Dilated Convolutional Network Research Articles

FRD-Net: a full-resolution dilated convolution network for retinal vessel segmentation

Accurate and automated retinal vessel segmentation is essential for performing diagnosis and surgical planning of retinal diseases. However, conventional U-shaped networks often suffer from segmentation errors when dealing with fine and low-contrast blood vessels due to the loss of continuous resolution in the encoding stage and the inability to recover the lost information in the decoding stage. To address this issue, this paper introduces an effective full-resolution retinal vessel segmentation network, namely FRD-Net, which consists of two core components: the backbone network and the multi-scale feature fusion module (MFFM). The backbone network achieves horizontal and vertical expansion through the interaction mechanism of multi-resolution dilated convolutions while preserving the complete image resolution. In the backbone network, the effective application of dilated convolutions with varying dilation rates, coupled with the utilization of dilated residual modules for integrating multi-scale feature maps from adjacent stages, facilitates continuous learning of multi-scale features to enhance high-level contextual information. Moreover, MFFM further enhances segmentation by fusing deeper multi-scale features with the original image, facilitating edge detail recovery for accurate vessel segmentation. In tests on multiple classical datasets,compared to state-of-the-art segmentation algorithms, FRD-Net achieves superior performance and generalization with fewer model parameters.

Cross Stage Partial Dilated Convolution Network for License Plate Recognition

Cross Stage Partial Dilated Convolution Network for License Plate Recognition

PDED-ConvLSTM: Pyramid Dilated Deeper Encoder–Decoder Convolutional LSTM for Arctic Sea Ice Concentration Prediction

Arctic sea ice concentration plays a key role in the global ecosystem. However, accurate prediction of Arctic sea ice concentration remains a challenging task due to its inherent nonlinearity and complex spatiotemporal correlations. To address these challenges, we propose an innovative encoder–decoder pyramid dilated convolutional long short-term memory network (DED-ConvLSTM). The model is constructed based on the convolutional long short-term memory network (ConvLSTM) and, for the first time, integrates the encoder–decoder architecture of ConvLSTM (ED-ConvLSTM) with a pyramidal dilated convolution strategy. This approach aims to efficiently capture the spatiotemporal properties of the sea ice concentration and to enhance the identification of its nonlinear relationships. By applying convolutional layers with different dilation rates, the PDED-ConvLSTM model can capture spatial features at multiple scales and increase the receptive field without losing resolution. Further, the integration of the pyramid convolution module significantly enhances the model’s ability to understand complex spatiotemporal relationships, resulting in notable improvements in prediction accuracy and generalization ability. The experimental results show that the sea ice concentration distribution predicted by the PDED-ConvLSTM model is in high agreement with ground-based observations, with the residuals between the predictions and observations maintained within a range from −20% to 20%. PDED-ConvLSTM outperforms other models in terms of prediction performance, reducing the RMSE by 3.6% compared to the traditional ConvLSTM model and also performing well over a five-month prediction period. These experiments demonstrate the potential of PDED-ConvLSTM in predicting Arctic sea ice concentrations, making it a viable tool to meet the requirements for accurate prediction and provide technical support for safe and efficient operations in the Arctic region.

A convolutional neural-network-based diagnostic framework for industrial bearing

Abstract. The problem of industrial bearing health monitoring and fault diagnosis has recently been a popular research topic. Extracting sufficient features from the input raw vibration signals and mapping them to the most likely fault labels is the essence of bearing fault diagnosis. This study proposes a novel framework for bearing defect diagnostics by merging dilated residual convolutional neural networks and attention mechanisms. In this framework, multiple parallel dilated convolutional networks can automatically learn rich fault features at each scale from vibration signals. Simultaneously, the attention approach boosts fault-related features and suppresses irrelevant ones, improving fault detection performance and generalization. According to the experimental results of two different bearing datasets, the framework achieves a higher accuracy and can accurately identify various types of faults.

