Deep Learning Scheme Research Articles

Effective Capacity Maximization in beyond 5G Vehicular Networks: A Hybrid Deep Transfer Learning Method

How to improve delay‐sensitive traffic throughput is an open issue in vehicular communication networks, where a great number of vehicle to infrastructure (V2I) and vehicle to vehicle (V2V) links coexist. To address this issue, this paper proposes to employ a hybrid deep transfer learning scheme to allocate radio resources. Specifically, the traffic throughput maximization problem is first formulated by considering interchannel interference and statistical delay guarantee. The effective capacity theory is then applied to develop a power allocation scheme on each channel reused by a V2I and a V2V link. Thereafter, a deep transfer learning scheme is proposed to obtain the optimal channel assignment for each V2I and V2V link. Simulation results validate that the proposed scheme provides a close performance guarantee compared to a globally optimal scheme. Besides, the proposed scheme can guarantee lower delay violation probability than the schemes aiming to maximize the channel capacity.

Classification of Positive COVID-19 CT Scans using Deep Learning

In medical imaging, computer vision researchers are faced with a variety of features for verifying the authenticity of classifiers for an accurate diagnosis. In response to the coronavirus 2019 (COVID-19) pandemic, new testing procedures, medical treatments, and vaccines are being developed rapidly. One potential diagnostic tool is a reverse-transcription polymerase chain reaction (RT-PCR). RT-PCR, typically a time-consuming process, was less sensitive to COVID-19 recognition in the disease’s early stages. Here we introduce an optimized deep learning (DL) scheme to distinguish COVID-19-infected patients from normal patients according to computed tomography (CT) scans. In the proposed method, contrast enhancement is used to improve the quality of the original images. A pretrained DenseNet-201 DL model is then trained using transfer learning. Two fully connected layers and an average pool are used for feature extraction. The extracted deep features are then optimized with a Firefly algorithm to select the most optimal learning features. Fusing the selected features is important to improving the accuracy of the approach; however, it directly affects the computational cost of the technique. In the proposed method, a new parallel high index technique is used to fuse two optimal vectors; the outcome is then passed on to an extreme learning machine for final classification. Experiments were conducted on a collected database of patients using a 70:30 training: Testing ratio. Our results indicated an average classification accuracy of 94.76% with the proposed approach. A comparison of the outcomes to several other DL models demonstrated the effectiveness of our DL method for classifying COVID-19 based on CT scans.

BEHAVE: Behavior-Aware, Intelligent, and Fair Resource Management for Heterogeneous Edge-IoT Systems

Data-driven approaches are envisioned to build future Edge-IoT systems that satisfy IoT devices demands for edge resources. However, significant challenges and technical barriers exist which complicate resource management of such systems. IoT devices can demonstrate a wide range of behaviors in the devices resource demand that are extremely difficult to manage. In addition, the management of resources fairly and efficiently by the edge in such a setting is a challenging task. In this paper, we develop a novel data-driven resource management framework named BEHAVE that intelligently and fairly allocates edge resources to IoT devices with consideration of their behavior of resource demand (BRD). BEHAVE aims to holistically address the management technical barriers by 1) building an efficient scheme for modeling and assessment of the BRD of IoT devices based on their resource requests and resource usage; 2) expanding a new Rational, Fair, and Truthful Resource Allocation (RFTA) model that binds the devices BRD and resource allocation to achieve fair allocation and encourage truthfulness in resource demand; and 3) developing an enhanced deep reinforcement learning (EDRL) scheme to achieve the RFTA goals. The evaluation results demonstrate BEHAVE's capability to analyze the IoT devices BRD and adjust its resource management policy accordingly.

