Dynamic Convolutional Neural Network Research Articles

Accuracy and efficiency of automatic tooth segmentation in digital dental models using deep learning.

This study evaluates the accuracy and efficiency of automatic tooth segmentation in digital dental models using deep learning. We developed a dynamic graph convolutional neural network (DGCNN)-based algorithm for automatic tooth segmentation and classification using 516 digital dental models. We segmented 30 digital dental models using three methods for comparison: (1) automatic tooth segmentation (AS) using the DGCNN-based algorithm from LaonSetup software, (2) landmark-based tooth segmentation (LS) using OrthoAnalyzer software, and (3) tooth designation and segmentation (DS) using Autolign software. We evaluated the segmentation success rate, mesiodistal (MD) width, clinical crown height (CCH), and segmentation time. For the AS, LS, and DS, the tooth segmentation success rates were 97.26%, 97.14%, and 87.86%, respectively (p < 0.001, post-hoc; AS, LS > DS), the means of MD widths were 8.51, 8.28, and 8.63 mm, respectively (p < 0.001, post hoc; DS > AS > LS), the means of CCHs were 7.58, 7.65, and 7.52 mm, respectively (p < 0.001, post-hoc; LS > DS, AS), and the means of segmentation times were 57.73, 424.17, and 150.73 s, respectively (p < 0.001, post-hoc; AS < DS < LS). Automatic tooth segmentation of a digital dental model using deep learning showed high segmentation success rate, accuracy, and efficiency; thus, it can be used for orthodontic diagnosis and appliance fabrication.

Dynamic convolutional neural network based e‐waste management and optimized collection planning

AbstractElectronic waste, also known as e‐waste, refers to electrical or electronic devices that are discarded from households and workplaces. These used e‐wastes are meant to be renovated, reused, recycled, or disposed of, and the processing of these wastes often causes disease and harms the environment. As a result, it is important to handle waste and collect it from the disposal site on a regular basis. Besides, in order to separate precious metals from discarded waste, it is important to identify them by category. Therefore, this article proposes a novel method known as e‐waste management by exploiting the dynamic convolutional neural network (DCNN). This enhances the classification accuracy with the aid of exactly mapping the features of the images. Meanwhile, the collection of waste can be optimized in order to reduce the distance and time. The e‐wastes in the smart garbage bin are frequently monitored by smartphone applications to collect the waste on time. Moreover, it also significantly reduces the training error, classification error, localization error, and validation error on the test images. The experimental depicts that the proposed method hones up the classification accuracy to the great extent.

Nonlinear Dynamic Soft Sensor Development with a Supervised Hybrid CNN-LSTM Network for Industrial Processes.

A soft sensor is a key component when a real-time measurement is unavailable for industrial processes. Recently, soft sensor models based on deep-learning techniques have been successfully applied to complex industrial processes with nonlinear and dynamic characteristics. However, the conventional deep-learning-based methods cannot guarantee that the quality-relevant features are included in the hidden states when the modeling samples are limited. To address this issue, a supervised hybrid network based on a dynamic convolutional neural network (CNN) and a long short-term memory (LSTM) network is designed by constructing multilayer dynamic CNN-LSTM with improved structures. In each time instant, data augmentation is implemented by dynamic expansion of the original samples. Moreover, multiple supervised hidden units are trained by adding quality variables as part of the layer input to acquire a better quality-related feature learning performance. The effectiveness of the proposed soft senor development is validated through two industrial applications, including a penicillin fermentation process and a debutanizer column.

Fast Prediction of Quasi-Periodic Array Using Dynamical Graph Convolutional Neural Networks

In this letter, a novel approach based on dynamic graph convolutional neural networks (DGCNNs) is proposed to analyze the quasi-periodic array consisting of the cells with different geometric parameters. Each vertex node in a graph represents a cell in quasi-periodic array and the features of each vertex node are determined by the geometric parameters of the corresponding cell. The connections in graph are initialized and dynamically updated to learn the mutual coupling among cells. DGCNNs are constructed and trained such that the neural networks can obtain the relationships among cells. The trained DGCNNs can predict the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">S</i> -parameters of a quasi-periodic array and the numerical experiments demonstrate the accuracy and efficiency of our method.

