Residual Convolution Research Articles

Parametric seasonal-trend autoregressive neural network for long-term crop price forecasting.

Crop price forecasting is difficult in that supply is not as elastic as demand, therefore, supply and demand should be stabilized through long-term forecasting and pre-response to the price. In this study, we propose a Parametric Seasonal-Trend Autoregressive Neural Network (PaSTANet), which is a hybrid model that includes both a multi-kernel residual convolution neural network model and a Gaussian seasonality-trend model. To compare the performance of the PaSTANet, we used daily data from the Garak market for four crops: onion, radish, Chinese cabbage, and green onion, and performed long-term price forecasts for one year in 2023. The PaSTANet shows good performance on all four crops compared to other conventional statistical and deep learning-based models. In particular, for onion, the (mean absolute error (MAE) for the long-term forecast of 2023 is 107, outperforming the second-best Prophet (152) by 29.6%. Chinese cabbage, radish, and green onion all outperform the existing models with MAE of 2008, 3703, and 557, respectively. Moreover, using the confidence interval, the predicted price was categorized into three intervals: probability, caution, and warning. Comparing the percentage of classified intervals about the true prices in our test set, we found that they accurately detect the large price volatility.

Dual-resolution transformer combined with multi-layer separable convolution fusion network for real-time semantic segmentation

Environmental perception is crucial for unmanned mobile platforms such as autonomous vehicles and robots. Precise and fast semantic segmentation of the surrounding scene is a key task to enhance this capability. Existing real-time semantic segmentation networks are typically based on convolutional neural networks (CNNs), which have achieved good results, but they still lack control over global context features. In recent years, the Transformer architecture has achieved significant success in capturing global context, which is beneficial for improving segmentation accuracy. However, Transformers tend to ignore local connections, and their computational complexity makes real-time segmentation challenging. We propose a lightweight real-time semantic segmentation network called DTMC-Net, which combines the advantages of CNNs and Transformers. We design a special residual convolution module called the Lightweight Multi-layer Separable Convolution Attention module (LMSCA) to reduce the parameter count and perform multi-scale feature fusion to capture local features effectively. We introduce the Simple Dual-Resolution Transformer (SDR Transformer) that utilizes lightweight attention mechanisms and residual feed forward networks to capture and maintain features, with multiple bilateral fusions between two branches to exchange information. The proposed Anti-artifact Aggregation Pyramid Pooling Module (AAPPM) optimizes the upsampling process, refines features, and performs multi-scale feature fusion again. DTMC-Net only contains 4.2M parameters and achieves good performance on multiple public datasets with different scenarios.

Dual encoding DDS‐UNet liver tumour segmentation based on multi‐scale deep and shallow feature fusion

AbstractThe fusion and utilization of multi‐scale deep and shallow features are of great significance in liver tumour segmentation. This study proposes a dual encoding DDS‐UNet liver tumour segmentation method based on multi‐scale deep and shallow feature fusion, aiming to fully achieve the fusion and utilization of deep and shallow features and achieve accurate segmentation. The proposed method mainly consists of residual convolution module fusion residual convolution (FRC), dual encoding end fusion module dual encoding end fusion (DEF), and skip connection fusion module jump connection fusion module (JCF). In the residual convolution module, a layer by layer fused residual convolution is used instead of traditional convolution to achieve better training. In the dual encoding end fusion module, multi‐scale feature fusion at the end provides more comprehensive contextual information, solves the loss of spatial geometric information. The skip connection fusion module reduces the interference of invalid features by changing the weights of spatial attention and channel attention on important features. The Dice coefficient, average intersection to union ratio, accuracy, recall, and accuracy indicators tested on the LiTS dataset were 90.37%, 90.16%, 93.78%, 94.91%, and 98.84%, respectively, which are superior to many advanced liver tumor segmentation methods.