GDJN: Global density joined network for abnormal event detection in surveillance videos

Detecting abnormal events in surveillance involves identifying unexpected behavior through video analysis. This involves recognizing patterns or deviations from normal behavior and taking actions to mitigate potential risks. However, the distribution of data can change over time, leading to concept drift, which can make it challenging to accurately detect abnormal events. To address this issue, a new approach using a global density network (GDN) has been proposed. The GDN allows for more efficient identification of object distributions in surveillance videos, leading to improved accuracy in abnormal event detection. The proposed method combines features extracted by a backbone network with a global density joined network (GDJN), which refines density features using dilated convolutional networks. A multistage long short-term memory (LSTM) network is then used to classify abnormal events. The experimental results are conducted on two datasets, UMN and UCSD Ped2. The achieved F1 scores were 93.42 and 94.46 respectively, with corresponding AUC values of 93.5 and 94.8.

Deep multi-scale dilated convolution network for coronary artery segmentation

Automatic segmentation of coronary arteries is of great significance for the rapid and accurate detection of cardiovascular diseases. Currently, deep learning has been successfully applied in the field of coronary artery segmentation. However, the branch structure of coronary arteries is thin, and the contrast between the blood vessels and the background is relatively low, making branches difficult to identify and the false positive rate is high. In response to these challenges, we proposed a multi-scale dilated convolution and deep information extraction network based on unet, which we called 3D-MDCNET. Firstly, adaptive scale expansion convolution modules are designed based on different layers. The advantage is to expand the receptive field and extract a larger range of information, thereby improving the continuity of small branches, while avoiding excessive computational costs. Secondly, the information from different layers of the decoder in Unet is fused with the first-stage segmentation results. Using multi-scale information fusion to enhance information expression, and applying the depth information extraction module to refine the results, effectively reducing the false positive rate. Finally, we introduce deep supervision as a mechanism to mitigate vanishing and exploding gradient problems caused by deep models. By conducting experiments on a benchmark dataset of coronary artery segmentation, our method indeed improves the continuity of small branch segmentation results while reducing the false positive rate. The proposed method has good segmentation performance and generalization ability, outperforming multiple state-of-the-art algorithms on various indicators.

Modified DeeplabV3+ with multi-level context attention mechanism for colonoscopy polyp segmentation

The development of automated methods for analyzing medical images of colon cancer is one of the main research fields. A colonoscopy is a medical treatment that enables a doctor to look for any abnormalities like polyps, cancer, or inflammatory tissue inside the colon and rectum. It falls under the category of gastrointestinal illnesses, and it claims the lives of almost two million people worldwide. Video endoscopy is an advanced medical imaging approach to diagnose gastrointestinal disorders such as inflammatory bowel, ulcerative colitis, esophagitis, and polyps. Medical video endoscopy generates several images, which must be reviewed by specialists. The difficulty of manual diagnosis has sparked research towards computer-aided techniques that can quickly and reliably diagnose all generated images. The proposed methodology establishes a framework for diagnosing coloscopy diseases. Endoscopists can lower the risk of polyps turning into cancer during colonoscopies by using more accurate computer-assisted polyp detection and segmentation. With the aim of creating a model that can automatically distinguish polyps from images, we presented a modified DeeplabV3+ model in this study to carry out segmentation tasks successfully and efficiently. The framework's encoder uses a pre-trained dilated convolutional residual network for optimal feature map resolution. The robustness of the modified model is tested against state-of-the-art segmentation approaches. In this work, we employed two publicly available datasets, CVC-Clinic DB and Kvasir-SEG, and obtained Dice similarity coefficients of 0.97 and 0.95, respectively. The results show that the improved DeeplabV3+ model improves segmentation efficiency and effectiveness in both software and hardware with only minor changes.

