Deep Learning-based Methods Research Articles

MGNDTI: A Drug-Target Interaction Prediction Framework Based on Multimodal Representation Learning and the Gating Mechanism.

Drug-Target Interaction (DTI) prediction facilitates acceleration of drug discovery and promotes drug repositioning. Most existing deep learning-based DTI prediction methods can better extract discriminative features for drugs and proteins, but they rarely consider multimodal features of drugs. Moreover, learning the interaction representations between drugs and targets needs further exploration. Here, we proposed a simple M ulti-modal G ating N etwork for DTI prediction, MGNDTI, based on multimodal representation learning and the gating mechanism. MGNDTI first learns the sequence representations of drugs and targets using different retentive networks. Next, it extracts molecular graph features of drugs through a graph convolutional network. Subsequently, it devises a multimodal gating network to obtain the joint representations of drugs and targets. Finally, it builds a fully connected network for computing the interaction probability. MGNDTI was benchmarked against seven state-of-the-art DTI prediction models (CPI-GNN, TransformerCPI, MolTrans, BACPI, CPGL, GIFDTI, and FOTF-CPI) using four data sets (i.e., Human, C. elegans, BioSNAP, and BindingDB) under four different experimental settings. Through evaluation with AUROC, AUPRC, accuracy, F1 score, and MCC, MGNDTI significantly outperformed the above seven methods. MGNDTI is a powerful tool for DTI prediction, showcasing its superior robustness and generalization ability on diverse data sets and different experimental settings. It is freely available at https://github.com/plhhnu/MGNDTI.

WaveFusionNet: Infrared and visible image fusion based on multi-scale feature encoder–decoder and discrete wavelet decomposition

To merge complementary information from multimodal images, such as thermal saliency from infrared images and texture details from visible images, traditional multi-scale transform-based methods have been extensively studied, with deep learning-based methods gaining significant popularity in recent years. However, there has been limited research on optimally combining the advantages of these two categories in fusion. In this paper, we propose a novel infrared and visible image fusion (IVIF) framework, WaveFusionNet, which integrates precise frequency feature decomposition from the discrete wavelet transform (DWT) with the comprehensive feature extraction from the multi-scale encoder. Firstly, we train an encoder–decoder network for multi-scale feature extraction and image reconstruction. DWT is used for down-sampling with minimal information loss by decomposing extracted features into low and high-frequency sub-bands. Next, a dual-band feature fusion (DBFF) module is trained to merge these sub-bands by integrating a spatial feature transform-based sub-network for low-frequency fusion and a maximum absolute value selection strategy for fusing high-frequencies. Finally, all fused sub-bands are fed into the pre-trained decoder to reconstruct the final image. Experimental results on three benchmark datasets (TNO, Roadscene, and MSRS) demonstrate that the proposed fusion method outperforms recent IVIF methods in both quantitative assessment and visual perception while maintaining competitive time complexity.

Transitions in dynamical regime and neural mode underlie perceptual decision-making.

Perceptual decision-making is the process by which an animal uses sensory stimuli to choose an action or mental proposition. This process is thought to be mediated by neurons organized as attractor networks 1,2 . However, whether attractor dynamics underlie decision behavior and the complex neuronal responses remains unclear. Here we use an unsupervised, deep learning-based method to discover decision-related dynamics from the simultaneous activity of neurons in frontal cortex and striatum of rats while they accumulate pulsatile auditory evidence. We found that trajectories evolved along two sequential regimes, the first dominated by sensory inputs, and the second dominated by the autonomous dynamics, with flow in a direction (i.e., "neural mode") largely orthogonal to that in the first regime. We propose that the second regime corresponds to decision commitment. We developed a simplified model that approximates the coupled transition in dynamics and neural mode and allows precise inference, from each trial's neural activity, of a putative internal decision commitment time in that trial. The simplified model captures diverse and complex single-neuron temporal profiles, such as ramping and stepping 3-5 . It also captures trial-averaged curved trajectories 6-8 , and reveals distinctions between brain regions. The putative neurally-inferred commitment times ("nTc") occurred at times broadly distributed across trials, and not time-locked to stimulus onset, offset, or response onset. Nevertheless, when trials were aligned to nTc, behavioral analysis showed that, as predicted by a decision commitment time, sensory evidence before nTc affected the subjects' decision, but evidence after nTc did not. Our results show that the formation of a perceptual choice involves a rapid, coordinated transition in both the dynamical regime and the neural mode of the decision process, and suggest the moment of commitment to be a useful entry point for dissecting mechanisms underlying rapid changes in internal state.

