Dilated Convolutional Neural Network Research Articles

SPATIOTEMPORAL GRAPH CONVOLUTIONAL NEURAL NETWORKS FOR METRO FLOW PREDICTION

Abstract. Forecasting urban metro flow accurately plays an important role for station management and passenger safety. Owing to the limitations of non-linearity and complexity of traffic flow data, traditional methods cannot satisfy the requirements of effectively capturing spatiotemporal dependencies at the metro network level, which makes it difficult to demonstrate high performance. In this paper, a novel deep learning method is proposed based on Graph Neural Networks (GNN), named STGCN-Metro (SpatioTemporal Graph Convolutional Network based on Metro network), to forecast the short-term inflow and outflow volumes of metro passengers. The proposed model is composed of two spatiotemporal convolutional blocks, which is integrated with the Dilated Convolutional Neural Network (DCNN) and Cluster-Graph Convolutional Network (Cluster-GCN). The DCNN is employed with different dilation rates to capture temporal dependence in larger receptive field. In addition, compare with GCN, the Cluster-GCN is applied the graph clustering algorithms to reduce computational resources considering spatial heterogeneity. A real-world dataset collected in Shanghai metro stations is conducted for validation, and the results demonstrate that the proposed model achieves higher performance, outperforming some well-known baseline models.

An attention enhanced dilated CNN approach for cross-axis industrial robotics fault diagnosis

An industrial robot is a complex mechatronics system, whose failure is hard to diagnose based on monitoring data. Previous studies have reported various methods with deep network models to improve the accuracy of fault diagnosis, which can get an accurate prediction model when the amount of data sample is sufficient. However, the failure data is hard to obtain, which leads to the few-shot issue and the bad generalization ability of the model. Therefore, this paper proposes an attention enhanced dilated convolutional neural network (D-CNN) approach for the cross-axis industrial robotics fault diagnosis method. Firstly, key feature extraction and sliding window are adopted to pre-process the monitoring data of industrial robots before D-CNN is introduced to extract data features. And self-attention is used to enhance feature attention capability. Finally, the pre-trained model is used for transfer learning, and a small number of the dataset from another axis of the multi-axis industrial robot are used for fine-tuning experiments. The experimental results show that the proposed method can reach satisfactory fault diagnosis accuracy in both the source domain and target domain.

Chinese Clinical Named Entity Recognition with ALBERT and MHA Mechanism.

Traditional clinical named entity recognition methods fail to balance the effectiveness of feature extraction of unstructured text and the complexity of neural network models. We propose a model based on ALBERT and a multihead attention (MHA) mechanism to solve this problem. Structurally, the model first obtains character-level word embeddings through the ALBERT pretraining language model, then inputs the word embeddings into the iterated dilated convolutional neural network model to quickly extract global semantic information, and decodes the predicted labels through conditional random fields to obtain the optimal label sequence. Also, we apply the MHA mechanism to capture intercharacter dependencies from multiple aspects. Furthermore, we use the RAdam optimizer to boost the convergence speed and improve the generalization ability of our model. Experimental results show that our model achieves an F1 score of 85.63% on the CCKS-2019 dataset—an increase of 4.36% compared to the baseline model.

Multi-modal fusion of satellite and street-view images for urban village classification based on a dual-branch deep neural network

• A novel dual-branch deep neural network is proposed for urban village classification. • Multi-modal satellite and street-view images are integrated to increase the performance. • The proposed model yields a high accuracy of 92.61% in Jing-Jin-Ji region of China. • The dataset, Satellite & Street-view images for Urban Village classification (S 2 UV), has been released publicly. With the rapid urbanization process in China, numerous urban villages have been appeared, which are surrounded by the newly-built urban blocks. Due to the high population density, poor hygiene, chaotic waste discharge, and inadequate public facilities, urban villages have many negative impacts on both urban environment and management. The objective of this study is to propose a dual-branch deep learning model for multi-modal satellite and street-view data fusion to detect urban villages in Beijing, Tianjin and Shijiazhuang, which are the core cities of Jing-Jin-Ji region of China. Specifically, the proposed model consists of a satellite branch, a street-view branch and a gated-fusion module. As for the satellite branch, a Trans-MDCNN (multi-scale dilated convolutional neural network) is proposed to learn multi-level local features and global contextual features from high resolution satellite imagery, while for the street-view branch, an MVRAN (multi-view recurrent attention network) is constructed to learn and fuse multi-angle features from street-view images. A gated-fusion module is designed to aggregate the important features of the dual-branches. Experimental results indicate that the proposed model has achieved good performance with an overall accuracy (OA) of 92.61%. Ablation study shows that compared with satellite data alone, the integration of street-view images could increase the OA by about 2%. Besides, 1-D feature fusion outperforms its 2-D counterpart and the classic feature concatenation method. The proposed model also yields a better performance than other deep learning models. Finally, the dataset of this study, S 2 UV (Satellite & Street-view images for Urban Village classification), is publicly available: https://doi.org/10.11922/sciencedb.01410.

