Search

In this paper, research was conducted on anomaly detection of wheel flats. In the railway sector, conducting tests with actual railway vehicles is challenging due to safety concerns for passengers and maintenance issues as it is a public industry. Therefore, dynamics software was utilized. Next, STFT (short-time Fourier transform) was performed to create spectrogram images. In the case of railway vehicles, control, monitoring, and communication are performed through TCMS, but complex analysis and data processing are difficult because there are no devices such as GPUs. Furthermore, there are memory limitations. Therefore, in this paper, the relatively lightweight models LeNet-5, ResNet-20, and MobileNet-V3 were selected for deep learning experiments. At this time, the LeNet-5 and MobileNet-V3 models were modified from the basic architecture. Since railway vehicles are given preventive maintenance, it is difficult to obtain fault data. Therefore, semi-supervised learning was also performed. At this time, the Deep One Class Classification paper was referenced. The evaluation results indicated that the modified LeNet-5 and MobileNet-V3 models achieved approximately 97% and 96% accuracy, respectively. At this point, the LeNet-5 model showed a training time of 12 min faster than the MobileNet-V3 model. In addition, the semi-supervised learning results showed a significant outcome of approximately 94% accuracy when considering the railway maintenance environment. In conclusion, considering the railway vehicle maintenance environment and device specifications, it was inferred that the relatively simple and lightweight LeNet-5 model can be effectively utilized while using small images. [ABSTRACT FROM AUTHOR]

With the increasing number of households owning pets, the importance of sensor data for recognizing pet behavior has grown significantly. However, challenges arise due to the costs and reliability issues associated with data collection. This paper proposes a method for classifying pet behavior using cleaned meta pseudo labels to overcome these issues. The data for this study were collected using wearable devices equipped with accelerometers, gyroscopes, and magnetometers, and pet behaviors were classified into five categories. Utilizing this data, we analyzed the impact of the quantity of labeled data on accuracy and further enhanced the learning process by integrating an additional Distance Loss. This method effectively improves the learning process by removing noise from unlabeled data. Experimental results demonstrated that while the conventional supervised learning method achieved an accuracy of 82.9%, the existing meta pseudo labels method showed an accuracy of 86.2%, and the cleaned meta pseudo labels method proposed in this study surpassed these with an accuracy of 88.3%. These results hold significant implications for the development of pet monitoring systems, and the approach of this paper provides an effective solution for recognizing and classifying pet behavior in environments with insufficient labels. [ABSTRACT FROM AUTHOR]

Delamination represents one of the most significant and dangerous damages in composite plates. Recently, many papers have presented the capability of structural health monitoring (SHM) techniques for the investigation of structural delamination with various shapes and thickness depths. However, few studies have been conducted regarding the utilization of convolutional neural network (CNN) methods for automating the non-destructive testing (NDT) techniques database to identify the delamination size and depth. In this paper, an automated system qualified for distinguishing between pristine and damaged structures and classifying three classes of delamination with various depths is presented. This system includes a proposed CNN model and the Lamb wave technique. In this work, a unidirectional composite plate with three samples of delamination inserted at different depths was prepared for numerical and experimental investigations. In the numerical part, the guided wave propagation and interaction with three samples of delamination were studied to observe how the delamination depth can affect the scattered and trapped waves over the delamination region. This numerical study was validated experimentally using an efficient ultrasonic guided waves technique. This technique involved piezoelectric wafer active sensors (PWASs) and a scanning laser Doppler vibrometer (SLDV). Both numerical and experimental studies demonstrate that the delamination depth has a direct effect on the trapped waves' energy and distribution. Three different datasets were collected from the numerical and experimental studies, involving the numerical wavefield image dataset, experimental wavefield image dataset, and experimental wavenumber spectrum image dataset. These three datasets were used independently with the proposed CNN model to develop a system that can automatically classify four classes (pristine class and three different delamination classes). The results of all three datasets show the capability of the proposed CNN model for predicting the delamination depth with high accuracy. The proposed CNN model results of the three different datasets were validated using the GoogLeNet CNN. The results of both methods show an excellent agreement. The results proved the capability of the wavefield image and wavenumber spectrum datasets to be used as input data to the CNN for the detection of delamination depth. [ABSTRACT FROM AUTHOR]

Negative emotions of drivers may lead to some dangerous driving behaviors, which in turn lead to serious traffic accidents. However, most of the current studies on driver emotions use a single modality, such as EEG, eye trackers, and driving data. In complex situations, a single modality may not be able to fully consider a driver's complete emotional characteristics and provides poor robustness. In recent years, some studies have used multimodal thinking to monitor single emotions such as driver fatigue and anger, but in actual driving environments, negative emotions such as sadness, anger, fear, and fatigue all have a significant impact on driving safety. However, there are very few research cases using multimodal data to accurately predict drivers' comprehensive emotions. Therefore, based on the multi-modal idea, this paper aims to improve drivers' comprehensive emotion recognition. By combining the three modalities of a driver's voice, facial image, and video sequence, the six classification tasks of drivers' emotions are performed as follows: sadness, anger, fear, fatigue, happiness, and emotional neutrality. In order to accurately identify drivers' negative emotions to improve driving safety, this paper proposes a multi-modal fusion framework based on the CNN + Bi-LSTM + HAM to identify driver emotions. The framework fuses feature vectors of driver audio, facial expressions, and video sequences for comprehensive driver emotion recognition. Experiments have proved the effectiveness of the multi-modal data proposed in this paper for driver emotion recognition, and its recognition accuracy has reached 85.52%. At the same time, the validity of this method is verified by comparing experiments and evaluation indicators such as accuracy and F1 score. [ABSTRACT FROM AUTHOR]

The main purpose of this paper is to provide information on how to create a convolutional neural network (CNN) for extracting features from EEG signals. Our task was to understand the primary aspects of creating and fine-tuning CNNs for various application scenarios. We considered the characteristics of EEG signals, coupled with an exploration of various signal processing and data preparation techniques. These techniques include noise reduction, filtering, encoding, decoding, and dimension reduction, among others. In addition, we conduct an in-depth analysis of well-known CNN architectures, categorizing them into four distinct groups: standard implementation, recurrent convolutional, decoder architecture, and combined architecture. This paper further offers a comprehensive evaluation of these architectures, covering accuracy metrics, hyperparameters, and an appendix that contains a table outlining the parameters of commonly used CNN architectures for feature extraction from EEG signals. [ABSTRACT FROM AUTHOR]

