Robust Feature Representation Research Articles

Performance evaluation of attention-deep hashing based medical image retrieval in brain MRI datasets

BackgroundBrain MRI images pose significant challenges due to their complexity and voluminous data, which often hinder the accuracy of traditional image retrieval methods. In response, this research delves into a novel medical image retrieval approach grounded in attention mechanisms and deep hashing techniques. MethodologyThe study enhances the network's edge perception capability by incorporating a dual mixed attention module (CBAM-MA dual attention) into the convolutional neural network architecture. This addition optimally captures edge saliency and distributional features by introducing an extra attention layer atop CBAM. Furthermore, a novel triple loss function is introduced, amalgamating Euclidean and cosine distances. This fusion exploits the strengths of both distance metrics to comprehensively consider sample geometric attributes, thereby generating more robust and expressive feature representations crucial for preserving categorical details. ResultsExperimental evaluations showcase significant performance advancements in MRI image retrieval tasks using the proposed method. The average hit rate stands at 0.7783, average precision at 0.752246, and average inverse rank at 0.956721. These metrics collectively underscore the method's efficacy in enhancing the quality and expediting the diagnosis of brain diseases. ConclusionThe research underscores the critical role of attention and deep hashing techniques in augmenting the accuracy and efficiency of medical image retrieval, particularly in the context of brain MRI images. The proposed method's notable performance improvements pave the way for enhanced diagnostic capabilities, signaling a significant step forward in improving healthcare outcomes related to brain diseases.

Combining dual attention mechanism and efficient feature aggregation for road and vehicle segmentation from UAV imagery

Unmanned aerial vehicles (UAVs) have gained significant popularity in recent years due to their ability to capture high-resolution aerial imagery for various applications, including traffic monitoring, urban planning, and disaster management. Accurate road and vehicle segmentation from UAV imagery plays a crucial role in these applications. In this paper, we propose a novel approach combining dual attention mechanisms and efficient multi-layer feature aggregation to enhance the performance of road and vehicle segmentation from UAV imagery. Our approach integrates a spatial attention mechanism and a channel-wise attention mechanism to enable the model to selectively focus on relevant features for segmentation tasks. In conjunction with these attention mechanisms, we introduce an efficient multi-layer feature aggregation method that synthesizes and integrates multi-scale features at different levels of the network, resulting in a more robust and informative feature representation. Our proposed method is evaluated on the UAVid semantic segmentation dataset, showcasing its exceptional performance in comparison to renowned approaches such as U-Net, DeepLabv3+, and SegNet. The experimental results affirm that our approach surpasses these state-of-the-art methods in terms of segmentation accuracy.

Multi-granularity relationship reasoning network for high-fidelity 3D shape reconstruction

Monocular image-based 3D reconstruction is widely used in virtual reality, augmented reality, and autonomous driving, which benefits from the rapid development of deep learning approaches. Most of the available methods focused on reconstructing the overall shape of the object while ignoring some fine-grained details. Moreover, these methods make it hard to exactly reconstruct complex topological structures. In this paper, we propose a multi-granularity relationship reasoning network (MGRRNet), which aims to recover 3D shapes with high fidelity and rich details via the relationship reasoning between different granularity information. Specifically, our model captures the discriminative and detailed features at different granularities for extracting attentional regions. Then we perform the relationship reasoning between different granularities to reinforce the multi-granularity consistency and inter-granularity correlation. By doing this, our network is able to achieve robust feature representation and fine reconstruction. During the learning process, we jointly optimize procedures of different granularity feature representations via a sequence of inter-granularity cycle loss iterations. Extensive experimental results on two publicly available datasets justify that our approach achieves competitive performance compared to the state-of-the-art methods. Codes and all resources will be publicly available at https://github.com/Ray-tju/MGRRNet.