A Robot Pose Estimation Optimized Visual SLAM Algorithm Based on CO-HDC Instance Segmentation Network for Dynamic Scenes

In order to improve the accuracy of visual SLAM algorithms in a dynamic scene, instance segmentation is widely used to eliminate dynamic feature points. However, the existing segmentation technology has low accuracy, especially for the contour of the object, and the amount of calculation of instance segmentation is large, limiting the speed of visual SLAM based on instance segmentation. Therefore, this paper proposes a contour optimization hybrid dilated convolutional neural network (CO-HDC) algorithm, which can perform a lightweight calculation on the basis of improving the accuracy of contour segmentation. Firstly, a hybrid dilated convolutional neural network (HDC) is used to increase the receptive field, which is defined as the size of the region in the input that produces the feature. Secondly, the contour quality evaluation (CQE) algorithm is proposed to enhance the contour, retaining the highest quality contour and solving the problem of distinguishing dynamic feature points from static feature points at the contour. Finally, in order to match the mapping speed of visual SLAM, the Beetle Antennae Search Douglas–Peucker (BAS-DP) algorithm is proposed to lighten the contour extraction. The experimental results have demonstrated that the proposed visual SLAM based on the CO-HDC algorithm performs well in the field of pose estimation and map construction on the TUM dataset. Compared with ORB-SLAM2, the Root Mean Squared Error (Rmse) of the proposed method in absolute trajectory error is about 30 times smaller and is only 0.02 m.

Deep learning characterization of surface defects in the selective laser melting process

Surface defects in the selective laser melting (SLM) process adversely affect the surface quality of additive manufacturing workpieces. Although some studies have been conducted to classify or detect defects, the investigation of the distribution of defects has received relatively little attention. Currently, there are many studies which focus on the counting of objects and they typically adopt a backbone convolutional neural network to obtain an initial feature map, which contains more semantic information but loses some geometric details. In this paper, the distribution of surface defects and defect count estimation of additive manufactured components are first studied as well as presentation of a developed deep learning characterization method (DLCM) based on a detail-aware dilated convolutional neural network (DDCNN) incorporated with a fine details feature map extractor designed to obtain final fine semantic features. It features two dilated convolutional layer combination blocks (DCBs), which are proposed to fuse low-level features with semantic features. Data are acquired from the surface of a workpiece fabricated by SLM. A series of experiments have been conducted to validate the performance of the DLCM. Compared with the other main state-of-the-art methods, the proposed DLCM yields better results.

Pyramid dilated convolutional neural network for image denoising

Convolutional neural network has been successfully applied to image denoising. In particular, dilated convolution, which expands the network’s receptive field, has been widely used and has achieved good results in image denoising. Losing some image information, a standard network cannot effectively reconstruct tiny image details from noisy images. To solve this problem, we propose a pyramid dilated CNN, which mainly has three pyramid dilated convolutional blocks (PDCBs) and a gated fusion unit (GFU). PDCB uses dilated convolution to expand the network’s receptive field and the pyramid structure to obtain more image details. GFU fuses and enhances the feature maps from different blocks. Experiments demonstrate that the proposed method outperforms the comparative state-of-the-art denoising methods for gray and color images. In addition, the proposed method can effectively deal with real-world noisy images.

