteed by high performance values, which indicate that the obtained regression models have good internal and external predictive capacities.

Method of voltage setting for power battery simulator using successive nearest-neighbor interpolation

The power battery simulator is important equipment in new energy vehicle test platform and other industrial fields, and the battery model is the key to emulate the battery characteristics accurately. A novel method of given voltage for the power battery simulator is proposed in the paper. According to the state of charge (SOC) of the power battery, three sub-tables with different resolutions are established corresponding to the initial, stationary, and final stage of SOC. To ensure the accuracy of a given voltage under small data capacity, a successive nearest-neighbor interpolation (SNNI) algorithm is proposed for processing battery model data and the iterative calculation is carried out for the data obtained by looking up the table until it approaches the required value. Simulations and experiments are done on various capacity data tables, which are obtained by employed different sampling ratios to reduce the data capacity, and the influence of iteration number selection on the accuracy of the algorithm is discussed. The results show that, after 4–5 iterations, the given voltage accuracy is controlled within 0.03 V by the SNNI algorithm with only 1.6% capacity of the standard model table, which is greatly improved compared with the traditional method, and the feasibility is verified by executing the proposed algorism in a DSP system.