DEEP LEARNING TECHNIQUE BASED TEXT EMOTION CLASSIFICATION SYSTEM USING GENETIC ALGORITHM TECHNIQUE

In today’s era the research area of emotion detection has becomes trendy field of research. The data in text form is the very easy way to communicate among interaction of human-machine through the social networking sites, is one of the schemes to share users views. Recognition of human’s emotion through analyzing textual documents is useful and essential, but sometimes difficult because of the fact that it is not necessary to use emotion words directly in textual expressions. In this research, a deep learning technique based Text Emotion Classification (TEC) System using Genetic Algorithm (GA) as an optimization technique is presented. Initially, a lexicon dictionary is prepared based on the emotional words and different processes such as pre-processing, feature extraction, optimization and classification has been applied to classify the textual emotion. The test data has been trained using deep learning scheme named as Deep Neural Network (DNN) with optimization technique based on GA with a novel fitness function. The emotion; happy, sad and angry are identified as per the shared data on the social media platform. Most of the state-of-the-art in the previous research on textual emotion mining is mainly focused on without utilizing the feature selection concept, so we introduce the concept of feature selection using the GA and passes an input to DNN. At the last, we compare the performance of the proposed TEC system with existing work proposed by Chatterjee et al. terms of precision, recall and F-measure and we observed that the system got better emotion classification accuracy.

A Reliable Deep Learning Scheme for Nonlinear Reconstructions in Electrical Impedance Tomography

Significant progress has recently been made in applying deep learning methods to Electrical Impedance Tomography (EIT), which is a promising technique for non-invasive, bedside ionizing radiation free, and real-time monitoring of lung health. However, different with conventional methods solved from physical model, deep learning methods, as data-driven approaches, suffer from reliability problem, i.e., the “confidence level” is unknown when using deep learning methods as EIT solvers. In this work, a reliable deep learning scheme (RDLS) is proposed to deal with typical nonlinear EIT problems. A noticeable characteristic of the proposed RDLS is that, besides the reconstructions, pixel-based uncertainties of the results are also predicted with a Bayesian framework. Further, physical information based on a spectral analysis and back-propagated field are incorporated into the proposed RDLS to improve the robustness. By training with random inclusions, it is quantitatively shown that the proposed RDLS provides high-quality reconstructions and uncertainty qualifications, where the predicted uncertainties are verified by both linear and nonlinear correlations with true absolute errors that are only known when ground truth is known. Our approach provides fast, high-quality, and robust EIT imaging with pixel-based uncertainties in both numerical and lab-control experimental tests.

Deep Super-Resolution Network for rPPG Information Recovery and Noncontact Heart Rate Estimation

Remote photoplethysmography (rPPG) enables noncontact heart rate (HR) estimation using facial videos and exhibits significant convenience over traditional contact-based HR measurement approaches. However, its effectiveness depends mainly on the number of pixels of the facial region; thus, the estimation accuracy is easily degraded due to missing rPPG information when the resolution of facial images is low. Given that, in most real application scenarios, such as group-oriented and long-distance vital signs measurement, only low-resolution facial videos are available, finding a solution to recover rPPG information is a key to improve the performance of noncontact HR measurement. In this article, we propose a two-stage deep learning scheme to achieve accurate HR measurement from low-resolution facial videos. In the first stage, an rPPG information recovery network is proposed to recover the rPPG information during video super-resolution (SR) processing. The up-scaled images are then fed into a temporally aware HR measurement network in the second stage. An attention mechanism that reassigns the weights of the temporal information in each feature channel is designed to improve the measurement accuracy. The mean absolute percentage error achieves 3.320 and 4.231, while Pearson’s correlation coefficient reaches 0.930 and 0.892 on two datasets, respectively. Experimental results show that rPPG information could be effectively recovered through SR processing. Compared with the state-of-the-art approaches that treat high-resolution images as the input, the proposed method achieved closely comparable measurement performance using merely low-resolution images.

Multi-Stream Refining Network for Person Re-Identification

Viewpoint change, pose variation and background clutter have adverse impacts on similarity evaluation for person re-identification. Because of its distinction and reliability, person saliency has been applied to model person appearance characteristics. However, such valuable information is not fully exploited to compute similarities of person images with existing deep methods. To this end, we present a novel multi-stream refining based deep multi-task learning scheme that aggregates multi-stage salient embedding features in the network to boost the retrieval performance. Specifically, the backbone network is divided into four stages and a channel significance self-learning sub-module is introduced to strengthen the importance of saliency channels adaptively. Meanwhile, an enhancement sub-module is employed to extract the common information and different information from the channels. Finally, a multi-stream multi-task learning framework combining four-stage branches is adopted to learn discriminative features. Compared with the state-of-the-art approaches, our model achieves competitive performance on three publicly available datasets, i.e., Market-1501, MSMT17, and CUHK03. The experimental results demonstrate the superiority of our method, which achieves 95.67%/88.51%, 87.53%/65.54%, and 89.32%/78.99% on Rank-1/mAP, respectively.