TB-Net: Intra- and inter-video correlation learning for continuous sign language recognition

Visual feature extraction is the key to continuous sign language recognition (CSLR). However, current CSLR methods based on single-branch networks only rely on intra-video correlation learning to facilitate visual feature optimization. Obviously, this is not conducive to obtaining more robust visual feature representations. Hence, we pioneered a novel CSLR method via intra- and inter-video correlation learning with a two-branch network, named TB-Net. TB-Net explicitly establishes intra-video correlation between glosses and the most relevant video clips at each branch and then introduces inter-video correlation to enhance visual feature extraction at the branch confluence. Specifically, we introduce a contrastive learning-based inter-video correlation module (IEM), which co-optimizes visual features from both branches by designing inter- and intra-video losses to enhance their generalization. In addition, we propose an intra-video correlation module (IAM) based on a gloss-guided attention feature generator to adaptively build mappings between glosses and video clips, which in turn contributes to the acquisition of preliminary gloss-guided visual features within a single video. Extensive experiments on four public CSLR benchmarks show the superior performance of our method.

Discovery of Novel Aldose Reductase Inhibitors via the Integration of Ligand-Based and Structure-Based Virtual Screening with Experimental Validation.

Aldose reductase plays a central role in diabetes mellitus (DM) associated complications by converting glucose to sorbitol, resulting in a harmful increase of reactive oxygen species (ROS) in various tissues, such as the heart, vasculature, neurons, eyes, and kidneys. We employed a comprehensive approach, integrating both ligand- and structure-based virtual screening followed by experimental validation. Initially, candidate compounds were extracted from extensive drug and chemical libraries using the DeepChem's GraphConvMol algorithm, leveraging its capacity for robust molecular feature representation. Subsequent refinement employed molecular docking and molecular dynamics (MD) simulations, which are crucial for understanding compound-receptor interactions and dynamic behavior in a simulated physiological environment. Finally, the candidate compounds were subjected to experimental validation of their biological activity using an aldose reductase inhibitor screening kit. The comprehensive approach led to the identification of a promising compound, demonstrating significant potential as an aldose reductase inhibitor. This comprehensive approach not only yields a potential therapeutic intervention for DM-related complications but also establishes an integrated protocol for drug development, setting a new benchmark in the field.

Attention-based convolution neural network for magnetic tile surface defect classification and detection

Effectively identifying surface defects in magnetic tiles has proven to be highly challenging due to limited sample availability and irrelevant background interference, which also plays a crucial role in significantly influencing the lifespan and reliability of permanent magnet motors. To address these challenges, our study draws inspiration from a comprehensive analysis of the retinal attention mechanism and proposes three guiding criteria: multi-level resolution, what to look for, and where to look at. These criteria are utilized as foundational principles to enhance the representation learning capability of designed neural network structures through the incorporation of the retinal attention mechanism. Subsequently, based on these guiding criteria, we introduce a novel convolutional retinal attention block (CRAB) to learn discriminative and robust feature representations for magnetic tile surface defect classification and detection. The proposed CRAB comprises three modules: multi-resolution module (MRM), global attention aggregation module (GAAM), and local attention aggregation module (LAAM), designed to extract discriminative and robust features by refining meaningful information and suppressing redundant ones. Comprehensive experimental results across image classification and object detection tasks demonstrate that the proposed CRAB outperforms existing methods such as SE, ECA, and CBAM, and can effectively amplify the representation power across various backbone networks, including VGG-16, GoogLeNet, ResNet-18, and ResNet-50. An evaluation on surface defect classification and detection tasks for industrial magnetic tiles further shows that CRAB achieves accuracies of 99.50% and 96.98%, respectively. These results emphasize the promising application prospects of the proposed method in detecting industrial surface defects amid expansive and inconsequential backgrounds. The code of the proposed method is available at: https://github.com/KWflyer/CRAB.

UCTNet: Uncertainty-guided CNN-Transformer hybrid networks for medical image segmentation

Transformer, born for long-range dependency establishment, has been widely studied as a complementary of convolutional neural networks (CNNs) in medical image segmentation. However, existing CNN-Transformer hybrid approaches simply pursue implicit feature fusion without considering their underlying functional overlap. Medical images typically follow stable anatomical structures, making convolution capable of handling most segmentation targets. Without differentiation, enforcing transformers to operate self-attention for all image patches would result in severe redundancy, hindering global feature extraction. In this paper, we propose a simple yet effective hybrid network named UCTNet where transformers only focus on establishing global dependency for CNN’s unreliable regions predicted through uncertainty estimation. In this way, CNN and transformer are explicitly fused to minimize functional overlap. More importantly, with fewer regions to handle, UCTNet is of better convergence to learn more robust feature representations for hard examples. Extensive experiments on publicly-available datasets demonstrate the superiority of UCTNet against the state-of-the-art approaches, achieving 89.44%, 92.91%, and 91.15% in Dice similarity coefficient on Synapse, ACDC, and ISIC2018 respectively. Furthermore, such a CNN-Transformer hybrid strategy is highly extendable to other frameworks without introducing additional computational burdens. Code is available at https://github.com/innocence0206/UCTNet.