Fusion of a Static and Dynamic Convolutional Neural Network for Multiview 3D Point Cloud Classification

Three-dimensional (3D) point cloud classification methods based on deep learning have good classification performance; however, they adapt poorly to diverse datasets and their classification accuracy must be improved. Therefore, FSDCNet, a neural network model based on the fusion of static and dynamic convolution, is proposed and applied for multiview 3D point cloud classification in this paper. FSDCNet devises a view selection method with fixed and random viewpoints, which effectively avoids the overfitting caused by the traditional fixed viewpoint. A local feature extraction operator of dynamic and static convolution adaptive weight fusion was designed to improve the model’s adaptability to different types of datasets. To address the problems of large parameters and high computational complexity associated with the current methods of dynamic convolution, a lightweight and adaptive dynamic convolution operator was developed. In addition, FSDCNet builds a global attention pooling, integrating the most crucial information on different view features to the greatest extent. Due to these characteristics, FSDCNet is more adaptable, can extract more fine-grained detailed information, and can improve the classification accuracy of point cloud data. The proposed method was applied to the ModelNet40 and Sydney Urban Objects datasets. In these experiments, FSDCNet outperformed its counterparts, achieving state-of-the-art point cloud classification accuracy. For the ModelNet40 dataset, the overall accuracy (OA) and average accuracy (AA) of FSDCNet in a single view reached 93.8% and 91.2%, respectively, which were superior to those values for many other methods using 6 and 12 views. FSDCNet obtained the best results for 6 and 12 views, achieving 94.6%, 93.3%, 95.3%, and 93.6% in OA and AA metrics, respectively. For the Sydney Urban Objects dataset, FSDCNet achieved an OA and F1 score of 81.2% and 80.1% in a single view, respectively, which were higher than most of the compared methods. In 6 and 12 views, FSDCNet reached an OA of 85.3% and 83.6% and an F1 score of 85.5% and 83.7%, respectively.

SC-Dynamic R-CNN: A Self-Calibrated Dynamic R-CNN Model for Lung Cancer Lesion Detection.

Lung cancer has complex biological characteristics and a high degree of malignancy. It has always been the number one “killer” in cancer, threatening human life and health. The diagnosis and early treatment of lung cancer still require improvement and further development. With high morbidity and mortality, there is an urgent need for an accurate diagnosis method. However, the existing computer-aided detection system has a complicated process and low detection accuracy. To solve this problem, this paper proposed a two-stage detection method based on the dynamic region-based convolutional neural network (Dynamic R-CNN). We divide lung cancer into squamous cell carcinoma, adenocarcinoma, and small cell carcinoma. By adding the self-calibrated convolution module into the feature network, we extracted more abundant lung cancer features and proposed a new regression loss function to further improve the detection performance of lung cancer. After experimental verification, the mAP (mean average precision) of the model can reach 88.1% on the lung cancer dataset and it performed particularly well with a high IoU (intersection over union) threshold. This method has a good performance in the detection of lung cancer and can improve the efficiency of doctors' diagnoses. It can avoid false detection and miss detection to a certain extent.

Malicious behavior detection in cloud using self‐optimized dynamic kernel convolutional neural network

AbstractCloud computing is one example of a technological revolution that has drastically altered the way traditional commercial operations are conducted. Because cloud platforms are fundamentally revolutionizing the area of computing, they must provide safe transmission and storage of user data by assuring the confidentiality, integrity, and availability of the data. As a result, establishing a secure cloud computing framework is crucial, and security cannot only rely on the user's credentials being verified. To overcome this problem, this article presents a multi‐population genetic algorithm optimized dynamic kernel convolutional neural network to analyze the user's behavior to identify their malicious intent. This article describes a two‐stage malicious detection and prediction scheme. The proposed system's efficiency is measured by its ability to discriminate between nine different types of attacks in the UNSW‐NB15 dataset (abnormal behavior exhibited by users). The average true positive rate, false positive rate, precision, and F‐measure for the nine attack classes were 99.22%, 99.11%, 99.11%, and 99.22%, respectively, when evaluated using the UNSW‐NB15 dataset. The experimental results demonstrate that this technique is effective in detecting malicious behaviors in the cloud environment with higher accuracy.

