Bicubic Research Articles

Defect extraction method for additive manufactured parts with improved learning-based image super-resolution and the Canny algorithm.

Additive manufacturing (AM) is a highly competitive, low-cost, and high-degree-of-manufacturing technology. However, AM still has limitations because of some defects. Thus, defect detection technology is essential for quality enhancement in the AM process. Super-resolution (SR) technology can be utilized to improve defect image quality and enhance defect extraction performance. This study proposes a defect extraction method for additive manufactured parts with improved learning-based image SR and the Canny algorithm (LSRC), which is based on direct mapping methodology. The LSRC method is compared with the bicubic interpolation algorithm and the neighbor embedding (NE) algorithm in SR reconstruction quality and robustness. The experimental results show that the proposed LSRC method achieves satisfactory performance in terms of the averaged information entropy (E), standard deviation (SD), peak signal-to-noise ratio (PSNR), and structural similarity (SSIM), which are 7.259, 45.301, 27.723, and 0.822, respectively. The accordingly average improvement rates of the E, SD, PSNR, and SSIM, are 0.45%, 7.15%, 5.85%, and 6.35% in comparison with the bicubic interpolation algorithm, while the comparison data are 0.97%, 13.40%, 10.55%, and 15.35% in terms of the NE algorithm. This indicates that the LSRC method is significantly better than the comparison algorithm in reconstruction quality and robustness, which is of great significance for the extraction and analysis of key defect information of additive manufactured parts.

Magnetic grid resolution enhancement using machine learning: A case study from the Eastern Goldfields Superterrane

Densely sampled geophysical surveys are a key driver for mineral exploration, but sample density, and therefore grid resolution, is limited by survey cost. Consequently, computational methods are resorted to for upsampling, or ‘super-resolving’, of gridded geophysical survey data. However, existing approaches such as interpolation filters do not leverage high-resolution detail from pre-existing geophysical surveys. Through the application of state-of-the-art deep learning super-resolution architectures, accurate high-resolution grids are predicted from corresponding low-resolution grid priors. Specifically, the RDN and ESRGAN+ neural network architectures, which were originally developed for enhancing images, are applied to enhance magnetic surveys and trained with high-resolution and low-resolution magnetic grids of the same extent. 80 m cell size magnetic grids are upsampled to 20 m cell size using this method, and predicted value and structural accuracy comparisons to the corresponding ground truth 20 m grids are presented over two test sites in the Eastern Goldfields Superterrane, Western Australia. The method based on RDN achieves 53 % lower error compared to Bicubic interpolation. The case study demonstrates that the deep learning approach can improve the resolution and frequency content of geophysical surveys without requiring additional sampling expense, while remaining accurate against known ground truth surveys. Super-resolution may assist in the interpretation of low-resolution survey grids, however these upsampled grids cannot perfectly recreate the accuracy of highly sampled surveys gridded at their optimal cell size.

Nanoindentation mapping defects filtration for heterogeneous materials using generative adversarial networks

Advanced composite materials with multiple phases and heterogeneous microstructure necessitate spatial mapping characterization of elastic modulus to develop constitutive relations and overall mechanical response. Such modulus mapping can be obtained using the nanoindentation technique, where the indenter tip raster over the selected microstructure region. Typically, a surface preparation procedure is done in the specimens to ensure proper contact between the indenter tip and sample surface. However, a near-perfect surface finish is unachievable in heterogeneous materials, primarily with ceramic reinforcements, due to the differential material removal rate during polishing. Thus, the nanoindenter records localized erroneous measurements due to differences in surface roughness and corresponding force response. This study establishes a novel deep learning-based strategy to rectify incorrect experimental spatial measurements acquire during nanoindentation modulus mapping. The integrated bicubic interpolation and generative adversarial networks (GANs) model was trained using 14 ceramic and 18 metallic data sets, each comprising 65,536 measurements. The developed algorithm was validated against experimental measurements on four unknown specimens. The standard deviation in measured elastic modulus reduces by ~50% in ceramics and ~ 72% in metallic samples. This computational framework proposes a novel approach to reducing uncertainty in materials' properties using state-of-the-art computer vision techniques.