A DNN robust video watermarking method in dual-tree complex wavelet transform domain

Deep learning is increasingly being applied in the field of robust watermarking. However, the existing deep learning-based video watermarking methods only uses spatial domain information as the input and the robustness against attacks such as H.264/AVC compression is still not strong. Therefore, this paper proposes a deep learning-based robust video watermarking method in dual-tree complex wavelet transform (DT-CWT) domain. The video frames are transformed into the DT-CWT domain and the suitable high-pass subbands are selected as candidate embedding positions. Then, the 2D and 3D convolutions are combined to extract both intra-frame spatial features and inter-frame temporal features for finding the stable and imperceptible coefficients for watermark embedding in the candidate positions. The convolutional attention module (CBAM) is used to further adjust the embedding coefficients and strengths. In addition, the attack layer, where a differentiable proxy is specially designed in this paper for the simulation of non-differentiable H.264/AVC compression, is introduced to generate distorted watermarked videos for improving the robustness against different attacks. Experimental results show that our method is superior to both the existing deep learning-based methods and traditional methods in the robustness against both spatial and temporal attacks while preserving high video quality. The source code is available at https://github.com/imagecbj/A-DNN-Robust-Video-Watermarking-Method-in-DT-CWT-Domain.

Anomaly detection for multivariate time series in IoT using discrete wavelet decomposition and dual graph attention networks

Effective anomaly detection in multivariate time series data is critical to ensuring the security of Internet of Things (IoT) devices and systems. However, building a high precision and low false positive rate anomaly detection model for the complex and volatile IoT environment is a challenging task. This is often due to issues such as a lack of anomaly labeling, high data volatility, and the complexity of device mechanisms. Traditional machine learning algorithms and sequence models frequently fail to account for feature correlation and temporal dependency in anomaly detection. Although deep learning-based anomaly detection methods have progressed, there is still room for improvement in precision, recall, and generalization ability. In this paper, we propose an anomaly detection model called Meta-MWDG to address these issues. The model is based on a multi-scale discrete wavelet decomposition and a dual graph attention network, which can effectively extract feature correlation and temporal dependency in multivariate time series data. Additionally, model-agnostic meta-learning (MAML) is introduced to improve the model’s generalization performance, enabling it to perform well on new tasks even with a few samples. A gated recurrent unit (GRU) is combined with a multi-head self-attention network to output both prediction and reconstruction results in a joint optimization strategy, improving the precision of anomaly detection. Extensive experimental studies demonstrate that Meta-MWDG outperforms the state-of-the-art methods in anomaly detection.

A domain adaptation framework for cross-modality SAR 3D reconstruction point clouds segmentation utilizing LiDAR data

Synthetic aperture radar (SAR) 3D point clouds reconstruction can eliminate the problems of layover in 2D SAR image projections, the recognition of the reconstructed point cloud can significantly enhance target identification and information extraction. Current SAR 3D point cloud segmentation methods, based on traditional machine learning clustering approaches, suffer from poor automation and accuracy. However, the absence of publicly labeled SAR datasets impedes the progress of deep learning-based SAR 3D point cloud segmentation methods. To tackle the aforementioned challenges, we introduce, for the first time, an alternative training approach for SAR 3D point cloud segmentation using LiDAR annotated data, which offers a more abundant sample pool, thus alleviating the lack of training sets. Nevertheless, a segmentation model trained on LiDAR point clouds exhibits a significant decline in performance when directly applied to SAR 3D reconstructed point clouds due to the cross-domain discrepancy. This research presents a pioneering domain adaptation 3D semantic segmentation framework to implement cross-modal learning for SAR point clouds. In our scheme, a simple yet effective technique is developed to achieve segmentation of SAR double-bounce scattering regions called SARDBS-Mix, which employs a mixing strategy, to capture the distinctive reflection characteristics of SAR data. Furthermore, we implement approaches including center alignment and normalization, local augmentation, and weighted cross entropy to mitigate LiDAR and SAR domain gap and class imbalances. The experimental results validate the feasibility and effectiveness of the proposed method for SAR 3D reconstructed point cloud segmentation.