Building a full-waveform inversion starting model from wells with dynamic time warping and convolutional neural networks

Seismic full-waveform inversion (FWI) needs a feasible starting model because otherwise it might converge to a local minimum and the inversion result might suffer from detrimental artifacts. We have built a feasible starting model from wells by applying dynamic time warping (DTW) localized rewarp and convolutional neural network (CNN) methods alternatively. We use the DTW localized rewarp method to extrapolate the velocities at well locations to the nonwell locations in the model space. Rewarping is conducted based on the local structural coherence, which is extracted from a migration image of an initial infeasible model. The extraction uses the DTW method. The purpose of velocity extrapolation is to provide sufficient training samples to train a CNN, which maps local spatial features on the migration image into the velocity quantities of each layer. Furthermore, we design an interactive workflow to reject inaccurate network predictions and to improve CNN prediction accuracy by incorporating the Monte Carlo dropout method. We have determined that our method is robust against kinematic incorrectness in the migration velocity model, and it is capable of producing a feasible FWI starting model.

Quality Judgment of 3D Face Point Cloud Based on Feature Fusion

3D face point cloud quality judgment is an important task in 3D face recognition and 3D construction, as an auxiliary task of face recognition to predict the quality of a given image. However, projection-based quality judgment methods always ignore the value of multi-view 2D image features and 3D point cloud features. We propose a Feature Fusion Network (FFN) to address the above problems. We first preprocess the 3D point cloud obtained by the CCD camera, cut out the face area, and then use the point cloud data for multi-angle rotation and the corresponding 2D plane depth map projection as input. Secondly, Dynamic Graph Convolutional Neural Network (DGCNN) was trained for point cloud learning and ShuffleNet was trained for image learning. Then, the middle layer features of the two network modules were extracted and concat to fine-tune the whole network. Finally, three fully connected layers were used to realize the five-class classification of the 3D face point cloud (excellent, ordinary, stripe, burr, deformation). The proposed FFN achieved the classification accuracy of 83.7%, which was 5.8% higher than that of ShuffleNet and 2.2% higher than that of DGCNN. The experimental results show that concat depth map features and point cloud features can achieve the complementary effect between different features.

DGCNN Network Architecture With Densely Connected Point Pairs in Multiscale Local Regions for ALS Point Cloud Classification

Deep learning in 3-D classification tasks focuses on the designs of comprehensive local aggregation operators. Airborne laser scanning (ALS) point clouds have their own characteristics: 1) the object overlaps in the vertical direction; 2) the spatial density is uneven; and 3) the objects present large-scale variations between different categories. However, the dynamic graph convolutional neural network (DGCNN) ignores the inherent properties of ALS point clouds. This study modifies the DGCNN network and proposes a new neural network module called Webconv, which densely connects point pairs in multiscale local regions to learn contextual information. One modified cross entropy loss function with variable weight is proposed to solve the problem of uneven category distributions in ALS points. Because preprocessing such as block partition ignores the context information of the whole region, the conditional random field is used to refine the point cloud. Our approach achieves state-of-the-art performance on the dataset of the 2019 IEEE GRSS Data Fusion Contest 3-D Point Cloud Classification Challenge and can be widely used for ALS point classification.

IÇ ORTAMLARDA KAPILARIN TESPİTİ İÇİN DERİN ÖĞRENME TEKNİKLERİNİN KARŞILAŞTIRILMASI

In indoor environments, the detection of doors (open, semi-opened, and closed) is a crucial task for a variety of fields such as robotics, computer vision, and architecture. The studies that are addressed the door detection problem can be divided into three major categories: 1) closed doors via visual data, 2) open doors via range data, and 3) open, semi-opened, and closed doors via point cloud data. Although some successful studies have been proposed being detected doors via visual and range data under specific circumstances, in this study, we exploited point cloud data due to its ability to describe the 3D characteristic of scenes. The main contribution of this study is two-fold. Firstly, we mainly intended to discover the potential of point-based deep learning architectures such as PointNet, PointNet++, Dynamic Graph Convolutional Neural Network (DGCNN), PointCNN, and Point2Sequence, in contrast to previous studies that generally defined a set of rules depending on the type of door and characteristics of the data, Secondly, the OGUROB DOORS dataset is constructed, which contains point cloud data captured in the GAZEBO simulation environment in different robot positions and orientations. We used precision, recall, and F1-score metrics to analyze the merit and demerit aspects of these architectures. Also, some visual results were given to describe the characteristics of these architectures. The test results showed that all architectures are capable of classifying open, semi-opened, and closed doors over 98% accuracy.