Sparse Dictionary-Based Magnetic Resonance Superresolution Imaging with Joint Loss Function Learning.

Magnetic resonance image has important application value in disease diagnosis. Due to the particularity of its imaging mechanism, the resolution of hardware imaging needs to be improved by increasing radiation intensity and radiation time. Excess radiation can cause the body to overheat and, in severe cases, inactivate the protein. This problem is expected to be solved by the image superresolution method based on joint dictionary learning, which has good superresolution performance. In the process of dictionary learning, the loss function will directly affect the dictionary performance. The general method only uses the cascade error as the optimization function in dictionary training, and the method does not consider the individual reconstruction error of high- and low-resolution image dictionary. In order to solve the above problem, In this paper, the loss function of dictionary learning is optimized. While ensuring that the coefficients are sufficiently sparse, the high- and low-resolution dictionaries are trained separately to reduce the error generated by the joint high- and low-resolution dictionary block pair and increase the high-resolution reconstruction error. Experiments on neck and ankle MR images show that the proposed algorithm has better superresolution reconstruction performance on ×2 and ×4 compared with bicubic interpolation, nearest neighbor, and original dictionary learning algorithms.

Inter-Slice Resolution Improvement Using Convolutional Neural Network with Orbital Bone Edge-Aware in Facial CT Images.

The 3D modeling of orbital bones in facial CT images is essential to provide a customized implant for reconstructions of orbit and related structures during surgery. However, 3D models of the orbital bone show an aliasing effect and disconnected thin bone in the inter-slice direction because the slice thickness is two to three times larger than the pixel spacing. To improve the inter-slice resolution of facial CT images, we propose a method based on a 2D convolutional neural network (CNN) that uses the spatial information on the sagittal and axial planes and the orbital bone edge-aware (OBE) loss. First, intermediate slices are generated on the sagittal plane. Second, the generated intermediate slices are transformed to an axial image, which is then compared with the original axial image. To generate intermediate slices with an accurate orbital bone structure, the OBE loss considering the orbital bone structure on the sagittal and axial planes is used. To improve the perceptual quality of the generated intermediate slices, the feature map difference loss is additionally used on the axial plane. In the experiment, the proposed method showed the best performance among bilinear and bicubic interpolations, 3D SRGAN, and a 2D CNN-based method. Experimental results confirmed that the proposed method can generate intermediate slices with clear edges of thin bones as well as cortical bones on both the sagittal and the axial plane.

Deep learning based correction of low performing pixel in computed tomography

Low Performing Pixel (LPP)/bad pixel in CT detectors cause ring and streaks artifacts, structured non-uniformities and deterioration of the image quality. These artifacts make the image unusable for diagnostic purposes. A missing/defective detector pixel translates to a channel missing across all views in sinogram domain and its effect gets spill over entire image in reconstruction domain as artifacts. Most of the existing ring and streak removal algorithms perform correction only in the reconstructed image domain. In this work, we propose a supervised deep learning algorithm that operates in sinogram domain to remove distortions cause by the LPP. This method leverages CT scan geometry, including conjugate ray information to learn the interpolation in sinogram domain. While the experiments are designed to cover the entire detector space, we emphasize on LPPs near detector iso-center as these have most adverse impact on image quality specially if the LPPs fall on the high frequency region (bone-tissue interface). We demonstrated efficacy of the proposed method using data acquired on GE RevACT multi-slice CT system with flat-panel detector. Experimental results on head scans show significant reduction in ring artifacts regardless of LPP location in the detector geometry. We have simulated isolated LPPs accounting for 5% and 10% of total channels. Detailed statistical analysis illustrates approximately 5dB improvement in SNR in both sinogram and reconstruction domain as compared to classical bicubic and Lagrange interpolation methods. Also, with reduction in ring and streak artifacts, the perceptual image quality is improved across all the test images.

Antiocclusion Visual Tracking Algorithm Combining Fully Convolutional Siamese Network and Correlation Filtering.