A Time–Frequency Residual Convolution Neural Network for the Fault Diagnosis of Rolling Bearings

A time–frequency residual convolution neural network (TFRCNN) was proposed to identify various rolling bearing fault types more efficiently. Three novel points about TFRCNN are presented as follows: First, by constructing a double-branch convolution network in the time domain and the frequency domain, the respective features in the time domain and the frequency domain were extracted to ensure the rich and complete feature representation of raw data sources. Second, specific residual structures were designed to prevent learning degradation of the deep network, and global average pooling was adopted to improve the network’s sparsity. Third, TFRCNN was better than the other models in terms of prediction accuracy, robustness, generalization ability, and convergence. The experimental results demonstrate that the prediction accuracy rate of TFRCNN, trained using mixing load data, reached 98.88 to 99.92% after optimizing the initial learning rate and choosing the optimizer and loss function. It was verified that TFRCNN can adaptively learn to extract deep fault features, accurately identify bearing fault conditions, and overcome the limitations of classical shallow feature extraction and classification methods, as well as common convolution neural networks. Hence, this investigation revealed TFRCNN’s potential for bearing fault diagnosis in practical engineering applications.

A Frangi filter aided deep learning approach for palaeochannel recognition

SUMMARY In view of the problem of identifying palaeochannels with high concealment and complex geological structures, this paper proposes an intelligent recognition method for palaeochannels based on Frangi filtering and deep learning. The methodology makes use of Maximum Entropy Wigner–Ville Distribution (MEWVD) method to process the original instantaneous amplitude data, which enhance the distinct features of micro palaeochannels in different sensitive frequency bands. The sample 2-D stratigraphic images generated from these data is labelled for the pre-training process of Attention R2U-Net neural network model. Subsequently, Frangi filter is used to identify and enhance the linear structures of river channels in target stratigraphic images, improving the identification effect of the neural network. Finally, RGB data fusion and 3-D visualization carving are performed on the identification data. This method not only eliminates redundant information using the Frangi filter but also proves that the Attention R2U-Net network model structure with attention mechanism and residual convolution structure can effectively improve the segmentation effect for river channels at different scales. Experimental examples show that this method achieves pixel-level feature segmentation of 3-D seismic data for palaeochannels, accurately depicting their shape, width, thickness, flow direction and other features, thus providing support for subsequent well deployment and horizontal well fracturing selection.

PLU-MCN: Perturbation learning enhanced U-shaped multi-graph convolutional network for traffic flow prediction

Traffic flow prediction is a challenging task in intelligent transportation systems. To improve the accuracy of traffic flow prediction, graph convolutional neural networks and traffic networks are usually combined to reveal dynamic spatial-temporal correlations. However, the existing graph convolution methods need to repeatedly preprocess data of the same type due to network fluctuations, hardware failures, and other troublesome factors. To address this challenge, a novel Perturbation Learning enhanced U-shaped Multi-graph Convolutional Network (PLU-MCN) is proposed. Specifically, PLU-MCN consists of a Perturbation Learning (PL) module and a U-shaped Multi-graph Convolutional Network (U-MCN) based on encoder-decoder architecture. The PL module is designed based on residual convolution and residual transpose convolution to learn perturbation-resistant features from historical time series in the presence of perturbations. The U-MCN consists of multi-graph convolutional modules based on U-Net to extract dynamic spatial-temporal features from the internal and external information of the perturbation-resistant features and output the final predicted values. Compared with existing state-of-the-art models, the PLU-MCN model demonstrated excellently predictive accuracy and reliability in experiments based on five real-world datasets. The key source code and data are available at https://github.com/Bounger2/PLUMCN.

Binary semantic segmentation for detection of prostate adenocarcinoma using an ensemble with attention and residual U-Net architectures.

An accurate determination of the Gleason Score (GS) or Gleason Pattern (GP) is crucial in the diagnosis of prostate cancer (PCa) because it is one of the criterion used to guide treatment decisions for prognostic-risk groups. However, the manually designation of GP by a pathologist using a microscope is prone to error and subject to significant inter-observer variability. Deep learning has been used to automatically differentiate GP on digitized slides, aiding pathologists and reducing inter-observer variability, especially in the early GP of cancer. This article presents a binary semantic segmentation for the GP of prostate adenocarcinoma. The segmentation separates benign and malignant tissues, with the malignant class consisting of adenocarcinoma GP3 and GP4 tissues annotated from 50 unique digitized whole slide images (WSIs) of prostate needle core biopsy specimens stained with hematoxylin and eosin. The pyramidal digitized WSIs were extracted into image patches with a size of 256 × 256 pixels at a magnification of 20×. An ensemble approach is proposed combining U-Net-based architectures, including traditional U-Net, attention-based U-Net, and residual attention-based U-Net. This work initially considers a PCa tissue analysis using a combination of attention gate units with residual convolution units. The performance evaluation revealed a mean Intersection-over-Union of 0.79 for the two classes, 0.88 for the benign class, and 0.70 for the malignant class. The proposed method was then used to produce pixel-level segmentation maps of PCa adenocarcinoma tissue slides in the testing set. We developed a screening tool to discriminate between benign and malignant prostate tissue in digitized images of needle biopsy samples using an AI approach. We aimed to identify malignant adenocarcinoma tissues from our own collected, annotated, and organized dataset. Our approach returned the performance which was accepted by the pathologists.