A Comparative Study of Point-Based Deep Learning Techniques for Semantic Classification in Search and Rescue Arenas

Post-disaster indoor environments, which occur after calamities such as floods, fires, and poisonous material spread, could include serious risks for search and rescue teams. For example, the building's structural integrity could be corrupted, and some harmful substances for humans and animals could exist. Exploiting robots could prevent search and rescue teams from these risks. Nevertheless, robots need to possess advanced techniques to produce high-level information from raw sensor data in these harsh environments. This study aims to explore the positive and negative aspects of point-based deep learning architectures for the semantic classification of ramps in search and rescue test arenas, which are proposed by the National Institute of Standards and Technology (NIST). Also, we take into account walls and terrain since they can provide crucial information for robots. In this study, we opted to utilize point cloud data that is robust against lousy illumination conditions, which is frequently encountered in post-disaster environments. We used the ESOGU RAMPS dataset that contains point cloud data captured from a simulated environment similar to NIST search and rescue arenas. We selected PointNet, PointNet++, Dynamic Graph Convolutional Neural Network (DGCNN), PointCNN, Point2Sequence, PointConv, and Shellnet point-based deep learning architectures to analyze their performance for semantic classification of ramps, walls, and terrain. The test results indicate that accuracy of semantic classification is over 90% for all architectures.

Fusion Models for Generalized Classification of Multi-Axial Human Movement: Validation in Sport Performance.

We introduce a set of input models for fusing information from ensembles of wearable sensors supporting human performance and telemedicine. Veracity is demonstrated in action classification related to sport, specifically strikes in boxing and taekwondo. Four input models, formulated to be compatible with a broad range of classifiers, are introduced and two diverse classifiers, dynamic time warping (DTW) and convolutional neural networks (CNNs) are implemented in conjunction with the input models. Seven classification models fusing information at the input-level, output-level, and a combination of both are formulated. Action classification for 18 boxing punches and 24 taekwondo kicks demonstrate our fusion classifiers outperform the best DTW and CNN uni-axial classifiers. Furthermore, although DTW is ostensibly an ideal choice for human movements experiencing non-linear variations, our results demonstrate deep learning fusion classifiers outperform DTW. This is a novel finding given that CNNs are normally designed for multi-dimensional data and do not specifically compensate for non-linear variations within signal classes. The generalized formulation enables subject-specific movement classification in a feature-blind fashion with trivial computational expense for trained CNNs. A commercial boxing system, ‘Corner’, has been produced for real-world mass-market use based on this investigation providing a basis for future telemedicine translation.

A dynamic graph convolutional neural network framework reveals new insights into connectome dysfunctions in ADHD

The pathological mechanism of attention deficit hyperactivity disorder (ADHD) is incompletely specified, which leads to difficulty in precise diagnosis. Functional magnetic resonance imaging (fMRI) has emerged as a common neuroimaging technique for studying the brain functional connectome. Most existing methods that have either ignored or simply utilized graph structure, do not fully leverage the potentially important topological information which may be useful in characterizing brain disorders. There is a crucial need for designing novel and efficient approaches which can capture such information. To this end, we propose a new dynamic graph convolutional network (dGCN), which is trained with sparse brain regional connections from dynamically calculated graph features. We also develop a novel convolutional readout layer to improve graph representation. Our extensive experimental analysis demonstrates significantly improved performance of dGCN for ADHD diagnosis compared with existing machine learning and deep learning methods. Visualizations of the salient regions of interest (ROIs) and connectivity based on informative features learned by our model show that the identified functional abnormalities mainly involve brain regions in temporal pole, gyrus rectus, and cerebellar gyri from temporal lobe, frontal lobe, and cerebellum, respectively. A positive correlation was further observed between the identified connectomic abnormalities and ADHD symptom severity. The proposed dGCN model shows great promise in providing a functional network-based precision diagnosis of ADHD and is also broadly applicable to brain connectome-based study of mental disorders.