Machine learning only uses single-channel grayscale features to model the target, and the filter solution process is relatively simple. When the target has a large change relative to the initial frame, the tracking effect is poor. When there is the same kind of target interference in the target search area, the tracking results will be poor. The tracking algorithm based on the fully convolutional Siamese network can solve these problems. By learning the similarity measurement function, the similarity between the template and the target search area is evaluated, and the target area is found according to the similarity. It adopts offline pre-training and does not update online for tracking, which has a faster tracking speed. According to this study, (1) considering the accuracy and speed, the target tracking algorithm based on correlation filtering performs well. A sample adaptive update model is introduced to eliminate unreliable samples, which effectively enhances the reliability of training samples. With simultaneous changes in illumination and scale, fast motion and in-plane rotation IPR can still be maintained. (2) Determined by calculating the Hessian matrix, in the Struck function, Bike3 parameter adjustment can achieve fast tracking, and Boat5 ensures that the system stability is maintained in the presence of interference factors. The position of the highest scoring point in the fine similarity score map of the same size as the search image is obtained by bicubic interpolation as the target position. (3) The parallax discontinuity caused by the object boundary cannot be directly processed as a smooth continuous parallax. The MeanShift vector obtained by calculating the target template feature and the feature to be searched can increase the accuracy by 53.1%, reduce the robustness by 31.8%, and reduce the error by 28.6% in the SiamVGG algorithm.

LSR: Lightening super-resolution deep network for low-light image enhancement

In long-range shooting during low-light, we can only obtain a Low-Resolution (LR) image with poor visibility due to the limitation of physical devices. To get high-quality distant images, optical lenses are the optimal choice, but they are quite expensive and bulky. Low-light enhancement and Super-Resolution (SR) are the necessity of mobile phones and surveillance cameras and have wide applications in video surveillance, remote sensing, and night photography. Usually, the images that are taken from long-range in low-light suffer the loss of details not only due to low photon count but also due to low Signal-to-Noise Ratio (SNR), which makes it a highly ill-posed problem. Therefore, to resolve these two problems simultaneously, we propose a Lightening Super-Resolution (LSR) deep network. The proposed network uses the back-projection to learn the enhanced and dark features in low-resolution space iteratively and up-sample the enhanced features at the last stage of the network to get the final enhanced and high-resolution image. In particular, to train the network, the low-light images of the publicly available LOw Light (LOL) dataset are down-sampled using bicubic interpolation, and ground truth images are used as enhanced high-resolution images. For a fair comparison, a series of already available SR networks have been trained on the mentioned dataset, and their performances are compared. The promising results open up many opportunities for future work.

MUST: A Multi-source Spatio-Temporal data fusion Model for short-term sea surface temperature prediction

Sea surface temperature (SST) plays an important role in various oceanic applications, including climate prediction, ocean environment monitoring and marine disaster warning. Although many approaches have been developed for predicting SST, most of them conduct the prediction only based on historical SST data. However, SST is essentially affected by many external factors, e.g., the short-wave radiation from the sun and the long-wave radiation from the atmosphere and ground, which are ignored by existing approaches. In this work, we proposed a Multi-source Spatio-Temporal data fusion model (MUST) to fuse multi-source data, including SST data and external factors, to improve the accuracy of short-term SST prediction. Concretely, MUST first introduces Bicubic Convolutional Interpolation (BCI) to address the issue of inconsistent spatial resolutions of multi-source data, then employs the Spatio-Temporal Dilated ConvLSTM (ST-DC) to learn the spatio-temporal features of SST and external factors, and finally fuses the learned features from multi-source data to predict SST with a Cross Data Fusion (CDF) component. As validated on two real datasets, MUST achieves much better performance than existing SST prediction approaches.

