Depthwise Separable Convolution Research Articles

Enhanced Feature Refinement Network Based on Depthwise Separable Convolution for Lightweight Image Super-Resolution

Image super-resolution (SR) techniques aim to enhance the clarity and realism of images. Recently, a wide range of excellent SR algorithms with powerful characterization capabilities have emerged and are widely used. However, there are still challenges and room for improvement in designing a lighter and more edge-friendly SR networks for hardware devices. In this paper, we propose a lightweight enhanced feature refinement network (EFRN) based on depthwise separatable convolution for SR reconstruction. The core network components consist of multiple enhanced feature refinement blocks (EFRB), which fully fuse channel features to extract more accurate low-frequency information based on the attention of different channels. In addition, a lightweight residual block (LRB) and a lightweight dual attention block (LDAB) are designed to enhance network information extraction with minimal parameter cost. We improve the feature refinement by using 1 × 1 convolution instead of a channel selection operation to reduce the dimensionality of the features and extract the refined features more efficiently. Finally, to achieve better reconstruction performance, the depth and number of channels of the network are expanded while keeping the total number of parameters at a low level. Extensive experiments have been conducted to demonstrate the superiority of our EFRN over other mainstream SR algorithms in terms of reconstruction results and the number of parameters.

ShiftTransUNet: An Efficient Deep Learning Model for Medical Image Segmentation Using ShiftViT Framework

Deep learning has significantly advanced the field of medical image segmentation. However, the complexity of network structures often leads to high computational demands, limiting their practical efficiency. To enhance the efficiency of image segmentation, this paper introduces an innovative, concise, and lightweight deep learning network. First, to reduce model complexity, we replaced the attention mechanism in the traditional vision transformer (ViT) structure with a shift operation, creating the ShiftViT architecture. This substitution significantly decreased computation and the number of parameters while preserving model performance. Second, to retain and enhance fine-grained features and facilitate more precise information transfer across different layers, we employed a full-scale progressive skip connection strategy. This approach effectively integrates multi-scale feature information, further enhancing model performance. Additionally, to further reduce network complexity, inspired by the independence of probabilities, we opted for depth-wise separable convolution over traditional convolution. This enhances the relative independence between layers. Together, these modifications achieved superior segmentation results on both the Synapse and Automated Cardiac Diagnostic Challenge (ACDC) datasets compared to mainstream models, with substantial advantages in terms of computational efficiency and parameter count. The proposed approach represents an effective solution for medical image applications with limited computational resources and holds great promise for clinical practice.

Transformer assisted by DSC and BiLSTM for bearing fault pattern recognition under strong noise interference

ABSTRACT Deep learning has greatly promoted intelligent fault diagnosis for bearings and has led to the development of many intelligent fault recognition methods based on deep neural networks. However, bearing fault signals in industrial applications are susceptible to noise interference, and intelligent recognition methods for bearing faults under strong noise interference are lacking. An intelligent fault pattern recognition model termed IFPR-MDL, which is based on a Transformer assisted by depthwise separable convolutional (DSC) and bidirectional long short-term memory (BiLSTM) networks, is designed to identify bearing faults under strong noise conditions. In this model, multiscale features are first extracted by a multiscale feature extraction module designed on the basis of DSC; then, contextual information is further extracted by a BiLSTM, and information from different scales is integrated by Transformer encoders. The DSC network facilitates model parameter reduction, the BiLSTM network has expertise in mining temporal features, and the Transformer network excels at capturing long-range feature correlations. This model fully utilises the advantages of these models and complements them to improve the anti-interference ability of bearing fault pattern recognition. Tests on several public datasets show that the proposed model has higher recognition accuracy, stronger generalisation ability, and noise resistance in strong noise situations.