Separation and imaging of diffractions using a dilated convolutional neural network

Seismic diffractions provide high-resolution details of small-scale geologic discontinuities, and diffraction imaging can be an important contribution to the exploration of faults, fractures, and cavities. However, reflections with strong energy generally mask the existence of weak diffractions in seismic records, and separating diffractions is necessary to see the full benefit of diffraction imaging. Here, a modified convolutional neural network (CNN) is used for separating diffractions. The input wavefields are modeled as the summation of individual diffractions and reflections, the network parameters are learned from multiple input data instances by minimizing the distance between the output of the network and the diffractions, and reflections are implicitly removed. To enhance the diffraction separation performance, we have used dilated convolutions to aggregate diffraction information, then combined batch normalization and residual learning to speed up the training process. The modified CNN can realize adaptive diffraction separation and avoid setting parameters in the subsequent application process. The synthetic Sigsbee2A model demonstrates the performance of the proposed method in removing high-slope reflections and imaging small-scale scatterers. The application to the Nankai field data further illustrates the ability of the methods to extract weak diffractions beneath the interference of other waves, revealing potential reservoir-related geologic features.

Annual dilated convolution neural network for newbuilding ship prices forecasting

Annual dilated convolution neural network for newbuilding ship prices forecasting

A data-driven method for multi-step-ahead prediction and long-term prognostics of proton exchange membrane fuel cell

Fuel cell technology has been rapidly developed in the last decade owing to its clean characteristic and high efficiency. Proton exchange membrane fuel cells (PEMFCs) are increasingly used in transportation applications and small stationary applications; however, the cost and the unsatisfying durability of the PEMFC stack have limited their successful commercialization and market penetration. In recent years, thanks to the availability and the quality of emerging data of PEMFCs, digitization is happening to offer possibilities to increase the productivity and the flexibility in fuel cell applications. Therefore, it is crucial to clarify the potential of digitization measures, how and where they can be applied, and their benefits. This paper focuses on the degradation performance of the PEMFC stacks and develops a data-driven intelligent method to predict both the short-term and long-term degradation. The dilated convolutional neural network is for the first time applied for predicting the time-dependent fuel cell performance and is proved to be more efficient than other recurrent networks. To deal with the long-term performance uncertainty, a conditional neural network is proposed. Results have shown that the proposed method can predict not only the degradation tendency, but also contain the degradation behaviour dynamics.

Low-dose CT image denoising using deep convolutional neural networks with extended receptive fields

How to reduce radiation dose while preserving the image quality as when using standard dose is an important topic in the computed tomography (CT) imaging domain because the quality of low-dose CT (LDCT) images is often strongly affected by noise and artifacts. Recently, there has been considerable interest in using deep learning as a post-processing step to improve the quality of reconstructed LDCT images. This paper provides, first, an overview of learning-based LDCT image denoising methods from patch-based early learning methods to state-of-the-art CNN-based ones and, then, a novel CNN-based method is presented. In the proposed method, preprocessing and post-processing techniques are integrated into a dilated convolutional neural network to extend receptive fields. Hence, large distance pixels in input images will participate in enriching feature maps of the learned model, leading to effective denoising. Experimental results showed that the proposed method is light, while its denoising effectiveness is competitive with well-known CNN-based models.

Aircraft Landing Gear Retraction/Extension System Fault Diagnosis with 1-D Dilated Convolutional Neural Network.

The faults of the landing gear retraction/extension(R/E) system can result in the deterioration of an aircraft’s maneuvering conditions; how to identify the faults of the landing gear R/E system has become a key issue for ensuring aircraft take-off and landing safety. In this paper, we aim to solve this problem by proposing the 1-D dilated convolutional neural network (1-DDCNN). Aiming at developing the limited feature information extraction and inaccurate diagnosis of the traditional 1-DCNN with a single feature, the 1-DDCNN selects multiple feature parameters to realize feature integration. The performance of the 1-DDCNN in feature extraction is explored. Importantly, using padding dilated convolution to multiply the receptive field of the convolution kernel, the 1-DDCNN can completely retain the feature information in the original signal. Experimental results demonstrated that the proposed method has high accuracy and robustness, which provides a novel idea for feature extraction and fault diagnosis of the landing gear R/E system.

Automated Segmentation of Mass Regions in DBT Images Using a Dilated DCNN Approach.