Image Enhancement-Based Detection with Small Infrared Targets

Today, target detection has an indispensable application in various fields. Infrared small-target detection, as a branch of target detection, can improve the perception capability of autonomous systems, and it has good application prospects in infrared alarm, automatic driving and other fields. There are many well-established algorithms that perform well in infrared small-target detection. Nevertheless, the current algorithms cannot achieve the expected detection effect in complex environments, such as background clutter, noise inundation or very small targets. We have designed an image enhancement-based detection algorithm to solve both problems through detail enhancement and target expansion. This method first enhances the mutation information, detail and edge information of the image and then improves the contrast between the target edge and the adjacent pixels to make the target more prominent. The enhancement improves the robustness of detection with background clutter or noise-flooded scenes. Moreover, bicubic interpolation is used on the input image, and the target pixels are expanded with upsampling, which enhances the detection effectiveness for tiny targets. From the results of qualitative and quantitative experiments, the algorithm proposed in this paper outperforms the existing work on various evaluation indicators.

Super-resolution reconstruction of GOSAT CO2 products using bicubic interpolation

Satellites provide global long-time series of spatio-temporal continuous CO2 observations. However, it is difficult to be applied to the study of small-scale carbon cycle because of its low spatial resolution. In this paper, the Greenhouse Gases Observing SATellite (GOSAT) XCO2 data are super-resolution reconstructed using bicubic interpolation, which improved the spatial resolution from 2.5° to 0.5°. CO2 measurements from ten selected TCCON sites are used to compare with the reconstructed GOSAT. Further, the high accuracy Orbiting Carbon Observatory-2 (OCO-2) data analysed by the combination of geographical grid statistics and kriging is used to evaluate the reconstructed data. The results show that compared with the original GOSAT data, the reconstructed GOSAT data not only improves the spatial resolution but also has little loss of the average accuracy. The mean error of original data has significant seasonal fluctuations with a peak from February to March and a trough from June to July.

Athlete Health Status Assessment Based on Edge Computing Model and Large-Scale Multimedia Health Data Analysis.

This study explores the athlete's health status evaluation system based on an edge computing model and large-scale multimedia health data analysis in order to improve the assessment effect of athletes' current health state. In addition to that, this research builds a wireless signal propagation model in edge areas using bicubic polynomial interpolation, improves the accuracy of DEM data by using bicubic polynomial interpolation, and digitally represents the sports field by utilizing DEM data. In addition to that, the model and DEM data are utilized in this paper to realize the site selection process for base station networking. The experimental analysis demonstrates that the athletes' health status evaluation system suggested in this research, which is based on the edge computing model and the large-scale multimedia health data analysis, has a good effect on the assessment of athletes' health.

IMPACT OF DEEP LEARNING-BASED SUPER-RESOLUTION ON BUILDING FOOTPRINT EXTRACTION

Abstract. Automated building footprints extraction from High Spatial Resolution (HSR) remote sensing images plays important roles in urban planning and management, and hazard and disease control. However, HSR images are not always available in practice. In these cases, super-resolution, especially deep learning (DL)-based methods, can provide higher spatial resolution images given lower resolution images. In a variety of remote sensing applications, DL based super-resolution methods are widely used. However, there are few studies focusing on the impact of DL-based super-resolution on building footprint extraction. As such, we present an exploration of this topic. Specifically, we first super-resolve the Massachusetts Building Dataset using bicubic interpolation, a pre-trained Super-Resolution CNN (SRCNN), a pre-trained Residual Channel Attention Network (RCAN), a pre-trained Residual Feature Aggregation Network (RFANet). Then, using the dataset under its original resolution, as well as the four different super-resolutions of the dataset, we employ the High-Resolution Network (HRNet) v2 to extract building footprints. Our experiments show that super-resolving either training or test datasets using the latest high-performance DL-based super-resolution method can improve the accuracy of building footprints extraction. Although SRCNN based building footprint extraction gives the highest Overall Accuracy, Intersection of Union and F1 score, we suggest using the latest super-resolution method to process images before building footprint extraction due to the fixed scale ratio of pre-trained SRCNN and low speed of convergence in training.