Ground subsidence prediction with high precision: a novel spatiotemporal prediction model with Interferometric Synthetic Aperture Radar technology

ABSTRACT As the extraction of mineral resources intensifies, ground subsidence in mining areas has escalated, posing substantial challenges to sustainable development and operational safety. This subsidence, resulting directly from mining activities, significantly compromises the safety of nearby residents by damaging residential structures and infrastructure. Thus, developing precise and dependable methods for predicting ground subsidence is crucial. This study introduces an innovative Cabs-Unet model, which enhances the U-Net architecture by integrating a Convolutional Block Attention Module (CBAM) and Depthwise Separable Convolutions (DSC). This model aims to predict the spatiotemporal dynamics of the Interferometric Synthetic Aperture Radar (InSAR) time series. Employing Small Baseline Subset Interferometric Synthetic Aperture Radar (SBAS InSAR) technology, we gathered and validated data on ground subsidence at the Pengzhuang coal mine from May 2017 to November 2021, covering 130 scenes, with its accuracy corroborated by levelling survey results. An empirical evaluation of the Cabs-Unet model in two distinct subsidence zones demonstrated superior performance over conventional methods like Convolutional Long Short-Term Memory (ConvLSTM) and Predictive Recurrent Neural Network (PredRNN), with Root Mean Square Error (RMSE) values of 1.44 and 1.70, respectively. These findings highlight the model’s efficacy in accurately predicting spatiotemporal InSAR ground subsidence. Further predictive analysis using InSAR data indicated an expected increase in subsidence, projecting cumulative declines of −457 mm in Area A and −1278 mm in Area B by 17 July 2022. Our model proves effective in assessing subsidence, promptly detecting potential risks and facilitating the rapid implementation of risk mitigation strategies.

Remaining useful life prediction method for cross- condition tools based on parallel fusion

Abstract The current tool remaining useful life (RUL) prediction models do not fully consider the importance of signals from different sensors and the differences in working conditions, which leads to a decline in the model’s robust performance under new conditions. This paper introduces a novel cross-condition RUL prediction approach that integrates parallel fusion, transfer learning, and deep learning. Considering the characteristics of different data sources, correlated signals are initially selected using correlation analysis and dynamic time warping (DTW). Subsequently, multi-scale Depthwise Separable Convolution and multi-attention mechanisms are combined to capture the importance of features at various scales. Finally, a dynamic domain adaptation technique is employed to adjust the importance weights of conditional distribution discrepancy and marginal distribution discrepancy, enabling the model to adapt to cross-condition scenarios.

Enhancing Skin Disease Diagnosis with TFFNet: A Two-Stream Feature Fusion Network Integrating CNNs and Self Attention Block

The skin of an individual serves as the primary defense mechanism for safe guarding vital organs in the body. Although this barrier effectively protects internal organs from a variety of threats, it is still prone to damage from viral, fungal, or dust-related illnesses. Even minor skin injuries possess the potential to escalate into more severe and hazardous conditions. A prompt and precise skin disease diagnosis becomes crucial in expediting the healing process for individuals grappling with skin-related issues. The objective of this study is to develop a system based on Convolutional Neural Network (CNN) that can accurately identify various skin diseases. The proposed architecture, known as TFFNet (Two-Stream Feature Fusion Network), integrates two simultaneous modules featuring a Self-Attention (SA) block. We employ Self Attention-Convolutional Neural Networks (SACNNs) and Depthwise Separable Convolution (DWSC) to establish a diagnostic system for skin diseases. In this method, two separate CNN models are joined together, and two parallel modules (M1 and M2) are added. This greatly reduces the total number of trainable parameters. In comparison to other deep learning methods outlined in existing literature, the proposed CNN exhibits a notably lower number of learned parameters, specifically around 7 million for classification purposes. The skin disease classification was carried out on three datasets—ISIC2016, ISIC2017, and HAM10000. The model achieved testing accuracies of 89.70%, 90.52%, and 90.12% on each respective dataset.