To overcome the limitations of conventional breast screening methods based on digital mammography, a quasi-3D imaging technique, digital breast tomosynthesis (DBT) has been developed in the field of breast cancer screening in recent years. In this work, a computer-aided architecture for mass regions segmentation in DBT images using a dilated deep convolutional neural network (DCNN) is developed. First, to improve the low contrast of breast tumour candidate regions and depress the background tissue noise in the DBT image effectively, the constraint matrix is established after top-hat transformation and multiplied with the DBT image. Second, input image patches are generated, and the data augmentation technique is performed to create the training data set for training a dilated DCNN architecture. Then the mass regions in DBT images are preliminarily segmented; each pixel is divided into two different kinds of labels. Finally, the postprocessing procedure removes all false-positives regions with less than 50 voxels. The final segmentation results are obtained by smoothing the boundaries of the mass regions with a median filter. The testing accuracy (ACC), sensitivity (SEN), and the area under the receiver operating curve (AUC) are adopted as the evaluation metrics, and the ACC, SEN, as well as AUC are 86.3%, 85.6%, and 0.852 for segmenting the mass regions in DBT images on the entire data set, respectively. The experimental results indicate that our proposed approach achieves promising results compared with other classical CAD-based frameworks.

Directionally separable dilated CNN with hierarchical attention feature fusion for hyperspectral image classification

ABSTRACT In recent years, the convolutional neural network (CNN) plays a vital role in hyperspectral image classification and performs more competitively than many other methods. However, in order to pursue better performance, most of existing CNN-based methods just simply stack rather deep convolutional layers. Although they improve the classification accuracy to a certain extent, they result in plenty of network parameters. In this paper, a light-weighted directionally separable dilated CNN with hierarchical attention feature fusion (DSD-HAFF) is proposed to solve these problems. First, two global dense dilated CNN branches that focus on two spatial directions separately are constructed to extract and reuse spatial information as much as possible. Second, a hierarchical attention feature fusion branch that consists of several coordinate attention blocks (CABs) is constructed. Hierarchical features from two directionally separable dilated CNN branches are adopted as inputs of CABs. In this way, the structure can not only fully incorporate hierarchical features, but also significantly reduce the network parameters. Meanwhile, the hierarchical attention feature fusion branch incorporates features from high-level to low-level in the kernel-number pyramid strategy. Experimental results on three popular benchmark datasets demonstrate that the DSD-HAFF achieves better performance and has a much smaller number of network parameters than the other state-of-the-art methods.

Obstacle type recognition in visual images via dilated convolutional neural network for unmanned surface vehicles

AbstractRecognition of obstacle type based on visual sensors is important for navigation by unmanned surface vehicles (USV), including path planning, obstacle avoidance, and reactive control. Conventional detection techniques may fail to distinguish obstacles that are similar in visual appearance in a cluttered environment. This work proposes a novel obstacle type recognition approach that combines a dilated operator with the deep-level features map of ResNet50 for autonomous navigation. First, visual images are collected and annotated from various different scenarios for USV test navigation. Second, the deep learning model, based on a dilated convolutional neural network, is set and trained. Dilated convolution allows the whole network to learn deep features with increased receptive field and further improves the performance of obstacle type recognition. Third, a series of evaluation parameters are utilised to evaluate the obtained model, such as the mean average precision (mAP), missing rate and detection speed. Finally, some experiments are designed to verify the accuracy of the proposed approach using visual images in a cluttered environment. Experimental results demonstrate that the dilated convolutional neural network obtains better recognition performance than the other methods, with an mAP of 88%.

D-GF-CNN Algorithm for Modulation Recognition

This paper presents a novel modulation recognition algorithm based on dilated convolutional neural network with a new defined GF regularization function, named as D-GF-CNN algorithm. Firstly, an asynchronous delay sampling (ADS) technique is introduced. Via the defined ADS, the received signal is converted into an asynchronous delay histogram (ADH). The ADH of different modulation signals has distinct characteristics, which provides great convenience for the neural network to identify the modulation mode. Then, the pixel point matrix of the ADH is convolved with the dilated convolution kernel of the convolutional neural network, and the automatic extraction of signal features is completed so that the manual feature extraction processing can be effectively avoided. Finally, a novel GF regularization function is given, which can improve the constraint ability of the loss function on the weight and effectively weaken the influence of network over-fitting on the modulation recognition accuracy. Theoretical analysis and simulation experiments show that the proposed algorithm provides several advantages, for example: (1) automatically extract features; (2) effectively prevent network over-fitting; (3) significantly improve recognition accuracy in the lower SNR scenarios.