Hybrid optimisation dependent deep belief network for lane detection

ABSTRACT Nowadays, in research introducing an advanced driver assistance system for improving driving is considered as the trending one. In this research, concentrate more on proposing lane detection model and assist in driving. This work develops a lane detection model through the deep learning scheme. The proposed scheme has two major phases, such as image transformation and lane detection. Initially, the proposed method obtains the multiple-lane images and transforms the image and this image transformation helps in classifier training. For detecting the lane from the bird’s view image, this work considers the Deep Convolution Neural Network (DCNN) classifier. A novel optimisation algorithm, namely Earth Worm-Crow Search Algorithm (EW-CSA), is developed in this work to assist the DCNN classifier with the optimal weights. The proposed algorithm is developed by modifying the Earth Worm Optimisation algorithm (EWA) with the properties of the Crow Search Algorithm (CSA). The proposed system is compared with other existing methods, in which the proposed method offers maximum sensitivity 0.9925, the detection accuracy of 0.99512, and specificity of 0.995.

A deep reinforcement learning based multi-criteria decision support system for optimizing textile chemical process

Textile manufacturing is a typical traditional industry involving high complexity in interconnected processes with limited capacity on the application of modern technologies. Decision-making in this domain generally takes multiple criteria into consideration, which usually arouses more complexity. To address this issue, the present paper proposes a decision support system that combines the intelligent data-based models of random forest (RF) and a human knowledge-based multi-criteria structure of analytical hierarchical process (AHP) in accordance with the objective and the subjective factors of the textile manufacturing process. More importantly, the textile chemical manufacturing process is described as the Markov decision process (MDP) paradigm, and a deep reinforcement learning scheme, the Deep Q-networks (DQN), is employed to optimize it. The effectiveness of this system has been validated in a case study of optimizing a textile ozonation process, showing that it can better master the challenging decision-making tasks in textile chemical manufacturing processes.

Situation-Aware Resource Allocation for Multi-Dimensional Intelligent Multiple Access: A Proactive Deep Learning Framework

To meet the ever-increasing communication services with diverse requirements, situation-aware intelligent utilization of multi-dimensional communication resources is becoming essential. In this paper, considering a time-division-duplex downlink cellular scenario, a deep learning-based framework for multi-dimensional intelligent multiple access (MD-IMA) scheme is developed for beyond 5G and 6G wireless networks to meet the real-time and diverse quality of service (QoS) requirements by fully utilizing the available radio resources in heterogeneous domains. To achieve intelligent operation of MD-IMA, the proposed deep learning scheme is achieved based on the convergence of long short term memory (LSTM) and deep reinforcement learning (DRL). Specifically, an LSTM neural network is used to predict the long-term network dynamics and inference changes in QoS requirements of the MD-IMA. Meanwhile, a deterministic policy gradient (DDPG) algorithm, a model-free DRL technique, is adopted to optimize the multi-dimensional radio resource allocation in real-time by dynamically following the fluctuations of the network situation. With the aid of the DDPG algorithm, radio resource management for MD-IMA can be achieved efficiently with reduced processing latency as compared to the conventional model-based approaches. Furthermore, the effectiveness of our proposed deep learning framework for MD-IMA is validated through real-world cellular traffic data-sets. The experimental results demonstrate that the proposed scheme can outperform state-of-the-art algorithms.

LSTM-Based Emotion Detection Using Physiological Signals: IoT Framework for Healthcare and Distance Learning in COVID-19.