Lightweight decoder U-net crack segmentation network based on depthwise separable convolution

Lightweight decoder U-net crack segmentation network based on depthwise separable convolution

Remaining useful life prognostics of bearings based on convolution attention networks and enhanced transformer

Rolling bearings are critical components of industrial equipment, and predicting their remaining useful life (RUL) is a challenging task. This article proposes a new end-to-end model (MDSCT) to achieve efficient and accurate prediction of bearing RUL. MDSCT uses the raw vibration signals collected by sensors for prediction, without relying on a large amount of prior knowledge. The feature extraction backbone of this model is composed of an MDSC attention module that integrates multiple parallel depth-wise separable convolutions and an efficient attention mechanism, combined with a tansformer encoder (PPSformer) optimized using patch embedding and probesparse self attention techniques. They are respectively used to capture local subtle features and global dependent features in degraded signals. This article also proposes an improved adaptive activation function AdaptH_Swish to enhance the model's ability to model nonlinear relationships. To comprehensively verify the comprehensive performance of the model, this paper conducted detailed ablation and comparative experiments using two standard datasets, PHM2012 and XJTU-SY. The experimental results not only confirm the rationality and efficiency of the model structure design, but also demonstrate significant advantages in three evaluation indicators compared to various existing methods, fully demonstrating the high generalization and strong robustness of the MDSCT model in bearing RUL prediction tasks.

OAR-UNet: Enhancing Long-Distance Dependencies for Head and Neck OAR Segmentation

Accurate segmentation of organs at risk (OARs) is a crucial step in the precise planning of radiotherapy for head and neck tumors. However, manual segmentation methods using CT images, which are still predominantly applied in clinical settings, are inefficient and expensive. Additionally, existing segmentation methods struggle with small organs and have difficulty managing the complex interdependencies between organs. To address these issues, this study proposed an OAR-UNet segmentation method based on a U-shaped architecture with two key designs. To tackle the challenge of segmenting small organs, a Local Feature Perception Module (LFPM) is developed to enhance the sensitivity of the method to subtle structures. Furthermore, a Cross-shaped Transformer Block (CSTB) with a cross-shaped attention mechanism is introduced to improve the ability of the model to capture and process long-distance dependency information. To accelerate the convergence of the Transformer, we designed a Local Encoding Module (LEM) based on depthwise separable convolutions. In our experimental evaluation, we utilized two publicly available datasets, SegRap2023 and PDDCA, achieving Dice coefficients of 78.22% and 89.42%, respectively. These results demonstrate that our method outperforms both previous classic methods and state-of-the-art (SOTA) methods.

Global and local feature extraction based on convolutional neural network residual learning for MR image denoising.

Given the different noise distribution information of global and local magnetic resonance (MR) images, this study aims to extend the current work on convolutional neural networks that preserve global structure and local details in MR image denoising tasks. This study proposed a parallel and serial network for denoising 3D MR images, called 3D-PSNet. We use the residual depthwise separable convolution block to learn the local information of the feature map, reduce the network parameters, and thus improve the training speed and parameter efficiency. In addition, we consider the feature extraction of the global image and utilize residual dilated convolution to process the feature map to expand the receptive field of the network and avoid the loss of global information. Finally, we combine both of them to form a parallel network. What's more, we integrate reinforced residual convolution blocks with dense connections to form serial network branches, which can remove redundant information and refine features to further obtain accurate noise information. The peak signal-to-noise ratio, structural similarity index measure, and root mean square error metrics of 3D-PSNet are as high as 47.79%, 99.81%, and 0.40%, respectively, achieving competitive denoising effect on three public datasets. The ablation experiments demonstrated the effectiveness of all the designed modules regarding all the evaluated metrics in both datasets. The proposed 3D-PSNet takes advantage of multi-scale receptive fields, local feature extraction and residual dense connections to more effectively restore the global structure and local fine features in MR images, and is expected to help doctors quickly and accurately diagnose patients' conditions.

