Convolution Step Research Articles

An Automatic Internal Wave Recognition Algorithm Based on CNN Applicable to an Ocean Data Buoy System

The application of internal wave recognition to the buoy system is of great significance to enhance the understanding of the ocean internal wave phenomenon and provide more accurate data and information support. This article proposes an automatic internal wave recognition algorithm based on convolutional neural networks (CNN), which is used in the tight-profile intelligent buoy system. The sea profile temperature data were collected using the Bailong buoy system in the Andaman Sea in 2018. The CNN network structure is applied to feature compression of ocean temperature profile data, reducing the input feature amount of the feature recognition network, thereby reducing the overall algorithm parameters and computational complexity. By adjusting the number of convolution kernels and the length of convolution steps, the original data features in the time domain and the space domain are compressed, respectively. The experimental results show that the identification accuracy and robustness of this method are clearly superior to those of other methods. Additionally, the parameter number and calculation amount of this algorithm are very tiny, which greatly improves the possibility of its deployment in the buoy system.

A study on the effect of different machine learning algorithms on soccer footwork recognition under trajectory tracking theory

Abstract This paper aims to investigate the effectiveness and influencing factors of different machine learning algorithms on soccer footwork recognition. In this paper, we use inertial sensors to obtain the basic data of soccer players’ movements, then convert them into initial data of footwork using pose representation and pose-solving filtering. The value of K mainly influences the classification accuracy of KNN, and the highest accuracy of 67.23% is achieved when K is 5. The classification accuracy of SVM is related to the choice of the distance function. The accuracy of CNN is mainly affected by the size of the convolutional kernel and the convolutional step size, and the highest accuracy is 73.82%. The machine learning-based soccer step recognition can improve the recognition accuracy of traditional physical methods and provide scientific sports guidance for soccer players’ training.

Spectral pre-processing and non-linear calibration with convolutional kernel partial least-squares. Teaching new tricks to an old dog

A recent trend in multivariate calibration of non-linear systems is to simplify data processing models, avoiding, if possible, some complex deep learning approaches. Contributing to this line of work, convolutional kernel partial least-squares (CKPLS) is introduced both for finding the best spectral pre-processing procedure for reducing the impact of radiation scattering and for handling non-linearities in the data. CKPLS is a combination of a previous convolutional step for pre-processing the spectra with the well-known kernel PLS regression model for coping with non-linear relationships between spectral signatures and analyte concentrations or target sample properties. The convolutional step is driven by particle swarm optimization (PSO), which estimates the coefficients of a moving window spectral pre-processing. This convolutional phase, previous to KPLS, is a viable alternative to the few available methods for finding the best mathematical pre-processing of the spectra, which is usually performed by trial and error. Analytical results concerning the calibration of selected analytes from partially selective spectra are employed to illustrate the performance of CKPLS. For this purpose, both simulated and experimental data sets have been employed, showing that automatic pre-processing of spectra is possible, with a success which is comparable to classical methods such as computing the spectral derivatives.

Efficient Discretization of Movement Kernels for Spatiotemporal Capture–Recapture

Spatially explicit capture–recapture (SECR) models treat detection probability as a function of the distance between each animal and its notional activity centre. Open-population variants of these models (open SECR) are increasingly used to estimate the vital rates (survival and recruitment) of spatial populations subject to turnover between sampling times. If activity centres also move between sampling times then modelling the movement can reduce bias in estimates of vital rates. The usual movement model in open SECR is a random walk with step length governed by a probability kernel. Space is discretized in open SECR for computational convenience, and in some implementations this includes truncation of the probability kernel. Computations for the movement submodel are nevertheless very time-consuming owing to the repeated convolution steps and the need to manage boundary effects. A novel ‘sparse’ discretized kernel is proposed that greatly reduces fitting time. The sparse kernel was tested by simulation and applied to two datasets. Differences between models fitted using the sparse and full kernels were minor and unlikely to matter in practice. The sparse kernel extends the practical limits of the movement modelling in open SECR to greater dispersal distances and greater spatial resolution. Supplementary materials accompanying this paper appear online.