Convolution Neural Network Bidirectional Long Short-Term Memory for Heartbeat Arrhythmia Classification

Arrhythmia is a heart condition that poses a severe threat to life and requires prompt medical attention. One of the challenges in detecting arrhythmias accurately is that incorrect diagnoses can have severe consequences. In light of this, it is critical to develop a solution that is both effective and reliable. In this study, we propose a residual Convolution Neural Network Bidirectional Long Short-Term Memory (DeepResidualBiLSTM) model for classifying Arrhythmia types, which addresses the vanishing gradient problem and captures the relevant features in the signals’ long dependencies. The model is characterized by its simplicity, stability, and ability to extract meaningful features effectively. Using two well-known datasets, the experimental results demonstrate exceptional accuracy, precision, and recall values of approximately 99.4% at the early stage of 20 epoch training. Furthermore, the model demonstrates a remarkable ability to discriminate between Arrhythmia classes under varying thresholds using the ROC curve metric, with a high value, in most cases, of 100% for accurately detecting positive cases.

A Method for Extracting Photovoltaic Panels from High-Resolution Optical Remote Sensing Images Guided by Prior Knowledge

The extraction of photovoltaic (PV) panels from remote sensing images is of great significance for estimating the power generation of solar photovoltaic systems and informing government decisions. The implementation of existing methods often struggles with complex background interference and confusion between the background and the PV panels. As a result, the completeness and edge clarity of PV panel extraction results are compromised. Moreover, most previous studies have overlooked the unique color characteristics of PV panels. To alleviate these deficiencies and limitations, a method for extracting photovoltaic panels from high-resolution optical remote sensing images guided by prior knowledge (PKGPVN) is proposed. Firstly, aiming to address the problems related to missed extractions and background misjudgments, a Photovoltaic Index (PVI) based on visible images in the three-band is constructed to serve as prior knowledge to differentiate between PV panels and non-PV panels. Secondly, in order to strengthen information interaction between shallow features and deep features and enhance the accuracy and integrity of results, a Residual Convolution Hybrid Attention Module (RCHAM) is introduced into the skip-connection of the encoding–decoding structure. Finally, for the purpose of reducing the phenomenon of blurred edges, a multilevel Feature Loss (FL) function is designed to monitor the prediction results at different scales. Comparative experiments are conducted with seven methods, including U-Net, on publicly available datasets. The experimental results show that our PKGPVN achieves superior performance in terms of evaluation metrics such as IoU (above 82%), Precision (above 91%), Recall (above 89%), and F1-score (above 90%) on the AIR-PV dataset. Additionally, the ablation experiments illustrate the effectiveness of our key parts. The proposed method reduces the phenomena of missed extractions and background misjudgments effectively while producing highly accurate results with clear boundaries.

Intelligent Detection and Segmentation of Space-Borne SAR Radio Frequency Interference

