Search

The maturity of 5G technology provides a guarantee for increasingly large communication networks, while the resources required for communication and computation are also increasing, and reasonable resource allocation can improve the efficiency of network communication and reduce the consumption of communication resources. Existing deep learning methods have been able to predict network traffic to a certain extent, so as to solve the communication efficiency and resource consumption problems in the field of integrated sensing, communication and computation (ISCC) through rational resource allocation. However, the following problems still exist: (1) The feature learning ability of the prediction model is insufficient, and the prediction accuracy needs to be improved. (2) Powerful and complex deep learning methods lead to an increase in the prediction cost of the model. To address these problems, this paper proposes a deep learning method based on a lightweight hybrid attention network. In order to capture the key features of 5G data more effectively, an efficient hybrid attention mechanism (EHA) is proposed. After this attention is applied to convolution, the key information can be well enhanced. We use depthwise separable convolution in feature extraction, which greatly improves the efficiency of lightweight convolution layer (LC) in feature extraction. Combined with the efficient hybrid attention mechanism (EHA), the proposed model has better lightweight properties. Experimental results show that the model proposed in this paper has lower RMSE and MAE values on the three datasets, as well as fewer parameters and computational effort compared to the baseline scheme. [ABSTRACT FROM AUTHOR]

• The paper presents a lightweight model with satisfactory performance in complex industrial noise environments. • Data fusion using data from multiple sensors for more complete fault information. • Design a unique grayscale feature map based on IMF components to obtain multi-sensor and multi-frequency band fault features. • Design a UL-GoogLeNet lightweight fault diagnosis model based on GoogLeNet and ULSAM. As artificial intelligence and sensor technology develop rapidly, intelligent fault diagnosis methods based on deep learning are widely used in industrial production. However, in practical industrial applications, the complex noise environment affects the performance of the diagnostic model, and the huge model parameters cannot meet the requirements of low cost and high performance in industrial production. To address the above problems, this paper proposes a lightweight intelligent fault diagnosis model using multi-sensor data fusion that not only meets the lightweight requirements of "small, light, and fast", but also realizes high accuracy diagnosis in noisy environments. Firstly, the vibration signals from different sensors of rolling bearings are processed using the variational mode decomposition (VMD) to design a unique method of constructing grayscale feature maps based on each intrinsic modal function (IMF) component. Then, the ultra-lightweight GoogLeNet model (UL-GoogLeNet) is constructed to adjust the traditional GoogLeNet structure, while the Ultra-lightweight subspace attention module (ULSAM) is introduced to reduce the model parameters and enhance the feature extraction capability. UL-GoogLeNet is trained and tested by dividing the grayscale feature maps into training and testing sets to realize the intelligent recognition of different fault types in rolling bearings. Experiments are conducted on two datasets and compared with multiple methods, and the final experimental results prove the effectiveness and superiority of the proposed method in this paper. [ABSTRACT FROM AUTHOR]

Recently, deep learning methods have achieved excellent performance in single image super-resolution (SISR), but most existing methods suffer from heavy computational costs and memory storage. To address this problem, some lightweight SR methods have been proposed, among which convolutional neural network (CNN) with attention mechanisms has received increasing attention. However, the existing CNN-based lightweight network and attention mechanism still have redundancy in feature extraction and enhancement. This paper presents Efficient Self-calibrated and Hierarchical Refinement Network (ESHR) for Lightweight Super-Resolution that includes three effective designs. First, we propose a more effective feature extraction basic module called Self-calibrated Residual Block (SRB), which can achieve superior performance when replacing other common convolution modules such as traditional convolution, BSConv and DSConv. Second, we propose a more lightweight spatial attention mechanism called Squeeze-Excitation Spatial Attention (SESA), which can interact and enhance features more effectively in spatial dimensions to improve the representation ability of the model. Third, we design an Efficient Hierarchical Refinement Block (EHRB) to integrate and utilize multi-level features, which can achieve better performance than single-level refinement without increasing the number of parameters. Extensive experiments demonstrate that the proposed method outperforms state-of-the-art CNN-based methods quantitatively and qualitatively. [ABSTRACT FROM AUTHOR]