Development of a chemically intuitive filter for chemical graph convolutional network

AbstractThis work proposes the use of (: identity matrix; : adjacency matrix), instead of , the normalized form of which has intensively been used for the construction of graph convolutional networks (GCNs), in deep‐learning chemistry. The performance of the GCN model with in its convolution step is at least on a par with the vanilla GCN that uses (: degree matrix of ) in various chemistry datasets, such as FreeSolv, ESOL, lipophilicity, and blood–brain barrier penetration datasets. It could be seen that the use of might be more chemically intuitive than the use of , potentially embracing the information on bond properties, such as dipole moment, and functional groups in a molecule. This work suggests unavoidable necessity of tackling molecular‐representation problems in deep‐learning chemistry from unprecedented angles of view for advanced development and construction of chemically intuitive deep‐learning models.

Wearable Sensor-Based Step Length Estimation During Overground Locomotion Using a Deep Convolutional Neural Network.

Step length is a critical gait parameter that allows a quantitative assessment of gait asymmetry. Gait asymmetry can lead to many potential health threats such as joint degeneration, difficult balance control, and gait inefficiency. Therefore, accurate step length estimation is essential to understand gait asymmetry and provide appropriate clinical interventions or gait training programs. The conventional method for step length measurement relies on using foot-mounted inertial measurement units (IMUs). However, this may not be suitable for real-world applications due to sensor signal drift and the potential obtrusiveness of using distal sensors. To overcome this challenge, we propose a deep convolutional neural network-based step length estimation using only proximal wearable sensors (hip goniometer, trunk IMU, and thigh IMU) capable of generalizing to various walking speeds. To evaluate this approach, we utilized treadmill data collected from sixteen able-bodied subjects at different walking speeds. We tested our optimized model on the overground walking data. Our CNN model estimated the step length with an average mean absolute error of 2.89 ± 0.89 cm across all subjects and walking speeds. Since wearable sensors and CNN models are easily deployable in real-time, our study findings can provide personalized real-time step length monitoring in wearable assistive devices and gait training programs.

MIFFuse: A Multi-Level Feature Fusion Network for Infrared and Visible Images

Image fusion operation is beneficial to many applications and is also one of the most common and critical computer vision challenges. The perfect infrared and visible image fusion results should include the important infrared targets while preserving visible textural detail information as much as possible. A novel infrared and visible image fusion framework is proposed for this purpose. In this paper, the proposed fusion network (MIFFuse) is an end-to-end, multi-level-based fusion network for infrared and visible images. The presented approach makes effective use of the intermediate convolution layer’s output features to preserve the primary image fusion information. We also build a cat_block to swap information between two paths to gain more sufficient information during the convolution steps. To reduce the model’s running time even further, the proposed method that reduces the number of feature channels while maintaining the accuracy of the fusion performance. Extensive experiments on the TNO and CVC-14 image fusion datasets show that our MIFFuse outperforms the other methods in terms of both subjective visual effects and quantitative metrics. Furthermore, MIFFuse is approximately twice as fast as the most recent state-of-the-art methods. Our code and models can be found at https://github.com/depeng6/MIFFuse .

Layer-wise relevance propagation of InteractionNet explains protein\u2013ligand interactions at the atom level

Development of deep-learning models for intermolecular noncovalent (NC) interactions between proteins and ligands has great potential in the chemical and pharmaceutical tasks, including structure–activity relationship and drug design. It still remains an open question how to convert the three-dimensional, structural information of a protein–ligand complex into a graph representation in the graph neural networks (GNNs). It is also difficult to know whether a trained GNN model learns the NC interactions properly. Herein, we propose a GNN architecture that learns two distinct graphs—one for the intramolecular covalent bonds in a protein and a ligand, and the other for the intermolecular NC interactions between the protein and the ligand—separately by the corresponding covalent and NC convolutional layers. The graph separation has some advantages, such as independent evaluation on the contribution of each convolutional step to the prediction of dissociation constants, and facile analysis of graph-building strategies for the NC interactions. In addition to its prediction performance that is comparable to that of a state-of-the art model, the analysis with an explainability strategy of layer-wise relevance propagation shows that our model successfully predicts the important characteristics of the NC interactions, especially in the aspect of hydrogen bonding, in the chemical interpretation of protein–ligand binding.

CnnAlpha: Protein disordered regions prediction by reduced amino acid alphabets and convolutional neural networks.