Tool wear condition monitoring is an important component of mechanical processing automation, and accurately identifying the wear status of tools can improve processing quality and production efficiency. This paper studied a new deep learning model, to identify the wear status of tools. The force signal was transformed into a two-dimensional image using continuous wavelet transform (CWT), short-time Fourier transform (STFT), and Gramian angular summation field (GASF) methods. The generated images were then fed into the proposed convolutional neural network (CNN) model for further analysis. The calculation results show that the accuracy of tool wear state recognition proposed in this paper was above 90%, which was higher than the accuracy of AlexNet, ResNet, and other models. The accuracy of the images generated using the CWT method and identified with the CNN model was the highest, which is attributed to the fact that the CWT method can extract local features of an image and is less affected by noise. Comparing the precision and recall values of the model, it was verified that the image obtained by the CWT method had the highest accuracy in identifying tool wear state. These results demonstrate the potential advantages of using a force signal transformed into a two-dimensional image for tool wear state recognition and of applying CNN models in this area. They also indicate the wide application prospects of this method in industrial production. [ABSTRACT FROM AUTHOR]

The measurement and analysis of vital signs are a subject of significant research interest, particularly for monitoring the driver's physiological state, which is of crucial importance for road safety. Various approaches have been proposed using contact techniques to measure vital signs. However, all of these methods are invasive and cumbersome for the driver. This paper proposes using a non-contact sensor based on continuous wave (CW) radar at 24 GHz to measure vital signs. We associate these measurements with distinct temporal neural networks to analyze the signals to detect and extract heart and respiration rates as well as classify the physiological state of the driver. This approach offers robust performance in estimating the exact values of heart and respiration rates and in classifying the driver's physiological state. It is non-invasive and requires no physical contact with the driver, making it particularly practical and safe. The results presented in this paper, derived from the use of a 1D Convolutional Neural Network (1D-CNN), a Temporal Convolutional Network (TCN), a Recurrent Neural Network particularly the Bidirectional Long Short-Term Memory (Bi-LSTM), and a Convolutional Recurrent Neural Network (CRNN). Among these, the CRNN emerged as the most effective Deep Learning approach for vital signal analysis. [ABSTRACT FROM AUTHOR]

In this paper, research was conducted on Deep Learning Wavefront Sensing (DLWS) neural networks using simulated atmospheric turbulence datasets, and a novel DLWS was proposed based on attention mechanisms and Convolutional Neural Networks (CNNs). The study encompassed both indoor experiments and kilometer-range laser transmission experiments employing DLWS. In terms of indoor experiments, data were collected and training was performed on the platform built by us. Subsequent comparative experiments with the Shack-Hartmann Wavefront Sensing (SHWS) method revealed that our DLWS model achieved accuracy on par with SHWS. For the kilometer-scale experiments, we directly applied the DLWS model obtained from the indoor platform, eliminating the need for new data collection or additional training. The DLWS predicts the wavefront from the beacon light PSF in real time and then uses it for aberration correction of the emitted laser. The results demonstrate a substantial improvement in the average peak intensity of the light spot at the target position after closed-loop correction, with a remarkable increase of 5.35 times compared to the open-loop configuration. [ABSTRACT FROM AUTHOR]

Deaf and hard-of-hearing people mainly communicate using sign language, which is a set of signs made using hand gestures combined with facial expressions to make meaningful and complete sentences. The problem that faces deaf and hard-of-hearing people is the lack of automatic tools that translate sign languages into written or spoken text, which has led to a communication gap between them and their communities. Most state-of-the-art vision-based sign language recognition approaches focus on translating non-Arabic sign languages, with few targeting the Arabic Sign Language (ArSL) and even fewer targeting the Saudi Sign Language (SSL). This paper proposes a mobile application that helps deaf and hard-of-hearing people in Saudi Arabia to communicate efficiently with their communities. The prototype is an Android-based mobile application that applies deep learning techniques to translate isolated SSL to text and audio and includes unique features that are not available in other related applications targeting ArSL. The proposed approach, when evaluated on a comprehensive dataset, has demonstrated its effectiveness by outperforming several state-of-the-art approaches and producing results that are comparable to these approaches. Moreover, testing the prototype on several deaf and hard-of-hearing users, in addition to hearing users, proved its usefulness. In the future, we aim to improve the accuracy of the model and enrich the application with more features. [ABSTRACT FROM AUTHOR]

In this paper, an innovative cyclic noise reduction method and an improved CAPON algorithm (also the called minimum variance distortionless response (MVDR) algorithm) are proposed to improve the accuracy and reliability of DOA (direction of arrival) estimation. By processing the eigenvalues obtained from the covariance matrix of the received signal, the signal-to-noise ratio (SNR) can be increased by up to 5 dB by the cyclic noise reduction method, which will improve the DOA estimation accuracy. The improved CAPON algorithm has a convolution neural network (CNN) structure, whose input is the processed covariance matrix of the received signal, and the CAPON spectral value is used as the training label to obtain the estimated spatial spectrum. It retains the advantages of the CAPON algorithm, which can achieve blind source estimation, and simulations show that the proposed algorithm exhibits a better performance than the traditional algorithm in conditions of various SNRs and snapshot numbers. The simulation results show that, based on a certain SNR, the root mean square error (RMSE) of the improved CAPON algorithm can be reduced from 0.86° to 0.8° compared to traditional algorithms, and the angle estimation error can be decreased by up to about 0.3°. With the help of the cyclic noise reduction method, the angle estimation error decreases from 0.04° to 0.02°. [ABSTRACT FROM AUTHOR]

The network intrusion detection system (NIDS) plays a crucial role as a security measure in addressing the increasing number of network threats. The majority of current research relies on feature-ready datasets that heavily depend on feature engineering. Conversely, the increasing complexity of network traffic and the ongoing evolution of attack techniques lead to a diminishing distinction between benign and malicious network behaviors. In this paper, we propose a novel end-to-end intrusion detection framework based on a contrastive learning approach. We design a hierarchical Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) model to facilitate the automated extraction of spatiotemporal features from raw traffic data. The integration of contrastive learning amplifies the distinction between benign and malicious network traffic in the representation space. The proposed method exhibits enhanced detection capabilities for unknown attacks in comparison to the approaches trained using the cross-entropy loss function. Experiments are carried out on the public datasets CIC-IDS2017 and CSE-CIC-IDS2018, demonstrating that our method can attain a detection accuracy of 99.9% for known attacks, thus achieving state-of-the-art performance. For unknown attacks, a weighted recall rate of 95% can be achieved. [ABSTRACT FROM AUTHOR]