Glaucoma is a prevalent eye disease that is responsible for causing blindness worldwide. To diagnose glaucoma, the cup-to-disc ratio (CDR) is an important factor. We introduce MRSNet, a novel segmentation network that incorporates encoding and decoding structures. The key innovation of this network is the application of convolutional block with large kernel convolutional attention to the task of medical image segmentation for optic cup and disc. By combining the benefits of residual and self-attention, our network achieves improved performance. The coding region of the network utilizes convolutional block with large kernel convolutional attention, enabling the extraction of multi-scale features with lower computational resources while also enhancing spatial attention. The self-attention layer acts as a transition between the encoding and decoding regions, capturing long connection information and providing additional image details. To further enhance segmentation performance, we employ a multi-resolution image combination approach and adaptively extract the input form using the compression and excitation module. Additionally, we propose a novel approach that combines the principle of consistency of deep supervision mechanisms with cross-entropy and Dice loss to guide the network towards accurate segmentation. In this study, we utilized a five-fold cross-validation method to train our network model. We then performed experimental validation and evaluation on three widely-used datasets, namely REFUGE, DRISHTI-GS, and RIM-ONE-r3. Our model achieved impressive results in the cup-to-disc ratio metric, which accurately reflects the segmentation effect. Specifically, we achieved scores of 0.0242, 0.0941, and 0.0158 for the aforementioned datasets, respectively. These scores outperformed some current classical algorithms. The experimental results demonstrate that the method proposed in this paper has the capability to extract more comprehensive information about the optic cup and disc, with the ability to generalize across different datasets. Furthermore, it shows that the convolutional block with large kernel convolutional attention module can be effectively utilized for the segmentation task of optic cup and disc. These findings provide a valuable research foundation for future researchers. [ABSTRACT FROM AUTHOR]

With the popularity of online shopping and the rise of various brands, logo detection is gradually coming to the forefront of researchers' minds. However, accurately detecting multiscale, similar, diverse and shape-shifting logos poses a challenge for this technology. The Swin transformer has created a milestone as a high-performance deep learning method across modalities, domains and tasks in various tasks of computer vision. We improve the Swin transformer for better image feature extraction and enhance the robustness for its application to logo detection. In this paper, we propose the PMST as the backbone of logo detection, and design a bypass-parallelizable shift module and a miniature window tandem shift strategy to further enhance image feature fusion and transfer between windows. The PMST achieves a 79.2 box AP on the LogoData dataset, surpassing most detection models. According to existing known methods, in logo detection, the PMST is the first backbone to put transformer to the detection. The results achieved state-of-the-art on the FlickrLogos-32 and FoodLogoDet-1500 datasets. It also achieves excellent efficiency on the ImageNet-1K, OpenBrands-80 and MS-COCO datasets. The code is available at https://github.com/blowhen/PMST. [ABSTRACT FROM AUTHOR]

Abstract The brain-like functionality of the artificial neural networks besides their great performance in various areas of scientific applications, make them a reliable tool to be employed in Audio-Visual Speech Recognition (AVSR) systems. The applications of such networks in the AVSR systems extend from the preliminary stage of feature extraction to the higher levels of information combination and speech modeling. In this paper, some carefully designed deep autoencoders are proposed to produce efficient bimodal features from the audio and visual stream inputs. The basic proposed structure is modified in three proceeding steps to make better usage of the presence of the visual information from the speakers' lips Region of Interest (ROI). The performance of the proposed structures is compared to both the unimodal and bimodal baselines in a professional phoneme recognition task, under different noisy audio conditions. This is done by employing a state-of-the-art DNN-HMM hybrid as the speech classifier. In comparison to the MFCC audio-only features, the finally proposed bimodal features cause an average relative reduction of 36.9% for a range of different noisy conditions, and also, a relative reduction of 19.2% for the clean condition in terms of the Phoneme Error Rates (PER). [ABSTRACT FROM AUTHOR]