EHOPN: A novel enhanced high-order pooling-based network for industrial fault detection

Recently, deep learning algorithms have been successfully applied to industrial fault detection because they are better at automatically extracting complex features and processing high-dimensional data than traditional methods. However, most existing deep learning-based fault detection methods only concentrate on extracting features from industrial process data without considering the crucial long-term temporal features and higher-order statistical information. To address this challenge, we proposed a novel enhanced higher-order pooling-based network (EHOPN) for industrial fault detection. First, the data pre-processing of the network is presented to capture the dynamic features of time-series process data and unify the high-dimensional data scale. Second, the EHOPN utilizes channel and temporal second-order pooling techniques to gather temporal and channel statistics information, facilitating the backbone network’s ability to capture complex inter-dependencies and long-term dynamics. Additionally, the high-order feature aggregation module is presented to aggregate global and local features, enhancing the network’s generalization ability. The proposed industrial fault detection approach is evaluated on the Tennessee Eastman benchmark and a real-world heavy-plate production process. Experimental results show that the proposed method is significantly better than comparison models in four evaluation metrics: accuracy, precision, recall, and F1-score, further proving the effectiveness of EHOPN.

Thyroid-DETR: Thyroid nodule detection model with transformer in ultrasound images

Deep learning-based object detection methods using ultrasound images to assist clinicians in rapid and accurate diagnosis of thyroid nodules has been an attractive area of research. Despite the great progress made by object detectors, these may still produce unsatisfactory performance due to certain factors, such as the presence of abundant background information and significant nodule size variations in ultrasound images of thyroid nodules. In this study, we draw on the idea of a Transformer and design a powerful thyroid nodule ultrasound imaging detection model, namely Thyroid-DETR. The proposed Thyroid-DETR has the following features: (i) a deformable convolution v2 is introduced flexibly capture small nodules of irregular shapes and small sizes, (ii) a multi-head self-attention structure is designed to acquire and learn broader correlated information in the shallow layer, (iii) a dual-stream training structure with BatchFormerV2 is developed to fully capture the effective information of medical images; (iv) fine-tuning methods to optimize the model are used to improve the diagnostic accuracy for detecting thyroid nodules. To evaluate the efficacy and effectiveness of the proposed approach. Extensive experiments and comprehensive comparisons are conducted on real thyroid nodule ultrasound images from the hospital. The results show that our method significantly outperforms the state-of-the-art methods.

Deep Learning-Based Methods for Multi-Class Rice Disease Detection Using Plant Images

Deep Learning-Based Methods for Multi-Class Rice Disease Detection Using Plant Images

Hybrid CNN-transformer network for efficient CSI feedback

In recent years, many deep learning-based methods have been utilized for the feedback of Channel State Information (CSI) in massive MIMO systems. The Transformer-based networks leverage global self-attention mechanisms that can effectively capture remote correlations between antennas, while Convolutional Neural Networks (CNNs) excel in acquiring local information. To balance the advantages of both, this paper proposes an Efficient Feature Aggregation Network called EFANet, which hybrid CNNs and Transformer. Specifically, we propose a Refined Window Multi-head Self-Attention (RW-MSA) through hybrid Convolutional Embedding Unit (CEU) and Window Multi-head Self-Attention (W-MSA) to reduce information loss between windows and achieve efficient feature aggregation. Additionally, we develop a Local Enhanced Feedforward Network (LEFN) to further integrate local information in the CSI matrix and model detailed features of different regions. Finally, the Compensation Unit (CU) is designed to further compensate for global-local features in the CSI matrix. Through the above design, the global and local features are fully interactive to reduce information loss. Numerous experiments have shown that the proposed method achieves better CSI reconstruction performance while reducing computational complexity.

Data augmentation aided excavator activity recognition using deep convolutional conditional generative adversarial networks

Excavator activity recognition is a crucial task with significant practical applications such as assessing production efficiency, injury prevention, and enabling intelligent control. However, implementing deep learning-based activity recognition methods presents considerable challenges due to the requirement of large training datasets. And sensor-based activity recognition demands more convenient and reliable sensors. This paper proposes a generalized data augmentation framework for excavator activity recognition. Main pump pressure and flow signals are generated using Deep Convolutional Conditional Generative Adversarial Networks (DC-CGAN). The proposed framework integrates Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) architectures for activity recognition. Results indicate that the framework outperforms traditional data augmentation methods regarding the quality of generated data and shallow and single deep networks regarding model accuracy and generalization. Additionally, the framework was implemented on three sizes of excavators and a wheel loader, demonstrating excellent versatility.