Differential Self-Feedback Dilated Convolution Network With Dual-Tree Channel Attention Mechanism for Hyperspectral Image Classification

Differential Self-Feedback Dilated Convolution Network With Dual-Tree Channel Attention Mechanism for Hyperspectral Image Classification

A Novel Efficient Dual-Gate Mixed Dilated Convolution Network for Multi-Scale Pedestrian Detection

With the increasing use of onboard high-speed computing systems, vehicle manufacturers are offering significant advanced features of driver assistance systems. Pedestrian detection is one of the major requirements of such systems, which commonly use cameras, radar, and ultrasonic sensors. Image recognition based on captured image streams is one of the powerful tools used for the detection of pedestrians, which exhibits similarities and distinguishing features compared to general object detection. Although pedestrian detection has advanced significantly along with deep learning, some issues still need to be addressed. Pedestrian detection is essential for several real-world applications and is an initial step in outdoor scene analysis. Typically, in a crowded situation, conventional detectors are unable to distinguish persons from each other successfully. This study presents a novel technique, based on the Dual Gate Mixed Dilated Convolution Network, to address this problem by adaptively filtering spatial areas where the patterns are still complicated and require further processing. The proposed technique manages obscured patterns while offering improved multiscale pedestrian recognition accuracy.

A lightweight spatiotemporal graph dilated convolutional network for urban sensor state prediction

Spatiotemporal prediction is one attractive research topic in urban computing, which is of great significance to urban planning and management. At present, there are many attempts to predict the spatiotemporal state of systems using various deep learning models. However, most existing models tend to improve prediction accuracy with larger parameter scale and time consumption, but ignoring ease of use in practice. To overcome this question, we propose a lightweight spatiotemporal graph dilated convolutional network called STGDN with satisfactory prediction accuracy and lower model complexity. More specifically, we propose a novel dilated convolution operator and integrate it into traditional causal convolutional networks and graph convolutional networks to greatly improve the efficiency of prediction. The proposed dilated convolution operator can significantly reduce the depth of the model, thereby reducing the parameter scale and improving the computational efficiency of the model. We conducted on multi experiments on three real-world spatiotemporal datasets (traffic dataset, PM2.5 dataset, and temperature dataset) to prove the effectiveness and advantage of our proposed STGDN. The experimental results show that the proposed STGDN model outperforms or achieves comparable prediction accuracy of the existing nine baselines with higher operational efficiency and fewer model parameters. Codes are available at anonymous private link on https://doi.org/10.6084/m9.figshare.23935683.

Self-supervised Learning for DNA sequences with circular dilated convolutional networks

DNA molecules commonly exhibit wide interactions between the nucleobases. Modeling the interactions is important for obtaining accurate sequence-based inference. Although many deep learning methods have recently been developed for modeling DNA sequences, they still suffer from two major issues: 1) most existing methods can handle only short DNA fragments and fail to capture long-range information; 2) current methods always require massive supervised labels, which are hard to obtain in practice. We propose a new method to address both issues. Our neural network employs circular dilated convolutions as building blocks in the backbone. As a result, our network can take long DNA sequences as input without any condensation. We also incorporate the neural network into a self-supervised learning framework to capture inherent information in DNA without expensive supervised labeling. We have tested our model in two DNA inference tasks, the human variant effect and the open chromatin region of plants, where the experimental results show that our method outperforms five other deep learning models. Our code is available at https://github.com/wiedersehne/cdilDNA.

Hard-rock tunnel lithology identification using multi-scale dilated convolutional attention network based on tunnel face images

Hard-rock tunnel lithology identification using multi-scale dilated convolutional attention network based on tunnel face images

A Hybrid Lane Detection Model for Wild Road Conditions

This paper presents a lane detection system in unstructured, complex environments in real time. We formulated lane detection as a binary segmentation problem and proposed a novel hybrid Convolutional Neural Network (CNN) architecture by fusing encoder-decoder network with a sequence of dilated convolution network. We computed weighted average of outputs of these branches for lane detection and introduced a new loss function for binary semantic segmentation to counter the imbalance between lane and non-lane pixels. The proposed model achieved 95.19% accuracy on TuSimple dataset. To evaluate the system with respect to unstructured road scenarios, we created an Indian Lane Dataset (ILD) with 6149 labelled images from India Driving Dataset (IDD). We reported mIoU (Intersection Over Union) of 0.31 on IDD which was higher than other state-of-the art lane detection models. Finally, we undertook an ablation study to understand effectiveness of two parallel branches, i.e., encoder-decoder and dilated convolution branches and the proposed loss function.