Robust ensemble person reidentification via orthogonal fusion with occlusion handling

Occlusion remains one of the major challenges in person reidentification (ReID) due to the diversity of poses and the variation of appearances. Developing novel architectures to improve the robustness of occlusion-aware person Re-ID requires new insights, especially on low-resolution edge cameras. We propose a deep ensemble model that harnesses both CNN and Transformer architectures to generate robust feature representations. To achieve robust Re-ID without manually labeling occluded regions, we propose to take an ensemble learning-based approach derived from the analogy between arbitrarily shaped occluded regions and robust feature representation. Using the orthogonality principle, our developed deep CNN model uses masked autoencoder (MAE) and global–local feature fusion for robust person identification. Furthermore, we present a part occlusion-aware transformer capable of learning feature space that is robust to occluded regions. Experimental results are reported on several Re-ID datasets to show the effectiveness of our developed ensemble model named orthogonal fusion with occlusion handling (OFOH). Compared to competing methods, the proposed OFOH approach has achieved competent rank-1 and mAP performance.

CD-MAE: Contrastive Dual-Masked Autoencoder Pre-Training Model for PCB CT Image Element Segmentation

Element detection is an important step in the process of the non-destructive testing of printed circuit boards (PCB) based on computed tomography (CT). Compared with the traditional manual detection method, the image semantic segmentation method based on deep learning greatly improves efficiency and accuracy. However, semantic segmentation models often require a large amount of data for supervised training to generalize better model performance. Unlike natural images, the PCB CT image annotation task is more time-consuming and laborious than the semantic segmentation task. In order to reduce the cost of labeling and improve the ability of the model to utilize unlabeled data, unsupervised pre-training is a very reasonable and necessary choice. The masked image reconstruction model represented by a masked autoencoder is pre-trained on the unlabeled data, learning a strong feature representation ability by recovering the masked image, and shows a good generalization ability in various downstream tasks. In the PCB CT image element segmentation task, considering the characteristics of the image, it is necessary to use a model with strong feature robustness in the pre-training stage to realize the representation learning on a large number of unlabeled PCB CT images. Based on the above purposes, we proposed a contrastive dual-masked autoencoder (CD-MAE) pre-training model, which can learn more robust feature representation on unlabeled PCB CT images. Our experiments show that the CD-MAE outperforms the baseline model and fully supervised models in the PCB CT element segmentation task.

Local feature expansion Vision Transformer model for bearing fault diagnosis under noise environments

Vision Transformer (ViT) shows potential in bearing fault diagnosis due to its multi-head self-attention mechanism and parallel feature extraction network which are efficient to achieve the robust complete feature representation of the fault. However, its adaption to the noise interference relies on the sufficient huge amount of training samples to prepare the local features of the fault and may suffer performance degradation when only a limited number of samples are available for the model training. To combat this challenge, an improved ViT diagnosis model based on the local feature expansion, i.e., LFE-ViT, is proposed. An auxiliary feature extraction block is introduced using a local feature expansion network and works as a parallel module with the ViT encoder. Through the enlargement of the receptive field, the multi-scale local features on a high dimensional space are available upon the limited samples. Then, through a feature embedding channel, the extracted local features are transmitted to the ViT encoder. Finally, by virtue of the multi-head self-attention mechanism to capture the time sequence global information, a fault diagnosis model comprising comprehensively local and global feature information is derived. Experimental validation on the bearing fault dataset from Case Western Reserve University shows that LFE-ViT has provided a rather satisfactory diagnosis performance under limited samples and noise environment.