Space-borne synthetic aperture radar (SAR), as an all-weather observation sensor, is an important means in modern information electronic warfare. Since SAR is a broadband active radar system, radio frequency interference (RFI) in the same frequency band will affect the normal observation of the SAR system. Untangling the above problem, this research explores a quick and accurate method for detecting and segmenting RFI-contaminated images. The purpose of the current method is to quickly detect the existence of RFI and to locate it in massive SAR data. Based on deep learning, the method shown in this paper realizes the existence of RFI by determining the presence or absence of interference in the image domain and then performs pixel-level image segmentation on Sentinel-1 RFI-affected quick-look images to locate RFI. Considering the need to quickly detect RFI in massive SAR data, an improved network based on MobileNet is proposed, which replaces some inverted residual blocks in the network with ghost blocks, reducing the number of network parameters and the inference time to 6.1 ms per image. Further, this paper also proposes an improved network called the Smart Interference Segmentation Network (SISNet), which is based on U2Net and replaces the convolution of the VGG blocks in U2Net with a residual convolution and introduces attention mechanisms and a modified RFB module to improve the segmentation mIoU to 87.46% on average. Experiment results and statistical analysis based on the MID dataset and PAIS dataset show that the proposed methods can achieve quicker detection than other CNNs while ensuring a certain accuracy and can significantly improve segmentation performance under the same conditions compared to the original U2Net and other semantic segmentation networks.

Graph Neural Network-Based Short‑Term Load Forecasting with Temporal Convolution

An accurate short-term load forecasting plays an important role in modern power system’s operation and economic development. However, short-term load forecasting is affected by multiple factors, and due to the complexity of the relationships between factors, the graph structure in this task is unknown. On the other hand, existing methods do not fully aggregating data information through the inherent relationships between various factors. In this paper, we propose a short-term load forecasting framework based on graph neural networks and dilated 1D-CNN, called GLFN-TC. GLFN-TC uses the graph learning module to automatically learn the relationships between variables to solve problem with unknown graph structure. GLFN-TC effectively handles temporal and spatial dependencies through two modules. In temporal convolution module, GLFN-TC uses dilated 1D-CNN to extract temporal dependencies from historical data of each node. In densely connected residual convolution module, in order to ensure that data information is not lost, GLFN-TC uses the graph convolution of densely connected residual to make full use of the data information of each graph convolution layer. Finally, the predicted values are obtained through the load forecasting module. We conducted five studies to verify the outperformance of GLFN-TC. In short-term load forecasting, using MSE as an example, the experimental results of GLFN-TC decreased by 0.0396, 0.0137, 0.0358, 0.0213 and 0.0337 compared to the optimal baseline method on ISO-NE, AT, AP, SH and NCENT datasets, respectively. Results show that GLFN-TC can achieve higher prediction accuracy than the existing common methods.

ERGPNet: lesion segmentation network for COVID-19 chest X-ray images based on embedded residual convolution and global perception.

The Segmentation of infected areas from COVID-19 chest X-ray (CXR) images is of great significance for the diagnosis and treatment of patients. However, accurately and effectively segmenting infected areas of CXR images is still challenging due to the inherent ambiguity of CXR images and the cross-scale variations in infected regions. To address these issues, this article proposes a ERGPNet based on embedded residuals and global perception, to segment lesion regions in COVID-19 CXR images. First, aiming at the inherent fuzziness of CXR images, an embedded residual convolution structure is proposed to enhance the ability of internal feature extraction. Second, a global information perception module is constructed to guide the network in generating long-distance information flow, alleviating the interferences of cross-scale variations on the algorithm's discrimination ability. Finally, the network's sensitivity to target regions is improved, and the interference of noise information is suppressed through the utilization of parallel spatial and serial channel attention modules. The interactions between each module fully establish the mapping relationship between feature representation and information decision-making and improve the accuracy of lesion segmentation. Extensive experiments on three datasets of COVID-19 CXR images, and the results demonstrate that the proposed method outperforms other state-of-the-art segmentation methods of CXR images.

Image Restoration via Low-Illumination to Normal-Illumination Networks Based on Retinex Theory.

Under low-illumination conditions, the quality of the images collected by the sensor is significantly impacted, and the images have visual problems such as noise, artifacts, and brightness reduction. Therefore, this paper proposes an effective network based on Retinex for low-illumination image enhancement. Inspired by Retinex theory, images are decomposed into two parts in the decomposition network, and sent to the sub-network for processing. The reconstruction network constructs global and local residual convolution blocks to denoize the reflection component. The enhancement network uses frequency information, combined with attention mechanism and residual density network to enhance contrast and improve the details of the illumination component. A large number of experiments on public datasets show that our method is superior to existing methods in both quantitative and visual aspects.