Intrinsically disordered regions (IDR) play an important role in key biological processes and are closely related to human diseases. IDRs have great potential to serve as targets for drug discovery, most notably in disordered binding regions. Accurate prediction of IDRs is challenging because their genome wide occurrence and a low ratio of disordered residues make them difficult targets for traditional classification techniques. Existing computational methods mostly rely on sequence profiles to improve accuracy which is time consuming and computationally expensive. This article describes an ab initio sequence-only prediction method-which tries to overcome the challenge of accurate prediction posed by IDRs-based on reduced amino acid alphabets and convolutional neural networks (CNNs). We experiment with six different 3-letter reduced alphabets. We argue that the dimensional reduction in the input alphabet facilitates the detection of complex patterns within the sequence by the convolutional step. Experimental results show that our proposed IDR predictor performs at the same level or outperforms other state-of-the-art methods in the same class, achieving accuracy levels of 0.76 and AUC of 0.85 on the publicly available Critical Assessment of protein Structure Prediction dataset (CASP10). Therefore, our method is suitable for proteome-wide disorder prediction yielding similar or better accuracy than existing approaches at a faster speed.

A deep neural network for fast and accurate scatter estimation in quantitative SPECT/CT under challenging scatter conditions.

A major challenge for accurate quantitative SPECT imaging of some radionuclides is the inadequacy of simple energy window-based scatter estimation methods, widely available on clinic systems. A deep learning approach for SPECT/CT scatter estimation is investigated as an alternative to computationally expensive Monte Carlo (MC) methods for challenging SPECT radionuclides, such as 90Y. A deep convolutional neural network (DCNN) was trained to separately estimate each scatter projection from the measured 90Y bremsstrahlung SPECT emission projection and CT attenuation projection that form the network inputs. The 13-layer deep architecture consisted of separate paths for the emission and attenuation projection that are concatenated before the final convolution steps. The training label consisted of MC-generated "true" scatter projections in phantoms (MC is needed only for training) with the mean square difference relative to the model output serving as the loss function. The test data set included a simulated sphere phantom with a lung insert, measurements of a liver phantom, and patients after 90Y radioembolization. OS-EM SPECT reconstruction without scatter correction (NO-SC), with the true scatter (TRUE-SC) (available for simulated data only), with the DCNN estimated scatter (DCNN-SC), and with a previously developed MC scatter model (MC-SC) were compared, including with 90Y PET when available. The contrast recovery (CR) vs. noise and lung insert residual error vs. noise curves for images reconstructed with DCNN-SC and MC-SC estimates were similar. At the same noise level of 10% (across multiple realizations), the average sphere CR was 24%, 52%, 55%, and 67% for NO-SC, MC-SC, DCNN-SC, and TRUE-SC, respectively. For the liver phantom, the average CR for liver inserts were 32%, 73%, and 65% for NO-SC, MC-SC, and DCNN-SC, respectively while the corresponding values for average contrast-to-noise ratio (visibility index) in low-concentration extra-hepatic inserts were 2, 19, and 61, respectively. In patients, there was high concordance between lesion-to-liver uptake ratios for SPECT reconstruction with DCNN-SC (median 4.8, range 0.02-13.8) compared with MC-SC (median 4.0, range 0.13-12.1; CCC = 0.98) and with 90Y PET (median 4.9, range 0.02-11.2; CCC = 0.96) while the concordance with NO-SC was poor (median 2.8, range 0.3-7.2; CCC = 0.59). The trained DCNN took ~ 40s (using a single i5 processor on a desktop computer) to generate the scatter estimates for all 128 views in a patient scan, compared to ~ 80min for the MC scatter model using 12 processors. For diverse 90Y test data that included patient studies, we demonstrated comparable performance between images reconstructed with deep learning and MC-based scatter estimates using metrics relevant for dosimetry and for safety. This approach that can be generalized to other radionuclides by changing the training data is well suited for real-time clinical use because of the high speed, orders of magnitude faster than MC, while maintaining high accuracy.

Live Semantic 3D Perception for Immersive Augmented Reality.

Semantic understanding of 3D environments is critical for both the unmanned system and the human involved virtual/augmented reality (VR/AR) immersive experience. Spatially-sparse convolution, taking advantage of the intrinsic sparsity of 3D point cloud data, makes high resolution 3D convolutional neural networks tractable with state-of-the-art results on 3D semantic segmentation problems. However, the exhaustive computations limits the practical usage of semantic 3D perception for VR/AR applications in portable devices. In this paper, we identify that the efficiency bottleneck lies in the unorganized memory access of the sparse convolution steps, i.e., the points are stored independently based on a predefined dictionary, which is inefficient due to the limited memory bandwidth of parallel computing devices (GPU). With the insight that points are continuous as 2D surfaces in 3D space, a chunk-based sparse convolution scheme is proposed to reuse the neighboring points within each spatially organized chunk. An efficient multi-layer adaptive fusion module is further proposed for employing the spatial consistency cue of 3D data to further reduce the computational burden. Quantitative experiments on public datasets demonstrate that our approach works 11× faster than previous approaches with competitive accuracy. By implementing both semantic and geometric 3D reconstruction simultaneously on a portable tablet device, we demo a foundation platform for immersive AR applications.