Terrain feature-aware deep learning network for digital elevation model superresolution

Neural networks (NNs) have demonstrated the potential to recover finer textural details from lower-resolution images by superresolution (SR). Given similar grid-based data structures, some researchers have transferred image SR methods to digital elevation models (DEMs). These efforts have yielded better results than traditional spatial interpolation methods. However, terrain data present inherently different characteristics and practical meanings compared with natural images. This makes it unsuitable for existing SR methods on perceptually visual features of images to be directly adopted for extracting terrain features. In this paper, we argue that the problem lies in the lack of explicit terrain feature modeling and thus propose a terrain feature-aware superresolution model (TfaSR) to guide DEM SR towards the extraction and optimization of terrain features. Specifically, a deep residual module and a deformable convolution module are integrated to extract deep and adaptive terrain features, respectively. In addition, explicit terrain feature-aware optimization is proposed to focus on local terrain feature refinement during training. Extensive experiments show that TfaSR achieves state-of-the-art performance in terrain feature preservation during DEM SR. Specifically, compared with the traditional bicubic interpolation method and existing neural network methods (SRGAN, SRResNet, and SRCNN), the RMSE of our results is improved by 1.1% to 23.8% when recovering the DEM from 120 m to 30 m, by 4.9% to 22.7% when recovering the DEM from 60 m to 30 m, and by 7.8% to 53.7% when recovering the DEM from 30 m to 10 m. The source code that has been developed is shared on Figshare (https://doi.org/10.6084/m9.figshare.19597201).

Impact localization with a weighted spectral cross correlation method

The impact localization is a typical inverse problem to identify the damage source of structures. To overcome the positioning problems of the low resolution caused by sparse reference points, a weighted spectral cross-correlation (WSCC) method is proposed. First, a mathematical model with a series of Fiber Bragg Grating (FBG) sensors is built to generate a database of transient frequency responses of each reference point when random shock happens. Second, to identify the spectrum difference between the impact point and the reference point, a cross correlation model is built by mean and normalization operations on the frequency domain. Third, to figure out the impact point among the adjacent reference points, bicubic spline interpolation is realized by subdividing the grid of the WSCC. To validate the accuracy of our suggested method, an impact localization system is designed with a multi FBG sensor network and a composite laminate. The experimental result shows that the error of our suggested method is almost ten times lower than that of other traditional methods. The weighted spectral cross-correlation has advantages in the field of structural health monitoring.

Bi Histogram Equalization Based Image Enhancement with Bicubic Interpolation

In image processing, enhancement histogram equalization is the widely used technique for contrast enhancement. However, this technique tends to change the brightness of the image. Here, the contrast and resolution of image were enhanced using the proposed Bi Histogram Equalization Based Image Enhancement with Bicubic Interpolation (BHBI) technique. Bi Histogram for contrast enhancement and bicubic interpolation for resolution enhancement has taken. Bi Histogram Equalization separates the input image's histogram into two based on input mean before equalizing them independently. Bicubic Interpolation can generate bigger or high-resolution image from one or more low resolution or smaller images. The performance of the BHBI method can be compared for some typical image’s cameraman lens that are applied to some existing enhancement methods like Adaptive Gamma Correction with Weighting Distribution (AGCWD), Adaptive Scale Adjustment Design of Unsharp Masking Filters (ASAUMF), Averaging Histogram Equalization (AVGHEQ), and Median-Mean Based Sub-Image-Clipped Histogram Equalization (MMSICHE). The performance of these existing techniques can be evaluated subjectively in terms of person illustration observation and measurably using Discrete Entropy (DE), Image Quality Index (IQI), Normalized Correlation Coefficient (NCC), Contrast Improvement Index (CII), and Absolute Mean Brightness Error (AMBE). The results obtained from the BHBI technique shows better when compared with respect to the various existing techniques.

Super-Resolution and Feature Extraction for Ocean Bathymetric Maps Using Sparse Coding.