MSCS: Multi-stage feature learning with channel-spatial attention mechanism for infrared and visible image fusion

The intention of infrared and visible image fusion is to combine the images captured by different modal sensors in the same scene to enhance its understanding. Deep learning has been proven its powerful application in image fusion due to its fine generalization, robustness, and representability of deep features. However, the performance of these deep learning-based methods heavily depends on the illumination condition. Especially in dark or exposed scenes, the fused results are over-smoothness and low-contrast, resulting in inaccuracy of object detection. To address these issues, this paper develops a multi-stage feature learning approach with channel-spatial attention mechanism, namely MSCS, for infrared and visible image fusion. The MSCS is composed of the following four key procedures: Firstly, the infrared and visible images are decomposed into illumination and reflectance components by a proposed network called as Retinex_Net. Then, the components are transported to an encoder for features coding. Next, we propose an adaptive fusion module with attention mechanisms to fuse the features. Finally, the fused image is generated by the decoder for decoding the fused features. Meanwhile, a novel fusion loss function and a multi-stage training strategy are proposed to train the above modules. The subjective and objective results of experiments on TNO, LLVIP and MSRS datasets illustrate that the proposed method is effective and performs better than the state-of-the-art fusion methods on achieving enjoyable results in dark or over-exposure scenes. And the results of further experiments on the fused images for object detection demonstrate that the fusion outputs produced by our MSCS are more beneficial for detection tasks.

ACSwinNet: A Deep Learning-Based Rigid Registration Method for Head-Neck CT-CBCT Images in Image-Guided Radiotherapy.

Accurate and precise rigid registration between head-neck computed tomography (CT) and cone-beam computed tomography (CBCT) images is crucial for correcting setup errors in image-guided radiotherapy (IGRT) for head and neck tumors. However, conventional registration methods that treat the head and neck as a single entity may not achieve the necessary accuracy for the head region, which is particularly sensitive to radiation in radiotherapy. We propose ACSwinNet, a deep learning-based method for head-neck CT-CBCT rigid registration, which aims to enhance the registration precision in the head region. Our approach integrates an anatomical constraint encoder with anatomical segmentations of tissues and organs to enhance the accuracy of rigid registration in the head region. We also employ a Swin Transformer-based network for registration in cases with large initial misalignment and a perceptual similarity metric network to address intensity discrepancies and artifacts between the CT and CBCT images. We validate the proposed method using a head-neck CT-CBCT dataset acquired from clinical patients. Compared with the conventional rigid method, our method exhibits lower target registration error (TRE) for landmarks in the head region (reduced from 2.14 ± 0.45 mm to 1.82 ± 0.39 mm), higher dice similarity coefficient (DSC) (increased from 0.743 ± 0.051 to 0.755 ± 0.053), and higher structural similarity index (increased from 0.854 ± 0.044 to 0.870 ± 0.043). Our proposed method effectively addresses the challenge of low registration accuracy in the head region, which has been a limitation of conventional methods. This demonstrates significant potential in improving the accuracy of IGRT for head and neck tumors.

A cosine similarity-based labeling technique for vulnerability type detection using source codes

Vulnerability detection is of great importance in providing reliability to software systems. Although existing methods achieve remarkable success in vulnerability detection, they have several disadvantages as follows: (1) The irrelevant information is removed from source codes, which have a high noise ratio, thereby utilizing deep learning methods and devising experiments featuring high accuracy. However, deep learning-based detection methods necessitate large-scale datasets. This results in computational hardship with respect to vulnerability detection in small-scale software systems. (2) The majority of the studies perform feature selection by processing vulnerability commits. Despite tremendous endeavors, there are few works detecting vulnerability with source codes. To solve these two problems, in this study, a novel labeling and vulnerability detection algorithm is proposed. The algorithm first exploits source codes with the help of a keyword vulnerability matrix. After that, an ultimate encoded matrix is generated by word2vec, thereby combining the labeling vector with the source code matrix to reveal a trainable dataset for a generalized linear model (GLM). Different from preceding studies, our method performs vulnerability detection without requiring vulnerability commits but using source codes. In addition to this, similar studies generally aim to bring sophisticated solutions for just one type of programming language. Conversely, our study develops vulnerability keywords for three programming languages including C#, Java, and C++, and creates the related labeling vectors by regarding the keyword matrix. The proposed method outperformed the baseline approaches for most of the experimental datasets with over 90% of the area under the curve (AUC). Further, there is a 7.7% margin between our method and the alternatives on average for Recall, Precision, and F1-score with respect to five types of vulnerabilities.