In recent years, computer vision has witnessed remarkable advancements in image classification, specifically in the domains of fully convolutional neural networks (FCNs) and self-attention mechanisms. Nevertheless, both approaches exhibit certain limitations. FCNs tend to prioritize local information, potentially overlooking crucial global contexts, whereas self-attention mechanisms are computationally intensive despite their adaptability. In order to surmount these challenges, this paper proposes cross-and-diagonal networks (CDNet), innovative network architecture that adeptly captures global information in images while preserving local details in a more computationally efficient manner. CDNet achieves this by establishing long-range relationships between pixels within an image, enabling the indirect acquisition of contextual information. This inventive indirect self-attention mechanism significantly enhances the network's capacity. In CDNet, a new attention mechanism named "cross and diagonal attention" is proposed. This mechanism adopts an indirect approach by integrating two distinct components, cross attention and diagonal attention. By computing attention in different directions, specifically vertical and diagonal, CDNet effectively establishes remote dependencies among pixels, resulting in improved performance in image classification tasks. Experimental results highlight several advantages of CDNet. Firstly, it introduces an indirect self-attention mechanism that can be effortlessly integrated as a module into any convolutional neural network (CNN). Additionally, the computational cost of the self-attention mechanism has been effectively reduced, resulting in improved overall computational efficiency. Lastly, CDNet attains state-of-the-art performance on three benchmark datasets for similar types of image classification networks. In essence, CDNet addresses the constraints of conventional approaches and provides an efficient and effective solution for capturing global context in image classification tasks. [ABSTRACT FROM AUTHOR]

Crowd counting, as a basic computer vision task, plays an important role in many fields such as video surveillance, accident prediction, public security, and intelligent transportation. At present, crowd counting tasks face various challenges. Firstly, due to the diversity of crowd distribution and increasing population density, there is a phenomenon of large-scale crowd aggregation in public places, sports stadiums, and stations, resulting in very serious occlusion. Secondly, when annotating large-scale datasets, positioning errors can also easily affect training results. In addition, the size of human head targets in dense images is not consistent, making it difficult to identify both near and far targets using only one network simultaneously. The existing crowd counting methods mainly use density plot regression methods. However, this framework does not distinguish the features between distant and near targets and cannot adaptively respond to scale changes. Therefore, the detection performance in areas with sparse population distribution is not good. To solve such problems, we propose an adaptive multi-scale far and near distance network based on the convolutional neural network (CNN) framework for counting dense populations and achieving a good balance between accuracy, inference speed, and performance. However, on the feature level, in order to enable the model to distinguish the differences between near and far features, we use stacked convolution layers to deepen the depth of the network, allocate different receptive fields according to the distance between the target and the camera, and fuse the features between nearby targets to enhance the feature extraction ability of pedestrians under nearby targets. Secondly, depth information is used to distinguish distant and near targets of different scales and the original image is cut into four different patches to perform pixel-level adaptive modeling on the population. In addition, we add density normalized average precision (nAP) indicators to analyze the accuracy of our method in spatial positioning. This paper validates the effectiveness of NF-Net on three challenging benchmarks in Shanghai Tech Part A and B, UCF_ CC_50, and UCF-QNRF datasets. Compared with SOTA, it has more significant performance in various scenarios. In the UCF-QNRF dataset, it is further validated that our method effectively solves the interference of complex backgrounds. [ABSTRACT FROM AUTHOR]

Recently, convolutional neural networks (CNNs) have been widely used in image denoising, and their performance has been enhanced through residual learning. However, previous research mostly focused on optimizing the network architecture of CNNs, ignoring the limitations of the commonly used residual learning. This paper identifies two of its limitations, which are the neglect of image information and the lack of effective consideration of image self-similarity. To solve these limitations, this paper proposes a compositional denoising network (CDN), which contains two sub-paths, the image information path (IIP) and the noise estimation path (NEP), respectively. IIP is trained via an image-to-image method to extract image information. For NEP, it utilizes image self-similarity from the perspective of training. This similarity-based training method constrains NEP to output similar estimated noise distributions for different image patches with a specific kind of noise. Finally, image information and noise distribution information are comprehensively considered for image denoising. Experimental results indicate that CDN outperforms other CNN-based methods in both synthetic and real-world image denoising, achieving state-of-the-art performance. [ABSTRACT FROM AUTHOR]

The Convolutional Neural Network (CNN) has demonstrated excellent performance in image recognition and has brought new opportunities for sign language recognition. However, the features undergo many nonlinear transformations while performing the convolutional operation and the traditional CNN models are insufficient in dealing with the correlation between images. In American Sign Language (ASL) recognition, J and Z with moving gestures bring recognition challenges. This paper proposes a novel Two-Stream Mixed (TSM) method with feature extraction and fusion operation to improve the correlation of feature expression between two time-consecutive images for the dynamic gestures. The proposed TSM-CNN system is composed of preprocessing, the TSM block, and CNN classifiers. Two consecutive images in the dynamic gesture are used as inputs of streams, and resizing, transformation, and augmentation are carried out in the preprocessing stage. The fusion feature map obtained by addition and concatenation in the TSM block is used as inputs of the classifiers. Finally, a classifier classifies images. The TSM-CNN model with the highest performance scores depending on three concatenation methods is selected as the definitive recognition model for ASL recognition. We design 4 CNN models with TSM: TSM-LeNet, TSM-AlexNet, TSM-ResNet18, and TSM-ResNet50. The experimental results show that the CNN models with the TSM are better than models without TSM. The TSM-ResNet50 has the best accuracy of 97.57% for MNIST and ASL datasets and is able to be applied to a RGB image sensing system for hearing-impaired people. [ABSTRACT FROM AUTHOR]