MJAR: A novel joint generalization-based diagnosis method for industrial robots with compound faults

Compound faults inevitably occur in multi-joint industrial robots resulting in excessive vibration. Intelligent diagnosis for the occurrence and position of fault joints can efficiently reduce the maintenance cost. Compared with laboratory disassembly parts, joint diagnosis of in-situ industrial robots is more challenging due to coupling of multiple subsystems and the industrial noise interference. This paper proposes a novel joint generalization-based fault diagnosis method for industrial robots via the multi-joint attention residual network (MJAR) model. Each joint signals are independently input to parallel residual convolution with unidirectional matrix kernel (ResCUM) in multi-joint decoupling attention (MJDA) module, which is exclusively designed combining residual and attention mechanism to provide decoupling capability of compound faults. And the multi-spatial reconstruction (MSR) module based on sparse sampling is design to provide multi-scale feature extraction adapted to real industrial signals. MJAR can diagnose unseen compound fault combinations without using any transfer strategy as the robot joint generalization-based fault diagnosis (RJGFD) framework. Integrated experiment results show that the MJAR is superior to some advanced congener DL methods.

DRCNN: decomposing residual convolutional neural networks for time series forecasting

Recent studies have shown great performance of Transformer-based models in long-term time series forecasting due to their ability in capturing long-term dependencies. However, Transformers have their limitations when training on small datasets because of their lack in necessary inductive bias for time series forecasting, and do not show significant benefits in short-time step forecasting as well as that in long-time step as the continuity of sequence is not focused on. In this paper, efficient designs in Transformers are reviewed and a design of decomposing residual convolution neural networks or DRCNN is proposed. The DRCNN method allows to utilize the continuity between data by decomposing data into residual and trend terms which are processed by a designed convolution block or DR-Block. DR-Block has its strength in extracting features by following the structural design of Transformers. In addition, by imitating the multi-head in Transformers, a Multi-head Sequence method is proposed such that the network is enabled to receive longer inputs and more accurate forecasts are obtained. The state-of-the-art performance of the presented model are demonstrated on several datasets.

Attention mechanism-guided residual convolution variational autoencoder for bearing fault diagnosis under noisy environments

Due to rolling bearings usually operate under fluctuating working conditions in practical engineering, the raw vibration signals generated by bearing faults have nonlinear and non-stationary characteristics. Additionally, there is a lot of noise interference in the collected bearing vibration signal, which indicates that it is difficult to extract bearing fault information and obtain a satisfactory diagnosis accuracy via using traditional method. Deep learning provides a shining road to address this issue. Nevertheless, traditional deep network model has the shortcomings of poor generalization performance and weak robustness in the feature learning. To improve fault recognition accuracy and obtain a favorable anti-noise robustness, this paper proposes a novel bearing fault diagnosis approach based on attention mechanism-guided residual convolutional variational autoencoder (AM-RCVAE). Firstly, the improved residual module is constructed to overcome the convergence difficulty problem caused by network degradation and promote the model generalization performance by replacing the batch normalization (BN) layer in the traditional residual module with the adaptive BN layer. Subsequently, by incorporating the convolutional block attention module and the improved residual module into convolutional variational autoencoder, a deep network model termed as AM-RCVAE is presented to automatically learn fault features from the original data and perform fault diagnosis tasks. The effectiveness of the proposed approach is verified via two experimental cases. Moreover, the recognition accuracy and diagnostic performance of the proposed approach have been certain improved compared with several representative methods.

An attention-based multi-scale convolution network for intelligent underwater acoustic signal recognition

Underwater acoustic signal identification is always an extremely challenging task due to the intricate marine environments and background noise interference. Existing techniques typically focus on handling only one type of specific features for recognition, overlooking a comprehensive representation of distinct vessel attributes and struggling to handle intricate scenarios. To address these challenges, we employ a band-pass filter bank for feature extraction. This process extracts multiple representations of the same data, each concentrating on distinct frequency components. Subsequently, the filtered multi-view representation is channeled into an attention-based multi-scale convolution network (AMSC-Net) for recognition purposes. AMSC-Net integrates a sequence of attention-based multi-scale convolution (AMSC) blocks to capture pivotal features tied to identifying diverse ship-radiated audio categories. Specifically, a set of parallel residual convolution layers with different kernels are designed to capture multi-scale features, and an adaptive channel attention module is followed to highlight dominant parts within global features. To enrich training data in terms of quantity and diversity, a data augmentation strategy is implemented during the training phase. Our proposals are evaluated on two real-world underwater acoustic signal datasets. Experimental results firmly establish the superiority of our proposed AMSC-Net, which substantially outperforms competitive DNN-based methods. Moreover, our proposals exhibit robustness across different noise levels, achieving an overall accuracy of 95.87% and 95.39% for the two datasets at 0 dB signal-to-noise ratio (SNR), respectively.