SPEED:Semantic Prior and Extremely Efficient Dilated Convolution Network for Real-Time Metal Surface Defects Detection

Automatic defect detection on the metal surface is a vital task for product inspection in industrial assembly lines or production processes. Owing to miscellaneous patterns of defects, inter-class similarity, intra-class difference, and fewer defect samples, achieving accurate and automatic detection remains a big challenge. What's more, since the rising demand for production efficiency, real-time detection is increasingly desirable. This article proposes a semantic prior and extremely efficient dilated convolution network, named SPEED, for pixel-wise detection on the metal surface, which aims to address the aforementioned issues. The architecture of SPEED involves the following: (i) a semantic prior (SP) branch, with shallow layer and prior mapping module to capture low-level details; and (ii) an extremely efficient dilation (EED) branch, with lightweight bottleneck to obtain high-level context. Furthermore, an aggregation module is designed to fuse both types of feature representation. Additionally, different level features of bottleneck are fused to improve the segmentation performance. Experimental results on three metal surface defect datasets indicate that the proposed method outperforms the state-of-the-art approaches in terms of the mean intersection of union, model parameters, FLOPs, and FPS. More specifically, SPEED achieves 92.34% mIoU on NEU-Seg, 88.65% mIoU on Severstal Strip Steel, and 63.91% mIoU on MT Defect.

Detection of driver drowsiness level using a hybrid learning model based on ECG signals.

Fatigue has a considerable impact on the driver's vehicle and even the driver's own operating ability. An intelligent algorithm is proposed for the problem that it is difficult to classify the degree of drowsiness generated by the driver during the driving process. By studying the driver's electrocardiogram (ECG) during driving, two models were established to jointly classify the ECG signals as awake, stress, and fatigue or drowsiness states for drowsiness levels. Firstly, the deep learning method was used to establish the model_1 to predict the drowsiness of the original ECG, and model_2 was developed using the combination of principal component analysis (PCA) and weighted K-nearest neighbor (WKNN) algorithm to classify the heart rate variability characteristics. Then, the drowsiness prediction results of the two models were weighted according to certain rules, and the hybrid learning model combining dilated convolution and bidirectional long short-term memory network with PCA and WKNN algorithm was established, and the mixed model was denoted as DiCNN-BiLSTM and PCA-WKNN (DBPW). Finally, the validity of the DBPW model was verified by simulation of the public database. The experimental results show that the average accuracy, sensitivity and F1 score of the test model in the dataset containing multiple drivers are 98.79, 98.81, and 98.79 % respectively, and the recognition accuracy for drowsiness or drowsiness state is 99.33 %. Using the proposed algorithm, it is possible to identify driver anomalies and provide new ideas for the development of intelligent vehicles.

A hybrid attention and dilated convolution framework for entity and relation extraction and mining

Mining entity and relation from unstructured text is important for knowledge graph construction and expansion. Recent approaches have achieved promising performance while still suffering from inherent limitations, such as the computation efficiency and redundancy of relation prediction. In this paper, we propose a novel hybrid attention and dilated convolution network (HADNet), an end-to-end solution for entity and relation extraction and mining. HADNet designs a novel encoder architecture integrated with an attention mechanism, dilated convolutions, and gated unit to further improve computation efficiency, which achieves an effective global receptive field while considering local context. For the decoder, we decompose the task into three phases, relation prediction, entity recognition and relation determination. We evaluate our proposed model using two public real-world datasets that the experimental results demonstrate the effectiveness of the proposed model.

Automatic Depression Detection via Learning and Fusing Features From Visual Cues

Depression is one of the most prevalent mental disorders, which seriously affects one’s life. Traditional depression diagnostics commonly depend on rating with scales, which can be labor-intensive and subjective. In this context, automatic depression detection (ADD), aiming to assist medical experts in their diagnosis and analysis, has been attracting more attention for its better objectivity and fewer laborious interventions. A typical ADD model detects depression via automatically extracting task-specific features from medical records, such as video sequences, and sending them into a classifier for assistive prediction. However, it remains challenging to effectively extract depression-specific information from long sequences, thereby hindering a satisfying accuracy. In this article, we propose a novel ADD method via learning and fusing features from visual cues. Specifically, we first construct temporal dilated convolutional network (TDCN), in which multiple dilated convolution blocks (DCBs) are designed and stacked, to learn the long-range temporal information from sequences. Then, the featurewise attention (FWA) module is adopted to fuse different features extracted from TDCNs. The module learns to assign weights for the feature channels, aiming to better incorporate different kinds of visual features and further enhance the detection accuracy. Our method achieves the state-of-the-art performance on the Distress Analysis Interview Corpus Wizard-of-Oz (DAIC_WOZ) dataset compared with other visual-feature-based methods, showing its effectiveness.