Organizations managing high-performance computing systems face a multitude of challenges, including overarching concerns such as overall energy consumption, microprocessor clock frequency limitations, and the escalating costs associated with chip production. Evidently, processor speeds have plateaued over the last decade, persisting within the range of 2 GHz to 5 GHz. Scholars assert that brain-inspired computing holds substantial promise for mitigating these challenges. The spiking neural network (SNN) particularly stands out for its commendable power efficiency when juxtaposed with conventional design paradigms. Nevertheless, our scrutiny has brought to light several pivotal challenges impeding the seamless implementation of large-scale neural networks (NNs) on silicon. These challenges encompass the absence of automated tools, the need for multifaceted domain expertise, and the inadequacy of existing algorithms to efficiently partition and place extensive SNN computations onto hardware infrastructure. In this paper, we posit the development of an automated tool flow capable of transmuting any NN into an SNN. This undertaking involves the creation of a novel graph-partitioning algorithm designed to strategically place SNNs on a network-on-chip (NoC), thereby paving the way for future energy-efficient and high-performance computing paradigms. The presented methodology showcases its effectiveness by successfully transforming ANN architectures into SNNs with a marginal average error penalty of merely 2.65%. The proposed graph-partitioning algorithm enables a 14.22% decrease in inter-synaptic communication and an 87.58% reduction in intra-synaptic communication, on average, underscoring the effectiveness of the proposed algorithm in optimizing NN communication pathways. Compared to a baseline graph-partitioning algorithm, the proposed approach exhibits an average decrease of 79.74% in latency and a 14.67% reduction in energy consumption. Using existing NoC tools, the energy-latency product of SNN architectures is, on average, 82.71% lower than that of the baseline architectures. [ABSTRACT FROM AUTHOR]

The need for efficient video coding technology is more important than ever in the current scenario where video applications are increasing worldwide, and Internet of Things (IoT) devices are becoming widespread. In this context, it is necessary to carefully review the recently completed MPEG-5 Essential Video Coding (EVC) standard because the EVC Baseline profile is customized to meet the specific requirements needed to process IoT video data in terms of low complexity. Nevertheless, the EVC Baseline profile has a notable disadvantage. Since it is a codec composed only of simple tools developed over 20 years, it tends to represent numerous coding artifacts. In particular, the presence of blocking artifacts at the block boundary is regarded as a critical issue that must be addressed. To address this, this paper proposes a post-filter using a block partitioning information-based Convolutional Neural Network (CNN). The proposed method in the experimental results objectively shows an approximately 0.57 dB for All-Intra (AI) and 0.37 dB for Low-Delay (LD) improvements in each configuration by the proposed method when compared to the pre-post-filter video, and the enhanced PSNR results in an overall bitrate reduction of 11.62% for AI and 10.91% for LD in the Luma and Chroma components, respectively. Due to the huge improvement in the PSNR, the proposed method significantly improved the visual quality subjectively, particularly in blocking artifacts at the coding block boundary. [ABSTRACT FROM AUTHOR]

This paper introduces a noise augmentation technique designed to enhance the robustness of state-of-the-art (SOTA) deep learning models against degraded image quality, a common challenge in long-term recording systems. Our method, demonstrated through the classification of digital holographic images, utilizes a novel approach to synthesize and apply random colored noise, addressing the typically encountered correlated noise patterns in such images. Empirical results show that our technique not only maintains classification accuracy in high-quality images but also significantly improves it when given noisy inputs without increasing the training time. This advancement demonstrates the potential of our approach for augmenting data for deep learning models to perform effectively in production under varied and suboptimal conditions. [ABSTRACT FROM AUTHOR]

The popularity and demand for high-quality date palm fruits (Phoenix dactylifera L.) have been growing, and their quality largely depends on the type of handling, storage, and processing methods. The current methods of geometric evaluation and classification of date palm fruits are characterised by high labour intensity and are usually performed mechanically, which may cause additional damage and reduce the quality and value of the product. Therefore, non-contact methods are being sought based on image analysis, with digital solutions controlling the evaluation and classification processes. The main objective of this paper is to develop an automatic classification model for varieties of date palm fruits using a convolutional neural network (CNN) based on two fundamental criteria, i.e., colour difference and evaluation of geometric parameters of dates. A CNN with a fixed architecture was built, marked as DateNET, consisting of a system of five alternating Conv2D, MaxPooling2D, and Dropout classes. The validation accuracy of the model presented in this study depended on the selection of classification criteria. It was 85.24% for fruit colour-based classification and 87.62% for the geometric parameters only; however, it increased considerably to 93.41% when both the colour and geometry of dates were considered. [ABSTRACT FROM AUTHOR]

Detecting and classifying vehicles as objects from images and videos is challenging in appearance-based representation, yet plays a significant role in the substantial real-time applications of Intelligent Transportation Systems (ITSs). The rapid development of Deep Learning (DL) has resulted in the computer-vision community demanding efficient, robust, and outstanding services to be built in various fields. This paper covers a wide range of vehicle detection and classification approaches and the application of these in estimating traffic density, real-time targets, toll management and other areas using DL architectures. Moreover, the paper also presents a detailed analysis of DL techniques, benchmark datasets, and preliminaries. A survey of some vital detection and classification applications, namely, vehicle detection and classification and performance, is conducted, with a detailed investigation of the challenges faced. The paper also addresses the promising technological advancements of the last few years. [ABSTRACT FROM AUTHOR]

The survival rate for sudden cardiac arrest (SCA) is low, and patients with long-term risks of SCA are not adequately alerted. Understanding SCA's characteristics will be key to developing preventive strategies. Many lives could be saved if SCA's early onset could be detected or predicted. Monitoring heart signals continuously is essential for diagnosing sporadic cardiac dysfunction. An electrocardiogram (ECG) can be used to continuously monitor heart function without having to go to the hospital. A zeolite-based dry electrode can provide safe on-skin ECG acquisition while the subject is out-of-hospital and facilitate long-term monitoring. To the ECG signal, a low-power 1 μW read-out circuit was designed and implemented in our prior work. However, having long-term ECG monitoring outside the hospital, i.e., high battery life, and low power consumption while transmission and reception of ECG signal are crucial. This paper proposes a prototype with a 10-bit resolution ADC and nRF24L01 transceivers placed 5 m apart. The system uses the 2.4 GHz worldwide ISM frequency band with GFSK modulation to wirelessly transmit digitized ECG bits at 250 kbps data rate to a physician's computer (or similar) for continuous monitoring of ECG signals; the power consumption is only 11.2 mW and 4.62 mW during transmission and reception, respectively, with a low bit error rate of ≤0.1%. Additionally, a subject-wise cross-validated, three-fold, optimized convolutional neural network (CNN) model using the Physionet-SCA dataset was implemented on NVIDIA Jetson to identify the irregular heartbeats yielding an accuracy of 89% with a run time of 5.31 s. Normal beat classification has an F1 score of 0.94 and a ROC score of 0.886. Thus, this paper integrates the ECG acquisition and processing unit with low-power wireless transmission and CNN model to detect irregular heartbeats. [ABSTRACT FROM AUTHOR]