Fine-grained image recognition method for digital media based on feature enhancement strategy

The emergence of digital media has changed the way people live and learn. In the context of the new era, digital media is gradually integrating into people’s life and learning. Digital media contains massive images, and fine-grained image recognition for digital media has become an important topic. The challenge of fine-grained image recognition is that the difference between different categories is small, and the difference between the same categories is sometimes large. This work designs a fine-grained image recognition based on feature enhancement (FIRFE). This extracts as much information as possible from fine-grained images under weak supervision to improve the recognition accuracy. When the existing methods extract image features, the feature extraction other than the most significant local feature is not enough. This deals with local features alone and ignores the relationship between features. First, this paper designs a feature enhancement and suppression module to process image features. Secondly, this paper designs pyramid residual convolution. This uses different scale convolution kernels to capture different levels of features in the scene. Thirdly, this paper uses the softpool method to rationally allocate the information weight in the pooling process. Fourth, this paper uses feature focus module to mine more features. This focuses on obtaining similar information in multiple local features as discriminant features to further improve the recognition. Fifthly, this paper carried out systematic experiments on the designed method. The proposed method achieves 94.3%/95.7% accuracy, 92.9%/94.1% recall, and 91.4%/92.2% F1 score on different datasets. This verified the superiority of this method for fine-grained image recognition of digital media.

Fault diagnosis of train traction motor bearing based on improved deep residual network

As one of the important components of train traction motor, the bearing plays a key role in ensuring the safety of train operation. At present, the existing fault diagnosis methods for train traction motor bearings have the disadvantages of poor feature extraction ability, slow diagnosis speed, and low diagnosis accuracy. To solve the above problems, an improved deep residual network (IDRN) is proposed in this paper. Firstly, a new residual convolution block (RCB) is proposed to improve the extraction ability of hidden features. Secondly, the squeeze and excitation block (SE) is embedded in the residual block to increase the sensitivity of the model to features. Lastly, Adan is selected as the optimizer to increase the convergence speed of IDRN. The train traction motor bearing bench built by our research team is selected for the experiment. The results show that the test accuracy of IDRN on the bearing dataset of train traction motor is 98.72%, which is at least 2.27% higher than other comparative models. The IDRN model converges with fewer iterations in the training process, shortening the entire experiment time. At the same time, the CWRU bearing dataset and the bearing dataset with natural defects are selected for comparative verification. The experimental results show that the IDRN model exhibits strong robustness and practicability on different datasets.

The High-Performance Design of a General Spiking Convolution Computation Unit for Supporting Neuromorphic Hardware Acceleration

Recently, the design of spiking neural networks (SNNs) processors based on FPGA have become a hot research topic for their low power and high efficiency. However, most existing works lack flexibility in configuration to accomplish efficient computation of deep SNNs with different network structures (e.g., residuals). In this brief, a high-performance general spiking convolution processing unit and the corresponding hardware architecture (named SCPU) is designed to support both standard and residual convolution; then, a neuromorphic processor containing SCPU (called FPGA-SCPP) is implemented on Xilinx Virtex-7 FPGA. Specifically, the model parameters are fused and quantified to reduce the computational amounts and simplify its hardware implementation. And a spike event processing unit is designed based on event computation and direct data flow to complete the standard and residual convolution computations. The experimental results show the FPGA-SCPP can effectively deploy different large-scale SNNs models such as VGG and ResNet, with the highest recognition accuracies of 92.45% and 68.55% on CIFAR-10/CIFAR-100, respectively. Our FPGA-SCPP has a computational power of only 1.7W and a frame rate of 40fps, which improves inference speed by about 6 times and reduces power consumption by 3 times compared to existing processors. Therefore, FPGA-SCPP can effectively accomplish the acceleration of deep SNNs, especially for models with spiking residual blocks.