Signal recognition technology is a currently active area in both civilian and military applications. Recently, deep learning has aroused extensive attempts in radar signal recognition due to its remarkable capability of automatic feature extraction. However, existing radar signal recognition networks overly depend on the probability-based decision model, resulting in poor robustness. This paper develops a novel deep metric learning frame to enhance the robustness of the recognition system. First, a multiscale atrous pyramid network (MAPNet) is proposed to efficiently learn high-resolution and distinct feature representation. Second, a variance loss is designed to constrain the intra-class feature distribution in metric space. Third, according to the distribution of training signals in metric space, recognition results are recalibrated to provide explicit rejection probabilities for unknowns. Extensive experiments and evaluations demonstrate that the proposed model can accurately classify known signals while robustly identifying unknown signals. The signal database and model can be freely accessed at https://github.com/bryantky/MAPNet. [ABSTRACT FROM AUTHOR]

Recent deep learning-based approaches on facial expression recognition face challenges on achieving optimal solution due to ineffective extraction of semantic expression features and difficulty in perceiving the positional relationship among facial expression features. In this paper we propose a novel framework, named Ventral-Dorsal Attention Capsule Network (VDACaps) to address the challenges. VDACaps adopts ResNet18 with a ventral channel attention and dilated spatial attention mechanism to strengthen the feature extraction. A novel attention layer containing multiple groups of ventral and dorsal attention is also designed to focus on the key features of facial expressions at the channel and spatial levels, exploring optimal sizes of receptive fields. Experiments on three well-known public expression datasets (CK+, JAFFE, and SFEW) have demonstrated that VDACaps achieves better or competitive performance on facial expression recognition against the state-of-the-art, with the accuracy of 98.98%, 98.44% and 54.07% respectively. [ABSTRACT FROM AUTHOR]

Deep learning-based voiceprint recognition methods rely heavily on adequate datasets, especially those closer to the natural environment and more complex under unconstrained conditions. Yet, the data types of open-source speech datasets are too homogeneous nowadays, and there are some differences with the address collected in natural application environments. For few Chinese datasets used, this paper proposes and produces an unconstrained Chinese speech dataset with richer data types closer to those collected in a natural environment. To address the inadequate extraction of acoustic features in the unconstrained speech dataset, a new two-dimensional convolutional residual network structure based on the attention mechanism is designed and applied to acoustic feature extraction. The residual block structure in the residual network is improved by the SE module and the CBAM module to obtain the SE-Cov2d and CSA-Cov2d models respectively. Finally, it is experimentally demonstrated that the attention mechanism can help the network focus on more critical feature information and fuse more differentiated features in feature extraction. • An unconstrained Chinese speech dataset proposed in a natural environment. • A new two-dimensional convolutional residual network structure designed and applied to acoustic feature extraction. • SE-Cov2d obtains the smallest EER values of 2.72% and 6.76% on both VoxCeleb and CN-Human datasets. [ABSTRACT FROM AUTHOR]

Radar specific emitter identification (SEI) plays important roles in spectrum resource management and Internet of Things (IoT) security. This term refers to the identification of different individual devices of the same type by analyzing the unintentional modulation features caused by hardware imperfections (i.e., fingerprint features). Increasingly many researchers are using deep learning (DL)-based methods for SEI. However, DL-based methods still have many problems, including model generalization, recognition efficiency, and adaptability in open electromagnetic environments. This paper proposes a one-dimensional convolutional neural network (1D CNN)—called a selective kernel residual network (SKResNet)—capable of autonomously adjusting the receptive field, to improve the feature extraction capability. The transient part is cut efficiently from a radar signal sequence as the input of SKResNet using a universal preprocessing method. In addition, considering that most existing methods obtain incorrect results while identifying new emitters without any prior knowledge, SKResNet was improved using OpenMax—an open-set recognition method—to develop Open-SKResNet. As this network could identify unknown emitters effectively, generating reliable classification results of known emitters simultaneously, it could be of great importance and have broad application prospects. [ABSTRACT FROM AUTHOR]

Automatic modulation recognition (AMR) detects the modulation scheme of the received signals for further signal processing without needing prior information, and provides the essential function when such information is missing. Recent breakthroughs in deep learning (DL) have laid the foundation for developing high-performance DL-AMR approaches for communications systems. Comparing with traditional modulation detection methods, DL-AMR approaches have achieved promising performance including high recognition accuracy and low false alarms due to the strong feature extraction and classification abilities of deep neural networks. Despite the promising potential, DL-AMR approaches also bring concerns to complexity and explainability, which affect the practical deployment in wireless communications systems. This paper aims to present a review of the current DL-AMR research, with a focus on appropriate DL models and benchmark datasets. We further provide comprehensive experiments to compare the state of the art models for single-input-single-output (SISO) systems from both accuracy and complexity perspectives, and propose to apply DL-AMR in the new multiple-input-multiple-output (MIMO) scenario with precoding. Finally, existing challenges and possible future research directions are discussed. [ABSTRACT FROM AUTHOR]