The comprehensive production of detailed bathymetric maps is important for disaster prevention, resource exploration, safe navigation, marine salvage, and monitoring of marine organisms. However, owing to observation difficulties, the amount of data on the world’s seabed topography is scarce. Therefore, it is essential to develop methods that effectively use the limited data. In this study, based on dictionary learning and sparse coding, we modified the super-resolution technique and applied it to seafloor topographical maps. Improving on the conventional method, before dictionary learning, we performed pre-processing to separate the teacher image into a low-frequency component that has a general structure and a high-frequency component that captures the detailed topographical features. We learn the topographical features by training the dictionary. As a result, the root-mean-square error (RMSE) was reduced by 30% compared with bicubic interpolation and accuracy was improved, especially in the rugged part of the terrain. The proposed method, which learns a dictionary to capture topographical features and reconstructs them using a dictionary, produces super-resolution with high interpretability.

심층학습 기반 초해상화 기법을 이용한 슬로싱 압력장 복원에 관한 연구

Deep-learning-based Super-Resolution (SR) methods were evaluated to reconstruct pressure fields with a high resolution from low-resolution images taken from a coarse grid simulation. In addition to a canonical SRCNN(super-resolution convolutional neural network) model, two modified models from SRCNN, adding an activation function (ReLU or Sigmoid function) to the output layer, were considered in the present study. High resolution images obtained by three models were more vivid and reliable qualitatively, compared with a conventional super-resolution method of bicubic interpolation. A quantitative comparison of statistical similarity showed that SRCNN model with Sigmoid function achieved best performance with less dependency on original resolution of input images.

Содержание курса «Геометрическое моделирование» для направления подготовки «Математика и компьютерные науки»

In this paper has been considered the main content and distinctive features of the “Geometric Modeling” training course for the “Mathematics and Computer Science” training program 02.03.01 (“Mathematical and Computer Modeling” specialization).
 The goal of the “Geometric Modeling” course study is the assimilation of mathematical methods for construction of geometric objects with complex curved shapes, and techniques for their computer visualization by using polygons of curves and surfaces. Methods for construction of structures’ curved shapes using spline representations, as well as techniques for construction of surfaces and volumetric geometries using motion operations and basic logical operations on geometric objects are considered. The spline representations include linear and bilinear splines, Hermite cubic splines and Hermite surfaces, natural cubic and bicubic interpolation splines, Bezier curves and surfaces, rational Bezier splines, B-splines and B-spline surfaces, NURBS-curves and NURBS-surfaces, transfinite interpolation methods, and splines of surfaces with triangular form. Logical operations for intersection of two spline curves, and intersection of two parametric surfaces are considered. The principles of scientific visualization and computer animation are considered in this course as well.
 Some examples for visualization of initial data and results of curves and surfaces construction in two- and three-dimensional spaces through the software shell developed by authors and used by students while doing tests have been demonstrated. The software shell has a web interface with the WebGL library graphic support. Tasks for four practical studies in a computer classroom, as well as several variations of homework are represented.
 The problems occurring in preparation materials for some course sections are discussed, as well as the practical importance of acquired knowledge for the further progress of students.
 The paper may be interesting for teachers of “Geometric Modeling” and “Computer Graphics” courses aimed to students with a specialization in mathematics and information, as well as to those who independently develop software interfaces for algorithms of geometric modeling.

Multi-path deep learning framework on discrete pressure points to predict velocity field of pump-jet propulsor

The research has provided a fast and accurate deep learning framework, which can predict the wake velocity field of pump-jet propulsor based on discrete pressure points at different conditions. Once the model is trained, the velocity field can be obtained by discrete pressure points quickly, which the pressure point can be obtained by pressure sensors in actual engineering or experiments. The bicubic interpolation method and hybrid multi-path deep learning model were used to realize the spatiotemporal mapping relationship between pressure point and velocity fields. The predicted difference for different numbers of discrete pressure points and the probability distribution of difference were investigated. The differences between the predicted velocity field and the real flow field have been analyzed by the error cloud map. The structural similarity index (SSIM), correlation coefficient and peak signal noise ratio (PSNR) were used to show the prediction accuracy and generalization ability of the method. The results show that the predicted value of the velocity field is in good agreement with the CFD velocity field whether it is the velocity profile or the change trend of the velocity gradient at different conditions. The highest PSNR value and SSIM value reached 40.085 dB and 0.9902, and the correlation coefficient exceeded 0.96, which shows that the predicted value has a strong correlation with real field and it is convincing.