A Deep Federated Learning-Based User Credit Evaluation Model Under Financial Internet of Things Scenarios

With the rapid development of Internet of Things (IoT) technology and the digital transformation of the financial industry, the financial IoT is becoming an important trend in the future financial field. In the era of financial IoT, a large amount of data information is recorded by sensors and devices, which poses new challenges and opportunities for user credit evaluation. In this context, conventional deep learning-based user credit evaluation methods often cannot meet privacy needs. The paper combines the privacy security ability of federated learning with deep learning, and proposes a deep federated learning-based user credit evaluation model under financial IoT scenarios. First, a particle swarm optimization-based backpropagation neural network algorithm is formulated to extract user credit evaluation features, in order to obtain user behavior patterns and credit features. Then, the XGBoost algorithm is employed to output credit evaluation results. As for the training process, the thought of federated learning is integrated to distribute the model to individual participants. In such mode, the model can be updated and trained without exposing user data to the central cloud. The experiments are conducted on real-world data to assess the proposal’s performance. The results show that the proposed credit evaluation framework can achieve better accuracy, under the guarantee of user privacy.

Integrating geographic knowledge into deep learning for spatiotemporal local climate zone mapping derived thermal environment exploration across Chinese climate zones

The Local Climate Zone (LCZ) scheme representing urban structure and land use pattern is essential for urban heat island (UHI) research. Fine-grained LCZ mapping considering spatial and temporal heterogeneity can provide a more precise characterization of surface thermal properties, thereby enabling a comprehensive analysis and understanding of spatiotemporal trends in climate change research. However, data-driven deep learning-based methods have limitations in coping with the complex urban landscapes of the real-world scenarios, including the spectral similarity of different LCZ and the geospatial heterogeneity of urban LCZ categories. In this study, we constructed a geographic knowledge base for enhanced LCZ characterization with the consideration of prior surface spatial information, including a set of spectral indices and urban morphological parameters (UMPs). Then, we integrated the explicable geographic knowledge base into a learnable deep learning framework in an end-to-end manner for accurate LCZ mapping by fusing multi-source heterogeneous data with a multi-level fusion strategy. The constructed Chinese Climate Zone Time Series LCZ (CClimate-TLCZ) dataset derived from Landsat-8 data with a 30 m spatial resolution, covering 18 representative cities in China, were used to evaluate the proposed framework. The experimental results demonstrate that the proposed framework achieved optimal outcomes across 18 cities, with an average overall accuracy exceeding 94 %, which is more than 20 % higher than that obtained by the standard WUDAPT method. Furthermore, the analysis of LCZ mapping-driven land surface temperature (LST) and surface UHI (SUHI) applications shows that cities within the same climate zone have similar LST distribution patterns, while significant heterogeneity exist between different zones. The annual consistency of LST patterns within each climate zone supports the validity of the LCZ classification scheme and accurate LCZ mapping for urban thermal environment studies. From 2016 to 2022, SUHI intensity initially increases and then decreases, indicating improvements in the urban thermal environment. These findings underscore the critical role of precise LCZ mapping in urban climate resilience and sustainable urban planning.

SemiH: DFT Hamiltonian neural network training with semi-supervised learning

Abstract Over the past decades, density functional theory (DFT) calculations have been utilized in various fields such as materials science and semiconductor devices. However, due to the inherent nature of DFT calculations, which rigorously consider interactions between atoms, they require significant computational cost. To address this, extensive research has recently focused on training neural networks to replace DFT calculations. However, previous methods for training neural networks necessitated an extensive number of DFT simulations to acquire the ground truth (Hamiltonians). Conversely, when dealing with a limited amount of training data, deep learning models often display increased errors in predicting Hamiltonians and band structures for testing data. This phenomenon poses the potential risk of generating inaccurate physical interpretations, including the emergence of unphysical branches within band structures. To tackle this challenge, we propose a novel deep learning-based method for calculating DFT Hamiltonians, specifically tailored to produce accurate results with limited training data. Our framework not only employs supervised learning with the calculated Hamiltonian but also generates pseudo Hamiltonians (targets for unlabeled data) and trains the neural networks on unlabeled data. Particularly, our approach, which leverages unlabeled data, is noteworthy as it marks the first attempt in the field of neural network Hamiltonians. Our framework showcases the superior performance of our framework compared to the state-of-the-art approach across various datasets, such as MoS2, Bi2Te3, HfO2, and InGaAs. Moreover, our framework demonstrates enhanced generalization performance by effectively utilizing unlabeled data, achieving noteworthy results when evaluated on data more complex than the training set, such as configurations with more atoms and temperature ranges outside the training data.