Fault Intelligent Diagnosis for Distribution Box in Hot Rolling Based on Depthwise Separable Convolution and Bi-LSTM

The finishing mill is a critical link in the hot rolling process, influencing the final product’s quality, and even economic efficiency. The distribution box of the finishing mill plays a vital role in power transmission and distribution. However, harsh operating conditions can frequently lead to distribution box damage and even failure. To diagnose faults in the distribution box promptly, a fault diagnosis network model is constructed in this paper. This model combines depthwise separable convolution and Bi-LSTM. Depthwise separable convolution and Bi-LSTM can extract both spatial and temporal features from signals. This structure enables comprehensive feature extraction and fully utilizes signal information. To verify the diagnostic capability of the model, five types of data are collected and used: the pitting of tooth flank, flat-headed sleeve tooth crack, gear surface crack, gear tooth surface spalling, and normal conditions. The model achieves an accuracy of 97.46% and incorporates a lightweight design, which enhances computational efficiency. Furthermore, the model maintains approximately 90% accuracy under three noise conditions. Based on these results, the proposed model can effectively diagnose faults in the distribution box, and reduce downtime in engineering.

CCM-Net: Color compensation and coordinate attention guided underwater image enhancement with multi-scale feature aggregation

Due to the light scattering and wavelength absorption in water, underwater images exhibit blurred details, low contrast, and color deviation. Existing underwater image enhancement methods are divided into traditional methods and deep learning-based methods. Traditional methods either rely on scene prior and lack robustness, or are not flexible enough resulting in poor enhancement effects. Deep learning methods have achieved good results in the field of underwater image enhancement due to their powerful feature representation ability. However, these methods cannot enhance underwater images with various degradations because they do not consider the inconsistent attenuation of different color channels and spatial regions. In this paper, we propose a novel asymmetric encoder-decoder network for underwater image enhancement, called CCM-Net. Concretely, we first introduce the prior knowledge-based encoder, which includes color compensation (CC) modules and feature extraction modules that consist of depth-wise separable convolution and global-local coordinate attention (GLCA). Then, we design a multi-scale feature aggregation (MFA) module to integrate shallow, middle, and deep features. Finally, we deploy a decoder to reconstruct the underwater images with the extracted features. Extensive experiments on publicly available datasets demonstrate that our CCM-Net effectively improves the visual quality of underwater images and achieves impressive performance.

Application of Instance Segmentation to Identifying Insect Concentrations in Data from an Entomological Radar

Entomological radar is one of the most effective tools for monitoring insect migration, capable of detecting migratory insects concentrated in layers and facilitating the analysis of insect migration behavior. However, traditional entomological radar, with its low resolution, can only provide a rough observation of layer concentrations. The advent of High-Resolution Phased Array Radar (HPAR) has transformed this situation. With its high range resolution and high data update rate, HPAR can generate detailed concentration spatiotemporal distribution heatmaps. This technology facilitates the detection of changes in insect concentrations across different time periods and altitudes, thereby enabling the observation of large-scale take-off, landing, and layering phenomena. However, the lack of effective techniques for extracting insect concentration data of different phenomena from these heatmaps significantly limits detailed analyses of insect migration patterns. This paper is the first to apply instance segmentation technology to the extraction of insect data, proposing a method for segmenting and extracting insect concentration data from spatiotemporal distribution heatmaps at different phenomena. To address the characteristics of concentrations in spatiotemporal distributions, we developed the Heatmap Feature Fusion Network (HFF-Net). In HFF-Net, we incorporate the Global Context (GC) module to enhance feature extraction of concentration distributions, utilize the Atrous Spatial Pyramid Pooling with Depthwise Separable Convolution (SASPP) module to extend the receptive field for understanding various spatiotemporal distributions of concentrations, and refine segmentation masks with the Deformable Convolution Mask Fusion (DCMF) module to enhance segmentation detail. Experimental results show that our proposed network can effectively segment concentrations of different phenomena from heatmaps, providing technical support for detailed and systematic studies of insect migration behavior.