A Dynamic Convolutional Neural Network Based Shared-Bike Demand Forecasting Model

Bike-sharing systems are becoming popular and generate a large volume of trajectory data. In a bike-sharing system, users can borrow and return bikes at different stations. In particular, a bike-sharing system will be affected by weather, the time period, and other dynamic factors, which challenges the scheduling of shared bikes. In this article, a new shared-bike demand forecasting model based on dynamic convolutional neural networks, called SDF , is proposed to predict the demand of shared bikes. SDF chooses the most relevant weather features from real weather data by using the Pearson correlation coefficient and transforms them into a two-dimensional dynamic feature matrix, taking into account the states of stations from historical data. The feature information in the matrix is extracted, learned, and trained with a newly proposed dynamic convolutional neural network to predict the demand of shared bikes in a dynamical and intelligent fashion. The phase of parameter update is optimized from three aspects: the loss function, optimization algorithm, and learning rate. Then, an accurate shared-bike demand forecasting model is designed based on the basic idea of minimizing the loss value. By comparing with classical machine learning models, the weight sharing strategy employed by SDF reduces the complexity of the network. It allows a high prediction accuracy to be achieved within a relatively short period of time. Extensive experiments are conducted on real-world bike-sharing datasets to evaluate SDF. The results show that SDF significantly outperforms classical machine learning models in prediction accuracy and efficiency.

Exploiting dynamic spatio-temporal graph convolutional neural networks for citywide traffic flows prediction

The prediction of crowd flows is an important urban computing issue whose purpose is to predict the future number of incoming and outgoing people in regions. Measuring the complicated spatial–temporal dependencies with external factors, such as weather conditions and surrounding point-of-interest (POI) distribution is the most difficult aspect of predicting crowd flows movement. To overcome the above issue, this paper advises a unified dynamic deep spatio-temporal neural network model based on convolutional neural networks and long short-term memory, termed as (DHSTNet) to simultaneously predict crowd flows in every region of a city. The DHSTNet model is made up of four separate components: a recent, daily, weekly, and an external branch component. Our proposed approach simultaneously assigns various weights to different branches and integrates the four properties’ outputs to generate final predictions. Moreover, to verify the generalization and scalability of the proposed model, we apply a Graph Convolutional Network (GCN) based on Long Short Term Memory (LSTM) with the previously published model, termed as GCN-DHSTNet; to capture the spatial patterns and short-term temporal features; and to illustrate its exceptional accomplishment in predicting the traffic crowd flows. The GCN-DHSTNet model not only depicts the spatio-temporal dependencies but also reveals the influence of different time granularity, which are recent, daily, weekly periodicity and external properties, respectively. Finally, a fully connected neural network is utilized to fuse the spatio-temporal features and external properties together. Using two different real-world traffic datasets, our evaluation suggests that the proposed GCN-DHSTNet method is approximately 7.9%–27.2% and 11.2%–11.9% better than the AAtt-DHSTNet method in terms of RMSE and MAPE metrics, respectively. Furthermore, AAtt-DHSTNet outperforms other state-of-the-art methods.

3DLEB-Net: Label-Efficient Deep Learning-Based Semantic Segmentation of Building Point Clouds at LoD3 Level

In current research, fully supervised Deep Learning (DL) techniques are employed to train a segmentation network to be applied to point clouds of buildings. However, training such networks requires large amounts of fine-labeled buildings’ point-cloud data, presenting a major challenge in practice because they are difficult to obtain. Consequently, the application of fully supervised DL for semantic segmentation of buildings’ point clouds at LoD3 level is severely limited. In order to reduce the number of required annotated labels, we proposed a novel label-efficient DL network that obtains per-point semantic labels of LoD3 buildings’ point clouds with limited supervision, named 3DLEB-Net. In general, it consists of two steps. The first step (Autoencoder, AE) is composed of a Dynamic Graph Convolutional Neural Network (DGCNN) encoder and a folding-based decoder. It is designed to extract discriminative global and local features from input point clouds by faithfully reconstructing them without any label. The second step is the semantic segmentation network. By supplying a small amount of task-specific supervision, a segmentation network is proposed for semantically segmenting the encoded features acquired from the pre-trained AE. Experimentally, we evaluated our approach based on the Architectural Cultural Heritage (ArCH) dataset. Compared to the fully supervised DL methods, we found that our model achieved state-of-the-art results on the unseen scenes, with only 10% of labeled training data from fully supervised methods as input. Moreover, we conducted a series of ablation studies to show the effectiveness of the design choices of our model.