Human activity recognition (HAR) using wearable sensors enables continuous monitoring for healthcare applications. However, the conventional centralised training of deep learning models on sensor data poses challenges related to privacy, communication costs, and on-device efficiency. This paper proposes a federated learning framework integrating spiking neural networks (SNNs) with long short-term memory (LSTM) networks for energy-efficient and privacy-preserving HAR. The hybrid spiking-LSTM (S-LSTM) model synergistically combines the event-driven efficiency of SNNs and the sequence modelling capability of LSTMs. The model is trained using surrogate gradient learning and backpropagation through time, enabling fully supervised end-to-end learning. Extensive evaluations of two public datasets demonstrate that the proposed approach outperforms LSTM, CNN, and S-CNN models in accuracy and energy efficiency. For instance, the proposed S-LSTM achieved an accuracy of 97.36% and 89.69% for indoor and outdoor scenarios, respectively. Furthermore, the results also showed a significant improvement in energy efficiency of 32.30%, compared to simple LSTM. Additionally, we highlight the significance of personalisation in HAR, where fine-tuning with local data enhances model accuracy by up to 9% for individual users. [ABSTRACT FROM AUTHOR]

The escalating global water usage and the increasing strain on major cities due to water shortages highlights the critical need for efficient water management practices. In water-stressed regions worldwide, significant water wastage is primarily attributed to leakages, inefficient use, and aging infrastructure. Undetected water leakages in buildings' pipelines contribute to the water waste problem. To address this issue, an effective water leak detection method is required. In this paper, we explore the application of edge computing in smart buildings to enhance water management. By integrating sensors and embedded Machine Learning models, known as TinyML, smart water management systems can collect real-time data, analyze it, and make accurate decisions for efficient water utilization. The transition to TinyML enables faster and more cost-effective local decision-making, reducing the dependence on centralized entities. In this work, we propose a solution that can be adapted for effective leakage detection in real-world scenarios with minimum human intervention using TinyML. We follow an approach that is similar to a typical machine learning lifecycle in production, spanning stages including data collection, training, hyperparameter tuning, offline evaluation and model optimization for on-device resource efficiency before deployment. In this work, we considered an existing water leakage acoustic dataset for polyvinyl chloride pipelines. To prepare the acoustic data for analysis, we performed preprocessing to transform it into scalograms. We devised a water leak detection method by applying transfer learning to five distinct Convolutional Neural Network (CNN) variants, which are namely EfficientNet, ResNet, AlexNet, MobileNet V1, and MobileNet V2. The CNN models were found to be able to detect leakages where a maximum testing accuracy, recall, precision, and F1 score of 97.45%, 98.57%, 96.70%, and 97.63%, respectively, were observed using the EfficientNet model. To enable seamless deployment on the Arduino Nano 33 BLE edge device, the EfficientNet model is compressed using quantization resulting in a low inference time of 1932 ms, a peak RAM usage of 255.3 kilobytes, and a flash usage requirement of merely 48.7 kilobytes. [ABSTRACT FROM AUTHOR]

A hypoglossal nerve stimulator (HGNS) is an invasive device that is used to treat obstructive sleep apnea (OSA) through electrical stimulation. The conventional implantable HGNS device consists of a stimuli generator, a breathing sensor, and electrodes connected to the hypoglossal nerve via leads. However, this implant is bulky and causes significant trauma. In this paper, we propose a minimally invasive HGNS based on an electrocardiogram (ECG) sensor and wireless power transfer (WPT), consisting of a wearable breathing monitor and an implantable stimulator. The breathing external monitor utilizes an ECG sensor to identify abnormal breathing patterns associated with OSA with 88.68% accuracy, achieved through the utilization of a convolutional neural network (CNN) algorithm. With a skin thickness of 5 mm and a receiving coil diameter of 9 mm, the power conversion efficiency was measured as 31.8%. The implantable device, on the other hand, is composed of a front-end CMOS power management module (PMM), a binary-phase-shift-keying (BPSK)-based data demodulator, and a bipolar biphasic current stimuli generator. The PMM, with a silicon area of 0.06 mm 2 (excluding PADs), demonstrated a power conversion efficiency of 77.5% when operating at a receiving frequency of 2 MHz. Furthermore, it offers three-voltage options (1.2 V, 1.8 V, and 3.1 V). Within the data receiver component, a low-power BPSK demodulator was ingeniously incorporated, consuming only 42 μ W when supplied with a voltage of 0.7 V. The performance was achieved through the implementation of the self-biased phase-locked-loop (PLL) technique. The stimuli generator delivers biphasic constant currents, providing a 5 bit programmable range spanning from 0 to 2.4 mA. The functionality of the proposed ECG- and WPT-based HGNS was validated, representing a highly promising solution for the effective management of OSA, all while minimizing the trauma and space requirements. [ABSTRACT FROM AUTHOR]

Alzheimer's disease (AD), a neuropsychiatric disorder, continually arises in the elderly. To date, no targeted medications have been developed for AD. Early and fast diagnosis of AD plays a pivotal role in identifying potential AD patients, enabling timely medical interventions, and mitigating disease progression. Computer-aided diagnosis (CAD) becomes possible with the burgeoning of deep learning. However, the existing CAD models for processing 3D Alzheimer's disease images usually have the problems of slow convergence, disappearance of gradient, and falling into local optimum. This makes the training of 3D diagnosis models need a lot of time, and the accuracy is often poor. In this paper, a novel 3D aggregated residual network with accelerated mirror descent optimization is proposed for diagnosing AD. First, a novel unbiased subgradient accelerated mirror descent (SAMD) optimization algorithm is proposed to speed up diagnosis network training. By optimizing the nonlinear projection process, our proposed algorithm can avoid the occurrence of the local optimum in the non-Euclidean distance metric. The most notable aspect is that, to the best of our knowledge, this is the pioneering attempt to optimize the AD diagnosis training process by improving the optimization algorithm. Then, we provide a rigorous proof of the SAMD's convergence, and the convergence of SAMD is better than any existing gradient descent algorithms. Finally, we use our proposed SAMD algorithm to train our proposed 3D aggregated residual network architecture (ARCNN). We employed the ADNI dataset to train ARCNN diagnostic models separately for the AD vs. NC task and the sMCI vs. pMCI task, followed by testing to evaluate the disease diagnostic outcomes. The results reveal that the accuracy can be improved in diagnosing AD, and the training speed can be accelerated. Our proposed method achieves 95.4% accuracy in AD diagnosis and 79.9% accuracy in MCI diagnosis; the best results contrasted with several state-of-the-art diagnosis methods. In addition, our proposed SAMD algorithm can save about 19% of the convergence time on average in the AD diagnosis model compared with the gradient descent algorithms, which is very momentous in clinic. [ABSTRACT FROM AUTHOR]