Deep self-taught learning framework for intrusion detection in cloud computing environment

<p>Cloud has become a target-rich environment for malicious attacks by cyber intruders. Security is a major concern and remains an obstacle to the adoption of cloud computing. The intrusion detection system (IDS) is regarded as defense-in-depth. Unfortunately, most machine learning approaches designed for cloud intrusion detection require large amounts of labeled attack samples, but in real practice, they are limited. Therefore, the key impetus of this work is to introduce self-taught learning (STL) combining stacked sparse autoencoder (SSAE) with long short-term memory (LSTM) as a candidate solution to learn the robust feature representation and efficiently improve the performance of IDS with respect to false alarm rate (FAR) and detection rate (DR). Accordingly, the proposed approach as a first step employs SSAE to achieve dimensional reduction by learning the discriminative features from network traffic. The approach adopts LSTM to recognize the intrusion with the features encoded by SSAE. To evaluate the detective performance of our model, a comprehensive set of experiments are conducted on NSL-KDD. Also, ablation experiments are conducted to show the contribution of each component of our approach. Further, the comparative analysis shows the efficacy of our approach against the existing approaches with an accuracy of 86.31%.</p>

Seizure prediction based on improved vision transformer model for EEG channel optimization

Epileptic seizures are unpredictable events caused by abnormal discharges of a patient’s brain cells. Extensive research has been conducted to develop seizure prediction algorithms based on long-term continuous electroencephalogram (EEG) signals. This paper describes a patient-specific seizure prediction method that can serve as a basis for the design of lightweight, wearable and effective seizure-prediction devices. We aim to achieve two objectives using this method. The first aim is to extract robust feature representations from multichannel EEG signals, and the second aim is to reduce the number of channels used for prediction by selecting an optimal set of channels from multichannel EEG signals while ensuring good prediction performance. We design a seizure-prediction algorithm based on a vision transformer (ViT) model. The algorithm selects channels that play a key role in seizure prediction from 22 channels of EEG signals. First, we perform a time-frequency analysis of processed time-series signals to obtain EEG spectrograms. We then segment the spectrograms of multiple channels into many non-overlapping patches of the same size, which are input into the channel selection layer of the proposed model, named Sel-JPM-ViT, enabling it to select channels. Application of the Sel-JPM-ViT model to the Boston Children’s Hospital–Massachusetts Institute of Technology scalp EEG dataset yields results using only three to six channels of EEG signals that are slightly better that the results obtained using 22 channels of EEG signals. Overall, the Sel-JPM-ViT model exhibits an average classification accuracy of 93.65%, an average sensitivity of 94.70% and an average specificity of 92.78%.

MeshCL: Towards robust 3D mesh analysis via contrastive learning

Recent years have witnessed the success of deep learning in 3D mesh analysis, including classification and segmentation. Previous works mainly rely on supervised learning, which trains models on the labeled 3D mesh data with the cross-entropy loss. However, in the supervised training, the very limited amount of 3D mesh data restricts the feature’s discriminative power and the model’s robustness. To address or at least alleviate this issue, we propose to regularize the learned representations by associating the data points in feature space via contrastive learning. To achieve this, we first explore mesh samples with diverse geometric structures, and then apply contrastive regularization into the positive and negative pairs sampled from the feature space. In this way, our method, MeshCL, can learn robust feature representations invariant to the geometric deformation for 3D manifold mesh. In detail, to augment the geometric diversity, we examine different data augmentation techniques for mesh data. Furthermore, considering the geometric characteristics of 3D mesh, we develop more efficient positive/negative pair sampling strategies for the dense prediction task. To verify the effectiveness of MeshCL, we study two typical mesh analysis tasks, classification and segmentation. Experimental results demonstrate that MeshCL not only achieves state-of-the-art performance on the two tasks mentioned above but also significantly improves the model’s robustness.

FireMatch: A semi-supervised video fire detection network based on consistency and distribution alignment

Deep learning techniques have greatly enhanced the performance of fire detection in videos. However, video-based fire detection models heavily rely on labeled data, and the process of data labeling is particularly costly and time-consuming, especially when dealing with videos. Considering the limited quantity of labeled video data, we propose a semi-supervised fire detection model called FireMatch, which is based on consistency regularization and adversarial distribution alignment. Specifically, we first combine consistency regularization with pseudo-label. For unlabeled data, we design video data augmentation to obtain corresponding weakly augmented and strongly augmented samples. The proposed model predicts weakly augmented samples and retains pseudo-label above a threshold, while training on strongly augmented samples to predict these pseudo-labels for learning more robust feature representations. Secondly, we generate video cross-set augmented samples by adversarial distribution alignment to expand the training data and alleviate the decline in classification performance caused by insufficient labeled data. Finally, we introduce a fairness loss to help the model produce diverse predictions for input samples, thereby addressing the issue of high confidence with the non-fire class in fire classification scenarios. The FireMatch achieved an accuracy of 76.92% and 91.80% on two real-world fire datasets, respectively. The experimental results demonstrate that the proposed method outperforms the current state-of-the-art semi-supervised classification methods.