Integrating feature optimization using a dynamic convolutional neural network for chemical process supervised fault classification

Chemical processes usually exhibit complex, high-dimensional, time-varying, and non-Gaussian characteristics, and the diagnosis of faults in chemical processes is particularly important. However, many current fault diagnosis methods do not consider the temporal correlation of process data, feature selection, and feature sequence arrangement. To solve this problem, this paper presents a fault diagnosis method using a dynamic convolutional neural network, based on a genetic algorithm (GA), for optimizing a feature sequence. First, the input data are transformed into a two-dimensional matrix by adding the dimension of time characteristics. Second, the GA is used to select the features, and the sequence of the selected features is optimized. Finally, the optimized feature sequence is input into the convolutional neural network (CNN) to obtain the final diagnosis results. The Tennessee Eastman chemical process is used for experimental analysis, and the proposed model is compared with the weighted cascade forest, deep belief network (DBN), optimized DBN, long short-term memory + CNN and feature selection using random forest models. The experimental results show that the proposed model has higher diagnostic accuracy. The average diagnosis rate of 20 faults is found to be 89.72%.

Effective Classification of Melting Curve in Real-time PCR Based on Dynamic Filter-based Convolutional Neural Network

Background: The effective classification of the melting curve is conducive to measure the specificity of the amplified products and the influence of invalid data on subsequent experiments is excluded. Objective: In this paper, a convolutional neural network (CNN) classification model based on dynamic filter is proposed, which can categorize the number of peaks in the melting curve image and distinguish the pollution data represented by the noise peaks. Methods: The main advantage of the proposed model is that it adopts the filter which changes with the input and uses the dynamic filter to capture more information in the image, making the network learning more accurate. In addition, the residual module is used to extract the characteristics of the melting curve, and the pooling operation is replaced with an atrous convolution to prevent the loss of context information. Result: In order to train the proposed model, a novel melting curve dataset is created, which includes a balanced dataset and an unbalanced dataset. The proposed method uses six classification-based assessment criteria to compare with seven representative methods based on deep learning. Experimental results show that the proposed method not only markedly outperforms the other state-of-the-art methods in accuracy, but also has much less running time. Conclusion: It evidently proves that the proposed method is suitable for judging the specificity of amplification products according to the melting curve. Simultaneously, it overcomes the difficulties of manual selection with low efficiency and artificial bias.

AN EXTRACTION METHOD FOR ROOF POINT CLOUD OF ANCIENT BUILDING USING DEEP LEARNING FRAMEWORK

Abstract. Chinese ancient architecture is a valuable heritage wealth, especially for roof that reflects the construction age, structural features and cultural connotation. Point cloud data, as a flexible representation with characteristics of fast, precise, non-contact, plays a crucial role in a variety of applications for ancient architectural heritage, such as 3D fine reconstruction, HBIM, disaster monitoring etc. However, there are still many limitations in data editing tasks that need to be worked out manually, which is time-consuming, labor-intensive and error-prone. In recent years, the theoretical advance on deep learning has stimulated the development of various domains, and digital heritage is not in exception. Whenever, deep learning algorithm need to consume a huge amount of labeled date to achieve the purpose for segmentation, resulting a actuality that high labor costs also be acquired. In this paper, inspired by the architectural style similarity between mimetic model and real building, we proposed a method supported by deep learning, which aims to give a solution for the point cloud automatic extraction of roof structure. Firstly, to generate real point cloud, Baoguang Temple, unmanned Aerial Vehicle (UAV) is presented to obtain image collections that are subsequently processed by reconstruction technology. Secondly, a modified Dynamic Graph Convolutional Neural Network (DGCNN) which can learn local features with taking advantage of an edge attention convolution is trained using simulated data and additional attributes of geometric attributes. The mimetic data is sampled from 3DMAX model surface. Finally, we try to extract roof structure of ancient building from real point clouds scenes utilizing the trained model. The experimental results show that the proposed method can extract the rooftop structure from real scene of Baoguang, which illustrates not only effectiveness of approach but also a fact that the simulated source perform potential value when real point cloud datasets are scarce.