Internet of Things (IoT) devices within smart cities, require innovative detection methods. This paper addresses this critical challenge by introducing a deep learning-based approach for the detection of network traffic attacks in IoT ecosystems. Leveraging the Kaggle dataset, our model integrates Convolutional Neural Networks (CNNs) and Gated Recurrent Units (GRUs) to capture both spatial and sequential features in network traffic data. We trained and evaluated our model over ten epochs, achieving an impressive overall accuracy rate of 99%. The classification report reveals the model's proficiency in distinguishing various attack categories, including 'Normal', 'DoS' (Denial of Service), 'Probe', 'U2R' (User to Root), and 'Sybil'. Additionally, the confusion matrix offers valuable insights into the model's performance across these attack types. In terms of overall accuracy, our model achieves an impressive accuracy rate of 99% across all attack categories. The weighted- average F1-score is also 99%, showcasing the model's robust performance in classifying network traffic attacks in IoT devices for smart cities. This advanced architecture exhibits the potential to fortify IoT device security in the complex landscape of smart cities, effectively contributing to the safeguarding of critical infrastructure [ABSTRACT FROM AUTHOR]

Medical image segmentation is crucial for medical image processing and the development of computer-aided diagnostics. In recent years, deep Convolutional Neural Networks (CNNs) have been widely adopted for medical image segmentation and have achieved significant success. UNet, which is based on CNNs, is the mainstream method used for medical image segmentation. However, its performance suffers owing to its inability to capture long-range dependencies. Transformers were initially designed for Natural Language Processing (NLP), and sequence-to-sequence applications have demonstrated the ability to capture long-range dependencies. However, their abilities to acquire local information are limited. Hybrid architectures of CNNs and Transformer, such as TransUNet, have been proposed to benefit from Transformer's long-range dependencies and CNNs' low-level details. Nevertheless, automatic medical image segmentation remains a challenging task due to factors such as blurred boundaries, the low-contrast tissue environment, and in the context of ultrasound, issues like speckle noise and attenuation. In this paper, we propose a new model that combines the strengths of both CNNs and Transformer, with network architectural improvements designed to enrich the feature representation captured by the skip connections and the decoder. To this end, we devised a new attention module called Three-Level Attention (TLA). This module is composed of an Attention Gate (AG), channel attention, and spatial normalization mechanism. The AG preserves structural information, whereas channel attention helps to model the interdependencies between channels. Spatial normalization employs the spatial coefficient of the Transformer to improve spatial attention akin to TransNorm. To further improve the skip connection and reduce the semantic gap, skip connections between the encoder and decoder were redesigned in a manner similar to that of the UNet++ dense connection. Moreover, deep supervision using a side-output channel was introduced, analogous to BASNet, which was originally used for saliency predictions. Two datasets from different modalities, a CT scan dataset and an ultrasound dataset, were used to evaluate the proposed UNet architecture. The experimental results showed that our model consistently improved the prediction performance of the UNet across different datasets. [ABSTRACT FROM AUTHOR]

Alzheimer's disease (AD) is a neurodegenerative disease that can cause dementia and result in a severe reduction in brain function, inhibiting simple tasks, especially if no preventative care is taken. Over 1 in 9 Americans suffer from AD-induced dementia, and unpaid care for people with AD-related dementia is valued at USD 271.6 billion. Hence, various approaches have been developed for early AD diagnosis to prevent its further progression. In this paper, we first review other approaches that could be used for the early detection of AD. We then give an overview of our dataset and propose a deep convolutional neural network (CNN) architecture consisting of 7,866,819 parameters. This model comprises three different convolutional branches, each having a different length. Each branch is comprised of different kernel sizes. This model can predict whether a patient is non-demented, mild-demented, or moderately demented with a 99.05% three-class accuracy. In summary, the deep CNN model demonstrated exceptional accuracy in the early diagnosis of AD, offering a significant advancement in the field and the potential to improve patient care. [ABSTRACT FROM AUTHOR]

In today's information age, how to accurately identify a person's identity and protect information security has become a hot topic of people from all walks of life. At present, a more convenient and secure solution to identity identification is undoubtedly biometric identification, but a single biometric identification cannot support increasingly complex and diversified authentication scenarios. Using multimodal biometric technology can improve the accuracy and safety of identification. This paper proposes a biometric method based on finger vein and face bimodal feature layer fusion, which uses a convolutional neural network (CNN), and the fusion occurs in the feature layer. The self-attention mechanism is used to obtain the weights of the two biometrics, and combined with the RESNET residual structure, the self-attention weight feature is cascaded with the bimodal fusion feature channel Concat. To prove the high efficiency of bimodal feature layer fusion, AlexNet and VGG-19 network models were selected in the experimental part for extracting finger vein and face image features as inputs to the feature fusion module. The extensive experiments show that the recognition accuracy of both models exceeds 98.4%, demonstrating the high efficiency of the bimodal feature fusion. [ABSTRACT FROM AUTHOR]