Human emotions are strongly coupled with physical and mental health of any individual. While emotions exbibit complex physiological and biological phenomenon, yet studies reveal that physiological signals can be used as an indirect measure of emotions. In unprecedented circumstances alike the coronavirus (Covid-19) outbreak, a remote Internet of Things (IoT) enabled solution, coupled with AI can interpret and communicate emotions to serve substantially in healthcare and related fields. This work proposes an integrated IoT framework that enables wireless communication of physiological signals to data processing hub where long short-term memory (LSTM)-based emotion recognition is performed. The proposed framework offers real-time communication and recognition of emotions that enables health monitoring and distance learning support amidst pandemics. In this study, the achieved results are very promising. In the proposed IoT protocols (TS-MAC and R-MAC), ultralow latency of 1 ms is achieved. R-MAC also offers improved reliability in comparison to state of the art. In addition, the proposed deep learning scheme offers high performance (n}{}fn-score) of 95%. The achieved results in communications and AI match the interdependency requirements of deep learning and IoT frameworks, thus ensuring the suitability of proposed work in distance learning, student engagement, healthcare, emotion support, and general wellbeing.

Image Descriptors for Weakly Annotated Histopathological Breast Cancer Data

Introduction: Cancerous Tissue Recognition (CTR) methodologies are continuously integrating advancements at the forefront of machine learning and computer vision, providing a variety of inference schemes for histopathological data. Histopathological data, in most cases, come in the form of high-resolution images, and thus methodologies operating at the patch level are more computationally attractive. Such methodologies capitalize on pixel level annotations (tissue delineations) from expert pathologists, which are then used to derive labels at the patch level. In this work, we envision a digital connected health system that augments the capabilities of the clinicians by providing powerful feature descriptors that may describe malignant regions.Material and Methods: We start with a patch level descriptor, termed Covariance-Kernel Descriptor (CKD), capable of compactly describing tissue architectures associated with carcinomas. To leverage the recognition capability of the CKDs to larger slide regions, we resort to a multiple instance learning framework. In that direction, we derive the Weakly Annotated Image Descriptor (WAID) as the parameters of classifier decision boundaries in a Multiple Instance Learning framework. The WAID is computed on bags of patches corresponding to larger image regions for which binary labels (malignant vs. benign) are provided, thus obviating the necessity for tissue delineations.Results: The CKD was seen to outperform all the considered descriptors, reaching classification accuracy (ACC) of 92.83%. and area under the curve (AUC) of 0.98. The CKD captures higher order correlations between features and was shown to achieve superior performance against a large collection of computer vision features on a private breast cancer dataset. The WAID outperform all other descriptors on the Breast Cancer Histopathological database (BreakHis) where correctly classified malignant (CCM) instances reached 91.27 and 92.00% at the patient and image level, respectively, without resorting to a deep learning scheme achieves state-of-the-art performance.Discussion: Our proposed derivation of the CKD and WAID can help medical experts accomplish their work accurately and faster than the current state-of-the-art.

Multi-frequency and multi-domain human activity recognition based on SFCW radar using deep learning

Deep learning and radar make it feasible to automatically recognize human activities in various lighting conditions, even occlusion case, which significantly promotes the application of activity recognition in the fields of security surveillance, health care, and so on. In this paper, an approach for human activity recognition (HAR) using deep learning is proposed based on stepped frequency continues wave (SFCW) radar. Specifically, SFCW radar is utilized to generate two types of characteristic representation domains, namely multiple frequencies of spectrograms in time–frequency domain and range maps in range domain. On the one hand, spectrograms and range maps provide different types of features. On the other hand, multi-frequency spectrograms furnish same type of features while with different scattering properties and frequency resolutions. Then a specific deep learning network including multiple parallel deep convolutional neural networks (DCNNs) and a sparse autoencoder is designed to extract and fuse these features associated with human activities from the multi-frequency spectrograms and rang map. In particular, each DCNN is aimed at extracting the detailed micro-Doppler features from a spectrogram, while sparse autoencoder learns prime range distribution features by compressing each range map to reduce complexity and improve robustness. Experimental results verify that the proposed deep learning scheme achieves 96.42% recognition accuracy about six types of activities by incorporating three frequencies of spectrograms and range map, and surpasses two existed methods depending on single-frequency spectrogram and combination of single-frequency spectrogram and range map.