• A deep-learning method for classification of the audio codecs is proposed. • The effect of variable bit-rates in the classification of fragments is analyzed. • Longest common substrings and subsequences can effectively classify audio codecs. • A large set of speech codecs with 20 codecs and 96 bit-rates is considered. A large portion of the Internet bandwidth is used for transmission of multimedia such as audio data. As the file sizes are usually much bigger than the maximum network packet size, the audio data are segmented into fragments. For eavesdropping or network surveillance purposes, the first step of a sniffer may be to determine the codec by which a fragment is generated. This problem is usually modeled as a multi-class classification problem. The basic methods for determining the codec type of each fragment rely on the metadata in the corresponding file header. However, in a non-cooperative context, the whole file is not available. Statistical features extracted from the fragments are generally used for solving this problem by employing machine learning methods. In this paper, we present an end to end scalable deep learning approach for the classification of audio codecs with variable bit-rates. This method is based on an efficient variant of convolutional neural networks, which learns hidden layer representations to encode input vectors. Moreover, by integrating the statistical features and Cascade Deep Forest method, we jointly optimize classification labels assignment and learn the features that are suitable for classification. In several experiments on a dataset of speech codecs, we demonstrate that our approach outperforms the state-of-the-art methods by a significant margin. [ABSTRACT FROM AUTHOR]

Specific emitter identification (SEI) is an important problem in the field of electronic intelligence. There are two major limitations in most existing SEI methods: First, the features should be artificially extracted, which requires specialized expertise; Second, various features are not merged effectively to improve performance. In this paper, an automatic multi-feature extraction and fusion method based on deep ensemble learning is proposed for SEI. This method extracts and fuses multiple features via data-driven strategies using convolutional neural networks (CNN). No additional constraints are required on the type and number of the features to be fused. Three typical characteristics of radar pulse signals, including amplitude, phase and spectrum asymmetry, are taken as example features for method description and performance verification. Experiment results demonstrate that, the proposed method performs well in automatic feature extraction, and it can significantly improve the SEI performance via multi-feature fusion. [ABSTRACT FROM AUTHOR]

• We propose a new data-driven filterbank for speech feature extraction in automatic speaker verification (ASV). • The proposed method uses voice frames of speech signal for computing the frequency warping scale. • We have also proposed a new method for computing filterbank frequency-responses using principal component analysis (PCA). • We have conducted experiments on NIST SRE 2001, 2002 and VoxCeleb1 with different ASV systems. • The proposed data-driven cepstral features yield improved recognition performance over baseline methods. Most of the speech processing applications use triangular filters spaced in mel-scale for feature extraction. In this paper, we propose a new data-driven filter design method which optimizes filter parameters from a given speech data. First, we introduce a frame-selection based approach for developing speech-signal-based frequency warping scale. Then, we propose a new method for computing the filter frequency responses by using principal component analysis (PCA). The main advantage of the proposed method over the recently introduced deep learning based methods is that it requires very limited amount of unlabeled speech-data. We demonstrate that the proposed filterbank has more speaker discriminative power than commonly used mel filterbank as well as existing data-driven filterbank. We conduct automatic speaker verification (ASV) experiments with different corpora using various classifier back-ends. We show that the acoustic features created with proposed filterbank are better than existing mel-frequency cepstral coefficients (MFCCs) and speech-signal-based frequency cepstral coefficients (SFCCs) in most cases. In the experiments with VoxCeleb1 and popular i-vector back-end, we observe 9.75% relative improvement in equal error rate (EER) over MFCCs. Similarly, the relative improvement is 4.43% with recently introduced x-vector system. We obtain further improvement using fusion of the proposed method with standard MFCC-based approach. [ABSTRACT FROM AUTHOR]