Since the 20th century, a rapid process of motorization has begun. The main goal of researchers, engineers and technology companies is to increase the safety and optimality of the movement of vehicles, as well as to reduce the environmental damage caused by the automotive industry. The difficulty of managing traffic flows is that cars are driven by a person and their behavior, even in similar situations, is different and difficult to predict. To solve this problem, ground-based unmanned vehicles are increasingly being developed and implemented; however, like any other intelligent system, it is necessary to train different road scenarios. Currently, an engineer is driving an unmanned vehicle for training and thousands of kilometers are being driven for training. Of course, this approach to training unmanned vehicles is very long, and it is impossible to reproduce all the scenarios that can be found in real operations on a real road. Based on this, we offer a simulator of a realistic urban environment which allows you to reduce the training time and allows you to generate all kinds of events. To implement such a simulator, it is necessary to develop a method that would allow recreating a realistic world in one passage with cameras (monocular) installed on board the vehicle. Based on this, the purpose of this work is to develop an intelligent vehicle recognition system using convolutional neural networks, which allows you to create mesh objects for further placement in the simulator. It is important to note that the resulting objects should be optimal in size so as not to overload the system, since a large number of road infrastructure objects are stored there. Also, neural complexity should not be excessive. In this paper, the general concept and classification of convolutional neural networks are given, which allow solving the problem of recognizing 3D objects in images. Based on the analysis, the existing neural network architectures do not solve the problems mentioned above. In this connection, the authors first of all carried out the design of the system according to the methodology of modeling business processes, and also modified and developed the architecture of the neural network, which allows classifying objects with sufficient accuracy, obtaining optimized mesh objects and reducing computational complexity. The methods proposed in this paper are used in a simulator of a realistic urban environment, which reduces the time and computational costs when training unmanned transport systems. [ABSTRACT FROM AUTHOR]

High-efficiency video coding (HEVC/H.265) is one of the most widely used video coding standards. HEVC introduces a quad-tree coding unit (CU) partition structure to improve video compression efficiency. The determination of the optimal CU partition is achieved through the brute-force search rate-distortion optimization method, which may result in high encoding complexity and hardware implementation challenges. To address this problem, this paper proposes a method that combines convolutional neural networks (CNN) with joint texture recognition to reduce encoding complexity. First, a classification decision method based on the global and local texture features of the CU is proposed, efficiently dividing the CU into smooth and complex texture regions. Second, for the CUs in smooth texture regions, the partition is determined by terminating early. For the CUs in complex texture regions, a proposed CNN is used for predictive partitioning, thus avoiding the traditional recursive approach. Finally, combined with texture classification, the proposed CNN achieves a good balance between the coding complexity and the coding performance. The experimental results demonstrate that the proposed algorithm reduces computational complexity by 61.23%, while only increasing BD-BR by 1.86% and decreasing BD-PSNR by just 0.09 dB. [ABSTRACT FROM AUTHOR]

Cattle behavior classification technology holds a crucial position within the realm of smart cattle farming. Addressing the requisites of cattle behavior classification in the agricultural sector, this paper presents a novel cattle behavior classification network tailored for intricate environments. This network amalgamates the capabilities of CNN and Bi-LSTM. Initially, a data collection method is devised within an authentic farm setting, followed by the delineation of eight fundamental cattle behaviors. The foundational step involves utilizing VGG16 as the cornerstone of the CNN network, thereby extracting spatial feature vectors from each video data sequence. Subsequently, these features are channeled into a Bi-LSTM classification model, adept at unearthing semantic insights from temporal data in both directions. This process ensures precise recognition and categorization of cattle behaviors. To validate the model's efficacy, ablation experiments, generalization effect assessments, and comparative analyses under consistent experimental conditions are performed. These investigations, involving module replacements within the classification model and comprehensive analysis of ablation experiments, affirm the model's effectiveness. The self-constructed dataset about cattle is subjected to evaluation using cross-entropy loss, assessing the model's generalization efficacy across diverse subjects and viewing perspectives. Classification performance accuracy is quantified through the application of a confusion matrix. Furthermore, a set of comparison experiments is conducted, involving three pertinent deep learning models: MASK-RCNN, CNN-LSTM, and EfficientNet-LSTM. The outcomes of these experiments unequivocally substantiate the superiority of the proposed model. Empirical results underscore the CNN-Bi-LSTM model's commendable performance metrics: achieving 94.3% accuracy, 94.2% precision, and 93.4% recall while navigating challenges such as varying light conditions, occlusions, and environmental influences. The objective of this study is to employ a fusion of CNN and Bi-LSTM to autonomously extract features from multimodal data, thereby addressing the challenge of classifying cattle behaviors within intricate scenes. By surpassing the constraints imposed by conventional methodologies and the analysis of single-sensor data, this approach seeks to enhance the precision and generalizability of cattle behavior classification. The consequential practical, economic, and societal implications for the agricultural sector are of considerable significance. [ABSTRACT FROM AUTHOR]

Optimizing the performance of heating, ventilation, and air-conditioning (HVAC) systems is critical in today's energy-conscious world. Fan coil units (FCUs) play a critical role in providing comfort in various environments as an important component of HVAC systems. However, FCUs often experience failures that affect their efficiency and increase their energy consumption. In this context, deep learning (DL)-based fault detection offers a promising solution. By detecting faults early and preventing system failures, the efficiency of FCUs can be improved. This paper explores DL models as fault detectors for FCUs to enable smarter and more energy-efficient hotel buildings. We tested three contemporary DL modeling approaches: convolutional neural network (CNN), long short-term memory network (LSTM), and a combination of CNN and gated recurrent unit (GRU). The random forest model (RF) was additionally developed as a baseline benchmark. The fault detectors were tested on a real-world dataset obtained from the sensory measurement system installed in a hotel and additionally supplemented with simulated data via a physical model developed in TRNSYS. Three representative FCU faults, namely, a stuck valve, a reduction in airflow, and an FCU outage, were simulated with a much larger dataset than is typically utilized in similar studies. The results showed that the hybrid model, integrating CNN and GRU, performed best for all three observed faults. DL-based fault detectors outperformed the baseline RF model, confirming these solutions as viable components for energy-efficient hotels. [ABSTRACT FROM AUTHOR]

This paper presents a novel method based on a convolutional neural network to recover thermal time constants from a temperature–time curve after thermal excitation. The thermal time constants are then used to detect the pathological states of the skin. The thermal system is modeled as a Foster Network consisting of R-C thermal elements. Each component is represented by a time constant and an amplitude that can be retrieved using the deep learning system. The presented method was verified on artificially generated training data and then tested on real, measured thermographic signals from a patient suffering from psoriasis. The results show proper estimation both in time constants and in temperature evaluation over time. The error of the recovered time constants is below 1% for noiseless input data, and it does not exceed 5% for noisy signals. [ABSTRACT FROM AUTHOR]