Complicated robot activity recognition by quality-aware deep reinforcement learning

Automatic robot activity understanding plays an important role in human–computer interaction (HCI), especially in smart home service robots. Existing manipulator control methods, such as position control, vision-based control method, fail to meet the requirements of autonomous learning. Reinforcement learning can cope with the interaction of robot control and environment; however, the method should relearn the control method when the position of target object changes. To solve this problem, this paper proposes a quality model to utilize deep reinforcement learning scheme to achieve an end-to-end manipulator control. Specifically, we design a policy search algorithm to achieve automatic learning of manipulator. To avoid relearning of manipulator, we design convolutional neural network control scheme to remain the robustness of manipulator. Extensive experiment has shown the effectiveness of our proposed method.

A deep active learning system for species identification and counting in camera trap images

Abstract A typical camera trap survey may produce millions of images that require slow, expensive manual review. Consequently, critical conservation questions may be answered too slowly to support decision‐making. Recent studies demonstrated the potential for computer vision to dramatically increase efficiency in image‐based biodiversity surveys; however, the literature has focused on projects with a large set of labelled training images, and hence many projects with a smaller set of labelled images cannot benefit from existing machine learning techniques. Furthermore, even sizable projects have struggled to adopt computer vision methods because classification models overfit to specific image backgrounds (i.e. camera locations). In this paper, we combine the power of machine intelligence and human intelligence via a novel active learning system to minimize the manual work required to train a computer vision model. Furthermore, we utilize object detection models and transfer learning to prevent overfitting to camera locations. To our knowledge, this is the first work to apply an active learning approach to camera trap images. Our proposed scheme can match state‐of‐the‐art accuracy on a 3.2 million image dataset with as few as 14,100 manual labels, which means decreasing manual labelling effort by over 99.5%. Our trained models are also less dependent on background pixels, since they operate only on cropped regions around animals. The proposed active deep learning scheme can significantly reduce the manual labour required to extract information from camera trap images. Automation of information extraction will not only benefit existing camera trap projects, but can also catalyse the deployment of larger camera trap arrays.

Contention Resolution in Wi-Fi 6-Enabled Internet of Things Based on Deep Learning

Internet of Things (IoT) is expected to vastly increase the number of connected devices. As a result, a multitude of IoT devices transmit various information through wireless communication technology, such as the Wi-Fi technology, cellular mobile communication technology, low-power wide-area network (LPWAN) technology. However, even the latest Wi-Fi technology is still ready to accommodate these large amounts of data. Accurately setting the contention window (CW) value significantly affects the efficiency of the Wi-Fi network. Unfortunately, the standard collision resolution used by IEEE 802.11ax networks is nonscalable; thus, it cannot maintain stable throughput for an increasing number of stations, even when Wi-Fi 6 has been designed to improve performance in dense scenarios. To this end, we propose a CW control strategy for Wi-Fi 6 systems. This strategy leverages deep learning to search for optimal configuration of CW under different network conditions. Our deep neural network is trained by data generated from a Wi-Fi 6 simulation system with some varying key parameters, e.g., the number of nodes, short interframe space (SIFS), distributed interframe space (DIFS), and data transmission rate. Numerical results demonstrated that our deep learning scheme could always find the optimal CW adjustment multiple by adaptively perceiving the channel competition status. The finalized performance of our model has been significantly improved in terms of system throughput, average transmission delay, and packet retransmission rate. This makes Wi-Fi 6 better adapted to the access of a large number of IoT devices.

An Efficient Deep Learning Scheme To Predict the Electronic Structure of Materials and Molecules: The Example of Graphene-Derived Allotropes.