Editable Co-Speech Gesture Synthesis Enhanced with Individual Representative Gestures

Co-speech gesture synthesis is a challenging task due to the complexity and uncertainty between gestures and speech. Gestures that accompany speech (i.e., Co-Speech Gesture) are an essential part of natural and efficient embodied human communication, as they work in tandem with speech to convey information more effectively. Although data-driven approaches have improved gesture synthesis, existing deep learning-based methods use deterministic modeling which could lead to averaging out predicted gestures. Additionally, these methods lack control over gesture generation such as user editing of generated results. In this paper, we propose an editable gesture synthesis method based on a learned pose script, which disentangles gestures into individual representative and rhythmic gestures to produce high-quality, diverse and realistic poses. Specifically, we first detect the time of occurrence of gestures in video sequences and transform them into pose scripts. Regression models are then built to predict the pose scripts. Next, learned pose scripts are used for gesture synthesis, while rhythmic gestures are modeled using a variational auto-encoder and a one-dimensional convolutional network. Moreover, we introduce a large-scale Chinese co-speech gesture synthesis dataset with multimodal annotations for training and evaluation, which will be publicly available to facilitate future research. The proposed method allows for the re-editing of generated results by changing the pose scripts for applications such as interactive digital humans. The experimental results show that this method generates more quality, more diverse, and realistic gestures than other existing methods.

CasCIFF: A cross-domain information fusion framework tailored for cascade prediction in social networks

The explosive growth of the Internet has elevated social networks to a pivotal role in information propagation, reshaping conventional paradigms of information distribution. Utilizing the vast amount of data available online to model and predict this diffusion is of great importance in various fields. Existing approaches for information cascade prediction fall into three main categories: feature-driven methods, point process-based methods, and deep learning-based methods. Among them, deep learning-based methods, characterized by their superior learning and representation capabilities, mitigate the shortcomings inherent in the other methods.However, current deep learning methods still face several persistent challenges. In particular, accurate representation of user attributes remains problematic due to factors such as fake followers and complex network configurations. Previous algorithms that focused on the sequential order of user activations often neglected the rich insights offered by activation timing. Furthermore, these techniques often fail to holistically integrate temporal and structural aspects, thus missing the nuanced propagation trends inherent in information cascades.To address these issues, we propose the Cross-Domain Information Fusion Framework (CasCIFF), which is tailored for information cascade prediction. This framework exploits multi-hop neighborhood information to make user embeddings robust. When embedding cascades, the framework intentionally incorporates timestamps, endowing it with the ability to capture evolving patterns of information diffusion. In particular, CasCIFF seamlessly integrates the tasks of user classification and cascade prediction into a consolidated framework, thereby allowing the extraction of common features that prove useful for all tasks, a strategy anchored in the principles of multi-task learning. After extensive experiments conducted on publicly available datasets, the results demonstrate CasCIFF's superiority over established baseline methods in terms of prediction accuracy.

Automatic detection of earthquake triggered landslides using Sentinel-1 SAR imagery based on deep learning

ABSTRACT Earthquake Triggered Landslides (ETLs) are serious secondary hazards of earthquakes, causing severe casualties and property losses, and their rapid, automated and accurate detection is of great value. Due to the all-day and all-weather imaging capability of Synthetic Aperture Radar (SAR), ETL detection using SAR images is promising but faces the problem of insufficient accuracy. In this study, we proposed a deep learning-based ETL detection method to address this problem. Firstly, published ETL inventories and SAR images are combined to generate high-quality training datasets. Then, a landslide detection network of Multi-level Features Effective Weighting and Fusion (MFEWF) is proposed to effectively extract and fuze multi-level landslide features to identify SAR pixels within ETLs. Finally, the ETL boundary is determined based on these identified SAR pixels. This method is verified through three earthquake cases: the 2017 Mainling, China earthquake, the 2018 Palu, Indonesia earthquake and the 2018 Papua New Guinea earthquake. Results show that our method can effectively identify landslide boundaries with high accuracy (88.8%, 81.4% and 82.4% for the three cases), obviously outperforming other deep learning frameworks (e.g. DeepLabV3+). Using Sentinel-1 imagery to achieve such high accuracy in landslide detection, this study will improve emergency response to landslide disasters following earthquakes.