New Hybrid Deep Learning Approach Using BiGRU-BiLSTM and Multilayered Dilated CNN to Detect Arrhythmia

Deep learning methods have shown early progress in analyzing complicated ECG signals, especially in heartbeat classification and arrhythmia detection. There is still a great deal of space for further research on this area before reaching a definite decision. This study introduced a novel hybrid framework based on a bidirectional recurrent neural network (BiRNN) with a multilayered dilated convolution neural network (CNN) for arrhythmia classification. Initially, the raw ECG signals are filtered using Chebyshev Type II method and the Daubechies wavelet method is used to solve fractal problems and signal discontinuities. Then, a synthetic signal is generated using a generative adversarial network (GAN) to handle imbalanced signal classes. The proposed Bidirectional RNN with Dilated CNN (BRDC) architecture takes advantage of multilayered dilated CNN and bidirectional RNN units (bidirectional gated recurrent Units, BiGRU, bidirectional long short-term memory, BiLSTM) to generate fusion features and then, fusion features are classified in the fully connected layer. The PhysioNet 2017 challenge (MIT-BIH) dataset is used to train and validate the proposed approach. By combining fusion features with dilated CNN, the proposed approach outperforms the existing model for arrhythmia detection with 99.90 % accuracy, 98.41 % F1-score, 97.96 % precision, and 99.90 % recall. Overall, our hybrid BRDC model provides a cost-effective ECG signal reduction and high-performance automated recognition technique to identify arrhythmia. In the future, an automatic and cloud-based system with more arrhythmia data variance to test the model’s robustness will be given the highest priority.

A Deep Learning Approach for Crowd Counting in Highly Congested Scene

With the rapid progress of deep convolutional neural networks, several applications of crowd counting have been proposed and explored in the literature. In congested scene monitoring, a variety of crowd density estimating approaches has been developed. The understanding of highly congested scenes for crowd counting during Muslim gatherings of Hajj and Umrah is a challenging task, as a large number of individuals stand nearby and, it is hard for detection techniques to recognize them, as the crowd can vary from low density to high density. To deal with such highly congested scenes, we have proposed the Congested Scene Crowd Counting Network (CSCC-Net) using VGG-16 as a core network with its first ten layers due to its strong and robust transfer learning rate. A hole dilated convolutional neural network is used at the back end to widen the relevant field to extract a large range of information from the image without losing its original resolution. The dilated convolution neural network is mainly chosen to expand the kernel size without changing other parameters. Moreover, several loss functions have been applied to strengthen the evaluation accuracy of the model. Finally, the entire experiments have been evaluated using prominent data sets namely, ShanghaiTech parts A, B, UCF_CC_50, and UCF_QNRF. Our model has achieved remarkable results i.e., 68.0 and 9.0 MAE on ShanghaiTech parts A, B, 199.1 MAE on UCF_CC_50, and 99.8 on UCF_QNRF data sets respectively.

MSRD-Unet: Multiscale Residual Dilated U-Net for Medical Image Segmentation

Semantic segmentation is an exciting research topic in medical image analysis because it aims to detect objects in medical images. In recent years, approaches based on deep learning have shown a more reliable performance than traditional approaches in medical image segmentation. The U-Net network is one of the most successful end-to-end convolutional neural networks (CNNs) presented for medical image segmentation. This paper proposes a multiscale Residual Dilated convolution neural network (MSRD-UNet) based on U-Net. MSRD-UNet replaced the traditional convolution block with a novel deeper block that fuses multi-layer features using dilated and residual convolution. In addition, the squeeze and execution attention mechanism (SE) and the skip connections are redesigned to give a more reliable fusion of features. MSRD-UNet allows aggregation of contextual information, and the network goes without needing to increase the number of parameters or required floating-point operations (FLOPS). The proposed model was evaluated on three multimodal datasets: polyp, skin lesion, and nuclei segmentation. The obtained results proved that the MSDR-Unet model outperforms several state-of-the-art U-Net-based methods.

Comparison of various deep learning inpainting methods in smart colposcopy images

The specular reflection appears as white area on the cervical images that covers certain region of the images. It affects the quality of the cervical images causing difficulty for the physician to analyse the smart colposcopy images. In this paper, specular reflection is detected and removed from the cervical images, and these removed pixels are refilled using the inpainting methods. To fill the removed pixels, the deep learning inpainting algorithms like partial convolutional neural network, generative multicolumn convolutional neural network, and the dilated convolutional neural network to were used to get the complete and enhanced cervical images. The enhanced images are considered for lesion identification using the Bayes classifier. Based on the analysis, the partial convolution inpainting method gives higher quality with the PSNR value of 48.25 dB and SSIM value of 0.984. The enhanced images using the partial inpainting method identify the neoplasm with an accuracy of 98%.