Computations based on density functional theory (DFT) are transforming various aspects of materials research and discovery. However, the effort required to solve the central equation of DFT, namely the Kohn-Sham equation, which remains a major obstacle for studying large systems with hundreds of atoms in a practical amount of time with routine computational resources. Here, we propose a deep learning architecture that systematically learns the input-output behavior of the Kohn-Sham equation and predicts the electronic density of states, a primary output of DFT calculations, with unprecedented speed and chemical accuracy. The algorithm also adapts and progressively improves in predictive power and versatility as it is exposed to new diverse atomic configurations. We demonstrate this capability for a diverse set of carbon allotropes spanning a large configurational and phase space. The electronic density of states, along with the electronic charge density, may be used downstream to predict a variety of materials properties, bypassing the Kohn-Sham equation, leading to an ultrafast and high-fidelity DFT emulator.

An Improved Deep Learning Scheme for Solving 2-D and 3-D Inverse Scattering Problems

Reconstructing the exact electromagnetic property of unknown targets from the measured scattered field is challenging due to the intrinsic nonlinearity and ill-posedness. In this article, a new scheme, named the modified contrast scheme (MCS), is proposed to tackle nonlinear inverse scattering problems (ISPs). A local-wave amplifier coefficient is used to form the modified contrast, which is able to alleviate the global nonlinearity in original ISPs without decreasing the accuracy of the problem formulation. Moreover, the modified contrast is more suitable to be the input of the deep learning scheme, due to the unity bound of the modified contrast. The numerical results show that MCS with the modified contrast input performs well in both 2-D and 3-D testing examples in real time after offline training process, even in high-relative-permittivity cases. Compared with the dominant current scheme, a significant improvement is achieved in reconstructing high-contrast scatterers.

Incorporating Deep Learning With Word Embedding to Identify Plant Ubiquitylation Sites.

Protein ubiquitylation is an important posttranslational modification (PTM), which is involved in diverse biological processes and plays an essential role in the regulation of physiological mechanisms and diseases. The Protein Lysine Modifications Database (PLMD) has accumulated abundant ubiquitylated proteins with their substrate sites for more than 20 kinds of species. Numerous works have consequently developed a variety of ubiquitylation site prediction tools across all species, mainly relying on the predefined sequence features and machine learning algorithms. However, the difference in ubiquitylated patterns between these species stays unclear. In this work, the sequence-based characterization of ubiquitylated substrate sites has revealed remarkable differences among plants, animals, and fungi. Then an improved word-embedding scheme based on the transfer learning strategy was incorporated with the multilayer convolutional neural network (CNN) for identifying protein ubiquitylation sites. For the prediction of plant ubiquitylation sites, the proposed deep learning scheme could outperform the machine learning-based methods, with the accuracy of 75.6%, precision of 73.3%, recall of 76.7%, F-score of 0.7493, and 0.82 AUC on the independent testing set. Although the ubiquitylated specificity of substrate sites is complicated, this work has demonstrated that the application of the word-embedding method can enable the extraction of informative features and help the identification of ubiquitylated sites. To accelerate the investigation of protein ubiquitylation, the data sets and source code used in this study are freely available at https://github.com/wang-hong-fei/DL-plant-ubsites-prediction.

Decoding Hand Motor Imagery Tasks Within the Same Limb From EEG Signals Using Deep Learning

Motor imagery (MI) tasks of different body parts have been successfully decoded by conventional classifiers, such as LDA and SVM. On the other hand, decoding MI tasks within the same limb is a challenging problem with these classifiers; however, it would provide more options to control robotic devices. This work proposes to improve the hand MI tasks decoding within the same limb in a brain-computer interface using convolutional neural networks (CNNs); the CNN EEGNet, LDA, and SVM classifiers were evaluated for two (flexion/extension) and three (flexion/extension/grasping) MI tasks. Our approach is the first attempt to apply CNNs for solving this problem to our best knowledge. In addition, visual and electrotactile stimulation were included as BCI training reinforcement after the MI task similar to feedback sessions; then, they were compared. The EEGNet achieved maximum mean accuracies of 78.46% (±12.50%) and 76.72% (±11.67%) for two and three classes, respectively. Outperforming conventional classifiers with results around 60% and 48%, and similar works with results lower than 67% and 75%, respectively. Moreover, the electrical stimulation over the visual stimulus was not significant during the calibration session. The deep learning scheme enhanced the decoding of MI tasks within the same limb against the conventional framework.