Research on an Enhanced Multimodal Network for Specific Emitter Identification

Specific emitter identification (SEI) refers to the task of distinguishing similar emitters, especially those of the same type and transmission parameters, which is one of the most critical tasks of electronic warfare. However, SEI is still a challenging task when a feature has low physical representation. Feature representation largely determines the recognition results. Therefore, this article expects to move toward robust feature representation for SEI. Efficient multimodal strategies have great potential for applications using multimodal data and can further improve the performance of SEI. In this research, we introduce a multimodal emitter identification method that explores the application of multimodal data, time-series radar signals, and feature vector data to an enhanced transformer, which employs a conformer block to embed the raw data and integrates an efficient multimodal feature representation module. Moreover, we employ self-knowledge distillation to mitigate overconfident predictions and reduce intra-class variations. Our study reveals that multimodal data provide sufficient information for specific emitter identification. Simultaneously, we propose the CV-CutMixOut method to augment the time-domain signal. Extensive experiments on real radar datasets indicate that the proposed method achieves more accurate identification results and higher feature discriminability.

Heterogeneous fusion of biometric and deep physiological features for accurate porcine cough recognition.

Accurate identification of porcine cough plays a vital role in comprehensive respiratory health monitoring and diagnosis of pigs. It serves as a fundamental prerequisite for stress-free animal health management, reducing pig mortality rates, and improving the economic efficiency of the farming industry. Creating a representative multi-source signal signature for porcine cough is a crucial step toward automating its identification. To this end, a feature fusion method that combines the biological features extracted from the acoustic source segment with the deep physiological features derived from thermal source images is proposed in the paper. First, acoustic features from various domains are extracted from the sound source signals. To determine the most effective combination of sound source features, an SVM-based recursive feature elimination cross-validation algorithm (SVM-RFECV) is employed. Second, a shallow convolutional neural network (named ThermographicNet) is constructed to extract deep physiological features from the thermal source images. Finally, the two heterogeneous features are integrated at an early stage and input into a support vector machine (SVM) for porcine cough recognition. Through rigorous experimentation, the performance of the proposed fusion approach is evaluated, achieving an impressive accuracy of 98.79% in recognizing porcine cough. These results further underscore the effectiveness of combining acoustic source features with heterogeneous deep thermal source features, thereby establishing a robust feature representation for porcine cough recognition.

Unsupervised Joint Contrastive Learning for Aerial Person Re-Identification and Remote Sensing Image Classification

Unsupervised person re-identification (Re-ID) aims to match the query image of a person with images in the gallery without the use of supervision labels. Most existing methods usually generate pseudo-labels through clustering algorithms for contrastive learning, which inevitably results in noisy labels assigned to samples. In addition, methods that only apply contrastive learning at the clustering level fail to fully consider instance-level relationships between instances. Motivated by this, we propose a joint contrastive learning (JCL) framework for unsupervised person Re-ID. Our proposed method involves creating two memory banks to store features of cluster centroids and instances and applies cluster and instance-level contrastive learning, respectively, to jointly optimize the neural networks. The cluster-level contrastive loss is used to promote feature compactness within the same cluster and reinforce identity similarity. The instance-level contrastive loss is used to distinguish easily confused samples. In addition, we use a WaveBlock attention module (WAM), which can continuously wave feature map blocks and introduce attention mechanisms to produce more robust feature representations of a person without considerable information loss. Furthermore, we enhance the quality of our clustering by leveraging camera label information to eliminate clusters containing single camera captures. Extensive experimental results on two widely used person Re-ID datasets verify the effectiveness of our JCL method. Meanwhile, we also used two remote sensing datasets to demonstrate the generalizability of our method.

Robust Feature Representation Using Multi-Task Learning for Human Activity Recognition.