A New Computational Approach to Estimate the Subdivision Depth of n-Ary Subdivision Scheme

The n-ary subdivision scheme has traditionally been designed to generate smooth curve and surface from control polygon. In this paper, we propose a new subdivision depth computation technique for n-ary subdivision scheme. The existing techniques do not ensure the computation of subdivision depth unless some strong condition is assumed on the mask of the scheme. But our technique relaxes the effect of strong condition assumed on the mask of the scheme by increasing the number of convolution steps. Consequently, a more precise subdivision depth technique for a given error tolerance is presented in this paper.

Data Authentication Based on Discrete Random Convolutional Neural Network

Data authentication is an efficient way to enhance the security of data transmission, especially for the resource-limited terminal devices in edge computing system. Authentication through physical layer channel characteristics is a lightweight method compared with that based on cryptography. With the popularity of deep learning, this paper proposed a discrete random convolutional neural network applicable for the channel matrix, which extracts more feature information in convolution step without increasing the required memory by discrete random movement of the convolutional kernel. It’s suitable for protecting the data transmission between micro terminals in edge computing.

Acoustic Impedance Deblurring With a Deep Convolution Neural Network

Domain-specific methods for deblurring particular sorts of objects have gained increasing attention due to the ineffectiveness of generic methods. We present a simple and effective convolutional neural network that deblurs postinversion acoustic impedance images. The architecture of our model consists of a convolutional layer that highlights edges and contours related to interfaces between rock layers; a locally connected layer that performs a convolutional step with unshared weights; and, finally, two fully connected layers that perform a nonlinear estimation of acoustic impedance values. We use the updated Standford VI reservoir model as training data set, which is composed of 150 acoustic impedance sections, each section with 200 traces. In this letter, we adopt a strong supervised learning that exploit, trace by trace, the data set of the inverted and ground truth impedance images. We also present an analysis comparing the frequency bandwidth among the latent, blurry, and deblurred images. Furthermore, the peak signal-to-noise ratio is calculated and compared with a classical deblurring method. We additionally address the requirement of deep learning for the huge amount of training examples by inserting rectified linear units and keeping the network architecture simple. The experimental results demonstrate the efficacy of the proposed method.

Filtering of pulses from particle detectors by means of Singular Value Decomposition (SVD)

This paper presents a novel methodology to filter pulses coming from particle detectors. It is based on variable-in-time convolutions in which one of the operands is the input pulse and the other is a vector that changes with every convolution step. This is equivalent to multiply every incoming pulse by a filtering matrix. The coefficients of this matrix are computed by applying a Singular Value Decomposition (SVD) factorization over a set of training pulses. A detailed explanation of this SVD-filtering methodology, a noise filtering analysis, simulations and filtering of pulses coming from a neutron monitor were carried out to verify its feasibility.

A Simplified Threshold Dynamics Algorithm for Isotropic Surface Energies

We present a simplified version of the threshold dynamics algorithm given in Esedoḡlu and Otto (Commun Pure Appl Math 68(5):808–864, 2015). The new version still allows specifying $$N\atopwithdelims ()2$$ possibly distinct surface tensions and $$N\atopwithdelims ()2$$ possibly distinct mobilities for a network with N phases, but achieves this level of generality without the use of retardation functions. Instead, it employs linear combinations of Gaussians in the convolution step of the algorithm. Convolutions with only two distinct Gaussians is enough for the entire network, maintaining the efficiency of the original thresholding scheme. We discuss stability and convergence of the new algorithm, including some counterexamples in which convergence fails. The apparently convergent cases include unequal surface tensions given by the Read and Shockley model and its three dimensional extensions, along with equal mobilities, that are a very common choice in computational materials science.

Image salient object detection with refined deep features via convolution neural network

Recent advances in saliency detection have used deep learning to obtain high-level features to detect salient regions. These advances have demonstrated superior results over previous works that use handcrafted low-level features for saliency detection. We propose a convolutional neural network (CNN) model to learn high-level features for saliency detection. Compared to other methods, our method presents two merits. First, when performing features extraction, apart from the convolution and pooling step in our method, we add restricted Boltzmann machine into the CNN framework to obtain more accurate features in intermediate step. Second, in order to avoid manual annotation data, we add deep belief network classifier at the end of this model to classify salient and nonsalient regions. Quantitative and qualitative experiments on three benchmark datasets demonstrate that our method performs favorably against the state-of-the-art methods.