A Lightweight Pathological Gait Recognition Approach Based on a New Gait Template in Side-View and Improved Attention Mechanism.

As people age, abnormal gait recognition becomes a critical problem in the field of healthcare. Currently, some algorithms can classify gaits with different pathologies, but they cannot guarantee high accuracy while keeping the model lightweight. To address these issues, this paper proposes a lightweight network (NSVGT-ICBAM-FACN) based on the new side-view gait template (NSVGT), improved convolutional block attention module (ICBAM), and transfer learning that fuses convolutional features containing high-level information and attention features containing semantic information of interest to achieve robust pathological gait recognition. The NSVGT contains different levels of information such as gait shape, gait dynamics, and energy distribution at different parts of the body, which integrates and compensates for the strengths and limitations of each feature, making gait characterization more robust. The ICBAM employs parallel concatenation and depthwise separable convolution (DSC). The former strengthens the interaction between features. The latter improves the efficiency of processing gait information. In the classification head, we choose to employ DSC instead of global average pooling. This method preserves the spatial information and learns the weights of different locations, which solves the problem that the corner points and center points in the feature map have the same weight. The classification accuracies for this paper's model on the self-constructed dataset and GAIT-IST dataset are 98.43% and 98.69%, which are 0.77% and 0.59% higher than that of the SOTA model, respectively. The experiments demonstrate that the method achieves good balance between lightweightness and performance.

Multi-Scale Depthwise Separable Capsule Network for hyperspectral image classification.

Addressing the challenges in effectively extracting multi-scale features and preserving pose information during hyperspectral image (HSI) classification, a Multi-Scale Depthwise Separable Capsule Network (MDSC-Net) is proposed in this article for HSI classification. Initially, hierarchical features are extracted by MDSC-Net through the employment of parallel multi-scale convolutional kernels, while computational complexity is reduced via depthwise separable convolutions, thus reducing the overall computational load and achieving efficient feature extraction. Subsequently, to enhance the translational invariance of features and reduce the loss of pose information, features of various scales are processed in parallel by independent capsule networks, with improvements in max pooling achieved through dynamic routing. Lastly, features of different scales are concatenated and integrated through the concatenate operation, thereby facilitating precise analysis of multi-level information in the hyperspectral image classification process. Experimental comparisons demonstrate that MDSC-Net achieves average accuracies of 94%, 98%, and 99% on the Kennedy Space Center, University of Pavia, and Salinas datasets, respectively, indicating a significant performance advantage over recent HSI classification models and validating the effectiveness of the proposed model.

Cataract and glaucoma detection based on Transfer Learning using MobileNet

A serious eye condition called cataracts can cause blindness. Early and accurate cataract detection is the most effective method for reducing risk and averting blindness. The optic nerve head is harmed by the neurodegenerative condition known as glaucoma. Machine learning and deep learning systems for glaucoma and cataract detection have recently received much attention in research. The automatic detection of these diseases also depends on deep learning transfer learning platforms like VeggNet, ResNet, and MobilNet. The authors proposed MobileNetV1 and MobileNetV2 based on an optimized architecture building lightweight deep neural networks using depth-wise separable convolutions. The experiments used publicly available data sets with both cataract & normal and glaucoma & normal images, and the results showed that the proposed model had the highest accuracy compared to the other models.