Learning underlying patterns from sensory data is crucial in the Human Activity Recognition (HAR) task to avoid poor generalization when coping with unseen data. A key solution to such an issue is representation learning, which becomes essential when input signals contain activities with similar patterns or when patterns generated by different subjects for the same activity vary. To address these issues, we seek a solution to increase generalization by learning the underlying factors of each sensor signal. We develop a novel multi-channel asymmetric auto-encoder to recreate input signals precisely and extract indicative unsupervised futures. Further, we investigate the role of various activation functions in signal reconstruction to ensure the model preserves the patterns of each activity in the output. Our main contribution is that we propose a multi-task learning model to enhance representation learning through shared layers between signal reconstruction and the HAR task to improve the robustness of the model in coping with users not included in the training phase. The proposed model learns shared features between different tasks that are indeed the underlying factors of each input signal. We validate our multi-task learning model using several publicly available HAR datasets, UCI-HAR, MHealth, PAMAP2, and USC-HAD, and an in-house alpine skiing dataset collected in the wild, where our model achieved 99%, 99%, 95%, 88%, and 92% accuracy. Our proposed method shows consistent performance and good generalization on all the datasets compared to the state of the art.

An enhanced Swin Transformer for soccer player reidentification

The re-identification (ReID) of objects in images is a widely studied topic in computer vision, with significant relevance to various applications. The ReID of players in broadcast videos of team sports is the focus of this study. We specifically focus on identifying the same player in images taken at any given moment during a game from various camera angles. This work varies from other person ReID apps since the same team wears very similar clothes, there are few samples for each identification, and image resolutions are low. One of the hardest parts of object ReID is robust feature representation extraction. Despite the great success of current convolutional neural network-based (CNN) methods, most studies only consider learning representations from images, neglecting long-range dependency. Transformer-based model studies are increasing and yielding encouraging results. Transformers still have trouble extracting features from small objects and visual cues. To address these issues, we enhanced the Swin Transformer with the levering of CNNs. We created a regional feature extraction Swin Transformer (RFES) backbone to increase local feature extraction and small-scale object feature extraction. We also use three loss functions to handle imbalanced data and highlight challenging situations. Re-ranking with k-reciprocal encoding was used in this study's retrieval phase, and its assessment findings were provided. Finally, we conducted experiments on the Market-1501 and SoccerNet-v3 ReID datasets. Experimental results show that the proposed re-ID method reaches rank-1 accuracy of 96.2% with mAP: 89.1 and rank-1 accuracy of 84.1% with mAP: 86.7 on the Market-1501 and SoccerNet-v3 datasets, respectively, outperforming the state-of-the-art approaches.

Deep causal speech enhancement and recognition using efficient long-short term memory Recurrent Neural Network.

Long short-term memory (LSTM) has been effectively used to represent sequential data in recent years. However, LSTM still struggles with capturing the long-term temporal dependencies. In this paper, we propose an hourglass-shaped LSTM that is able to capture long-term temporal correlations by reducing the feature resolutions without data loss. We have used skip connections in non-adjacent layers to avoid gradient decay. In addition, an attention process is incorporated into skip connections to emphasize the essential spectral features and spectral regions. The proposed LSTM model is applied to speech enhancement and recognition applications. The proposed LSTM model uses no future information, resulting in a causal system suitable for real-time processing. The combined spectral feature sets are used to train the LSTM model for improved performance. Using the proposed model, the ideal ratio mask (IRM) is estimated as a training objective. The experimental evaluations using short-time objective intelligibility (STOI) and perceptual evaluation of speech quality (PESQ) have demonstrated that the proposed model with robust feature representation obtained higher speech intelligibility and perceptual quality. With the TIMIT, LibriSpeech, and VoiceBank datasets, the proposed model improved STOI by 16.21%, 16.41%, and 18.33% over noisy speech, whereas PESQ is improved by 31.1%, 32.9%, and 32%. In seen and unseen noisy situations, the proposed model outperformed existing deep neural networks (DNNs), including baseline LSTM, feedforward neural network (FDNN), convolutional neural network (CNN), and generative adversarial network (GAN). With the Kaldi toolkit for automated speech recognition (ASR), the proposed model significantly reduced the word error rates (WERs) and reached an average WER of 15.13% in noisy backgrounds.