In vitro–in vivo correlations for nicotine transdermal delivery systems evaluated by both in vitro skin permeation (IVPT) and in vivo serum pharmacokinetics under the influence of transient heat application

The in vitro permeation test (IVPT) has been widely used to characterize the bioavailability (BA) of compounds applied on the skin. In this study, we performed IVPT studies using excised human skin (in vitro) and harmonized in vivo human serum pharmacokinetic (PK) studies to evaluate the potential in vitro–in vivo correlation (IVIVC) of nicotine BA from two, matrix-type, nicotine transdermal delivery systems (TDS). The study designs used for both in vitro and in vivo studies included 1h of transient heat (42±2°C) application during early or late time periods post-dosing. The goal was to evaluate whether any IVIVC observed would be evident even under conditions of heat exposure, in order to investigate further whether IVPT may have the potential to serve as a possible surrogate method to evaluate the in vivo effects of heat on the bioavailability of a drug delivered from a TDS. The study results have demonstrated that the BA of nicotine characterized by the IVPT studies correlated with and was predictive of the in vivo BA of nicotine from the respective TDS, evaluated under the matched study designs and conditions. The comparisons of single parameters such as steady-state concentration, heat-induced increase in partial AUCs and post-treatment residual content of nicotine in TDS from the in vitro and in vivo data sets showed no significant differences (p≥0.05). In addition, a good point-to-point IVIVC (Level A correlation) for the entire study duration was achieved by predicting in vivo concentrations of nicotine using two approaches: Approach I requiring only an in vitro data set and Approach II involving deconvolution and convolution steps. The results of our work suggest that a well designed IVPT study with adequate controls can be a useful tool to evaluate the relative effects of heat on the BA of nicotine from TDS with different formulations.

Deep Learning of Graphs with Ngram Convolutional Neural Networks

Convolutional Neural Network (CNN) has gained attractions in image analytics and speech recognition in recent years. However, employing CNN for classification of graphs remains to be challenging. This paper presents the Ngram graph-block based convolutional neural network model for classification of graphs. Our Ngram deep learning framework consists of three novel components. First, we introduce the concept of $n$ -gram block to transform each raw graph object into a sequence of $n$ -gram blocks connected through overlapping regions. Second, we introduce a diagonal convolution step to extract local patterns and connectivity features hidden in these $n$ -gram blocks by performing $n$ -gram normalization. Finally, we develop deeper global patterns based on the local patterns and the ways that they respond to overlapping regions by building a $n$ -gram deep learning model using convolutional neural network. We evaluate the effectiveness of our approach by comparing it with the existing state of art methods using five real graph repositories from bioinformatics and social networks domains. Our results show that the Ngram approach outperforms existing methods with high accuracy and comparable performance.

Artifacts correction method for fan-beam CT with projections asymmetrically truncated on both sides

In fan-beam CT scanning, the scanned field of view is determined by the fan-beam angle and the effective length of a linear array detector. Therefore, when an object extends outside the scanned field of view, CT projection data acquired by the detector will not be complete and is truncated abruptly at the projection boundaries. Furthermore, due to mechanical misalignment of X-ray source, object and the linear array detector, the length of the truncated data on the left side is not equal to its length on the right side. This asymmetrical truncation on both sides will bring out dual bright-band artifacts in the reconstructed images. The purpose of this paper is to develop a method of eliminating these artifacts by projection extension techniques. We first extend the truncated projection unilaterally by using geometrical symmetry property of the fan-beam scanning. Through this method, the projection center of rotation is adjusted to the center of a virtual sinogram and the region of completely scanned field of view is enlarged. The projection values in the extended region equal their equivalent points in the raw sinogram. Thus, values of the supplemented projection data are not approximate and would not decrease the reconstruction accuracy. Next, “mirror extension” technique is preformed to avoid data discontinuity at the edges of a virtual sinogram. Extra data are supplemented to make the projection data on both boundaries attenuate to zero smoothly. Experimental results demonstrated that the dual bright-band artifacts were well eliminated and the correction method could be implemented within the convolution step of a filtered back-projection reconstruction with negligible computational expense.