Dual Parallel Branch Fusion Network for Road Segmentation in High-Resolution Optical Remote Sensing Imagery

Road segmentation from high-resolution (HR) remote sensing images plays a core role in a wide range of applications. Due to the complex background of HR images, most of the current methods struggle to extract a road network correctly and completely. Furthermore, they suffer from either the loss of context information or high redundancy of details information. To alleviate these problems, we employ a dual branch dilated pyramid network (DPBFN), which enables dual-branch feature passing between two parallel paths when it is merged to a typical road extraction structure. A DPBFN consists of three parts: a residual multi-scaled dilated convolutional network branch, a transformer branch, and a fusion module. Constructing pyramid features through parallel multi-scale dilated convolution operations with multi-head attention block can enhance road features while suppressing redundant information. Both branches after fusing can solve shadow or vision occlusions and maintain the continuity of the road network, especially on a complex background. Experiments were carried out on three datasets of HR images to showcase the stable performance of the proposed method, and the results are compared with those of other methods. The OA in the three data sets of Massachusetts, Deep Globe, and GF-2 can reach more than 98.26%, 95.25%, and 95.66%, respectively, which has a significant improvement compared with the traditional CNN network. The results and explanation analysis via Grad-CAMs showcase the effective performance in accurately extracting road segments from a complex scene.

Deep spatio-temporal 3D dilated dense neural network for traffic flow prediction

Traffic flow prediction is increasingly vital for the administration of metropolitan areas. Many research on spatio-temporal networks have been explored but the impacts of both spatial and temporal flexibility, complex spatial correlation has not been considered simultaneously. We present the Spatio-Temporal 3D Multiscale Dilated Dense Network (ST-3DMDDN), a novel 3D Convolutional Neural Network (3DCNN) deep learning neural network for the road level and region level traffic flow prediction. It uses autocorrelation analysis’ early fusion method for importance sampling, a 3D multiscale dilated convolutional network to capture nearby and remote correlations simultaneously, and a densely connected network for deeper feature extraction. Considering traffic flow’s heterogeneity, a new block called the “Spatial and Channel Recalibrate” (SCR) is designed to accurately analyze the correlation contributions. The ST-3DMDDN model is evaluated using three real traffic flows, and the findings indicate that our approach surpasses the performance of the baseline approaches.

PDC-Net: parallel dilated convolutional network with channel attention mechanism for pituitary adenoma segmentation.

Accurate segmentation of the medical image is the basis and premise of intelligent diagnosis and treatment, which has a wide range of clinical application value. However, the robustness and effectiveness of medical image segmentation algorithms remains a challenging subject due to the unbalanced categories, blurred boundaries, highly variable anatomical structures and lack of training samples. For this reason, we present a parallel dilated convolutional network (PDC-Net) to address the pituitary adenoma segmentation in magnetic resonance imaging images. Firstly, the standard convolution block in U-Net is replaced by a basic convolution operation and a parallel dilated convolutional module (PDCM), to extract the multi-level feature information of different dilations. Furthermore, the channel attention mechanism (CAM) is integrated to enhance the ability of the network to distinguish between lesions and non-lesions in pituitary adenoma. Then, we introduce residual connections at each layer of the encoder-decoder, which can solve the problem of gradient disappearance and network performance degradation caused by network deepening. Finally, we employ the dice loss to deal with the class imbalance problem in samples. By testing on the self-established patient dataset from Quzhou People's Hospital, the experiment achieves 90.92% of Sensitivity, 99.68% of Specificity, 88.45% of Dice value and 79.43% of Intersection over Union (IoU).