Lightweight U-Net based on depthwise separable convolution for cloud detection onboard nanosatellite

Lightweight U-Net based on depthwise separable convolution for cloud detection onboard nanosatellite

A novel approach for automatic classification of macular degeneration OCT images

Age-related macular degeneration (AMD) and diabetic macular edema (DME) are significant causes of blindness worldwide. The prevalence of these diseases is steadily increasing due to population aging. Therefore, early diagnosis and prevention are crucial for effective treatment. Classification of Macular Degeneration OCT Images is a widely used method for assessing retinal lesions. However, there are two main challenges in OCT image classification: incomplete image feature extraction and lack of prominence in important positional features. To address these challenges, we proposed a deep learning neural network model called MSA-Net, which incorporates our proposed multi-scale architecture and spatial attention mechanism. Our multi-scale architecture is based on depthwise separable convolution, which ensures comprehensive feature extraction from multiple scales while minimizing the growth of model parameters. The spatial attention mechanism is aim to highlight the important positional features in the images, which emphasizes the representation of macular region features in OCT images. We test MSA-NET on the NEH dataset and the UCSD dataset, performing three-class (CNV, DURSEN, and NORMAL) and four-class (CNV, DURSEN, DME, and NORMAL) classification tasks. On the NEH dataset, the accuracy, sensitivity, and specificity are 98.1%, 97.9%, and 98.0%, respectively. After fine-tuning on the UCSD dataset, the accuracy, sensitivity, and specificity are 96.7%, 96.7%, and 98.9%, respectively. Experimental results demonstrate the excellent classification performance and generalization ability of our model compared to previous models and recent well-known OCT classification models, establishing it as a highly competitive intelligence classification approach in the field of macular degeneration.

Improved Taillight Detection Model for Intelligent Vehicle Lane-Change Decision-Making Based on YOLOv8

With the rapid advancement of autonomous driving technology, the recognition of vehicle lane-changing can provide effective environmental parameters for vehicle motion planning, decision-making and control, and has become a key task for intelligent vehicles. In this paper, an improved method for vehicle taillight detection and intent recognition based on YOLOv8 (You Only Look Once version 8) is proposed. Firstly, the CARAFE (Context-Aware Reassembly Operator) module is introduced to address fine perception issues of small targets, enhancing taillight detection accuracy. Secondly, the TriAtt (Triplet Attention Mechanism) module is employed to improve the model’s focus on key features, particularly in the identification of positive samples, thereby increasing model robustness. Finally, by optimizing the EfficientP2Head (a small object auxiliary head based on depth-wise separable convolutions) module, the detection capability for small targets is further strengthened while maintaining the model’s practicality and lightweight characteristics. Upon evaluation, the enhanced algorithm demonstrates impressive results, achieving a precision rate of 93.27%, a recall rate of 79.86%, and a mean average precision (mAP) of 85.48%, which shows that the proposed method could effectively achieve taillight detection.

MSRA-Net: multi-channel semantic-aware and residual attention mechanism network for unsupervised 3D image registration

Objective. Convolutional neural network (CNN) is developing rapidly in the field of medical image registration, and the proposed U-Net further improves the precision of registration. However, this method may discard certain important information in the process of encoding and decoding steps, consequently leading to a decline in accuracy. To solve this problem, a multi-channel semantic-aware and residual attention mechanism network (MSRA-Net) is proposed in this paper. Approach. Our proposed network achieves efficient information aggregation by cleverly extracting the features of different channels. Firstly, a context-aware module (CAM) is designed to extract valuable contextual information. And the depth-wise separable convolution is employed in the CAM to alleviate the computational burden. Then, a new multi-channel semantic-aware module (MCSAM) is designed for more comprehensive fusion of up-sampling features. Additionally, the residual attention module is introduced in the up-sampling process to extract more semantic information and minimize information loss. Main results. This study utilizes Dice score, average symmetric surface distance and negative Jacobian determinant evaluation metrics to evaluate the influence of registration. The experimental results demonstrate that our proposed MSRA-Net has the highest accuracy compared to several state-of-the-art methods. Moreover, our network has demonstrated the highest Dice score across multiple datasets, thereby indicating that the superior generalization capabilities of our model. Significance. The proposed MSRA-Net offers a novel approach to improve medical image registration accuracy, with implications for various clinical applications. Our implementation is available at https://github.com/shy922/MSRA-Net.