Neurological disorders have an extreme impact on global health, affecting an estimated one billion individuals worldwide. According to the World Health Organization (WHO), these neurological disorders contribute to approximately six million deaths annually, representing a significant burden. Early and accurate identification of brain pathological features in electroencephalogram (EEG) recordings is crucial for the diagnosis and management of these disorders. However, manual evaluation of EEG recordings is not only time-consuming but also requires specialized skills. This problem is exacerbated by the scarcity of trained neurologists in the healthcare sector, especially in low- and middle-income countries. These factors emphasize the necessity for automated diagnostic processes. With the advancement of machine learning algorithms, there is a great interest in automating the process of early diagnoses using EEGs. Therefore, this paper presents a novel deep learning model consisting of two distinct paths, WaveNet–Long Short-Term Memory (LSTM) and LSTM, for the automatic detection of abnormal raw EEG data. Through multiple ablation experiments, we demonstrated the effectiveness and importance of all parts of our proposed model. The performance of our proposed model was evaluated using TUH abnormal EEG Corpus V.2.0.0. (TUAB) and achieved a high classification accuracy of 88.76%, which is higher than in the existing state-of-the-art research studies. Moreover, we demonstrated the generalization of our proposed model by evaluating it on another independent dataset, TUEP, without any hyperparameter tuning or adjustment. The obtained accuracy was 97.45% for the classification between normal and abnormal EEG recordings, confirming the robustness of our proposed model. [ABSTRACT FROM AUTHOR]

This paper explores the utilization of smart device sensors for the purpose of vehicle recognition. Currently a ubiquitous aspect of people's lives, smart devices can conveniently record details about walking, biking, jogging, and stepping, including physiological data, via often built-in phone activity recognition processes. This paper examines research on intelligent transportation systems to uncover how smart device sensor data may be used for vehicle recognition research, and fit within its growing body of literature. Here, we use the accelerometer and gyroscope, which can be commonly found in a smart phone, to detect the class of a vehicle. We collected data from cars, buses, trains, and bikes using a smartphone, and we designed a 1D CNN model leveraging the residual connection for vehicle recognition. The model achieved more than 98% accuracy in prediction. Moreover, we also provide future research directions based on our study. [ABSTRACT FROM AUTHOR]

High-precision land cover maps of remote sensing images based on an intelligent extraction method are an important research field for many scholars. In recent years, deep learning represented by convolutional neural networks has been introduced into the field of land cover remote sensing mapping. In view of the problem that a convolution operation is good at extracting local features but has limitations in modeling long-distance dependence relationships, a semantic segmentation network, DE-UNet, with a dual encoder is proposed in this paper. The Swin Transformer and convolutional neural network are used to design the hybrid architecture. The Swin Transformer pays attention to multi-scale global features and learns local features through the convolutional neural network. Integrated features take into account both global and local context information. In the experiment, remote sensing images from UAVs were used to test three deep learning models including DE-UNet. DE-UNet achieved the highest classification accuracy, and the average overall accuracy was 0.28% and 4.81% higher than UNet and UNet++, respectively. It shows that the introduction of a Transformer enhances the model fitting ability. [ABSTRACT FROM AUTHOR]

Structural health monitoring (SHM) systems are used to analyze the health of infrastructures such as bridges, using data from various types of sensors. While SHM systems consist of various stages, feature extraction and pattern recognition steps are the most important. Consequently, signal processing techniques in the feature extraction stage and machine learning algorithms in the pattern recognition stage play an effective role in analyzing the health of bridges. In other words, there exists a plethora of signal processing techniques and machine learning algorithms, and the selection of the appropriate technique/algorithm is guided by the limitations of each technique/algorithm. The selection also depends on the requirements of SHM in terms of damage identification level and operating conditions. This has provided the motivation to conduct a Systematic literature review (SLR) of feature extraction techniques and pattern recognition algorithms for the structural health monitoring of bridges. The existing literature reviews describe the current trends in the field with different focus aspects. However, a systematic literature review that presents an in-depth comparative study of different applications of machine learning algorithms in the field of SHM of bridges does not exist. Furthermore, there is a lack of analytical studies that investigate the SHM systems in terms of several design considerations including feature extraction techniques, analytical approaches (classification/ regression), operational functionality levels (diagnosis/prognosis) and system implementation techniques (data-driven/model-based). Consequently, this paper identifies 45 recent research practices (during 2016–2023), pertaining to feature extraction techniques and pattern recognition algorithms in SHM for bridges through an SLR process. First, the identified research studies are classified into three different categories: supervised learning algorithms, neural networks and a combination of both. Subsequently, an in-depth analysis of various machine learning algorithms is performed in each category. Moreover, the analysis of selected research studies (total = 45) in terms of feature extraction techniques is made, and 25 different techniques are identified. Furthermore, this article also explores other design considerations like analytical approaches in the pattern recognition process, operational functionality and system implementation. It is expected that the outcomes of this research may facilitate the researchers and practitioners of the domain during the selection of appropriate feature extraction techniques, machine learning algorithms and other design considerations according to the SHM system requirements. [ABSTRACT FROM AUTHOR]

There are some irregular and disordered noise points in large-scale point clouds, and the accuracy of existing large-scale point cloud classification methods still needs further improvement. This paper proposes a network named MFTR-Net, which considers the local point cloud's eigenvalue calculation. The eigenvalues of 3D point cloud data and the 2D eigenvalues of projected point clouds on different planes are calculated to express the local feature relationship between adjacent point clouds. A regular point cloud feature image is constructed and inputs into the designed convolutional neural network. The network adds TargetDrop to be more robust. The experimental result shows that our methods can learn more high-dimensional feature information, further improving point cloud classification, and our approach can achieve 98.0% accuracy with the Oakland 3D dataset. [ABSTRACT FROM AUTHOR]

Speaker Recognition (SR) is a common task in AI-based sound analysis, involving structurally different methodologies such as Deep Learning or "traditional" Machine Learning (ML). In this paper, we compared and explored the two methodologies on the DEMoS dataset consisting of 8869 audio files of 58 speakers in different emotional states. A custom CNN is compared to several pre-trained nets using image inputs of spectrograms and Cepstral-temporal (MFCC) graphs. AML approach based on acoustic feature extraction, selection and multi-class classification by means of a Naïve Bayes model is also considered. Results show how a custom, less deep CNN trained on grayscale spectrogram images obtain the most accurate results, 90.15% on grayscale spectrograms and 83.17% on colored MFCC. AlexNet provides comparable results, reaching 89.28% on spectrograms and 83.43% on MFCC.The Naïve Bayes classifier provides a 87.09% accuracy and a 0.985 average AUC while being faster to train and more interpretable. Feature selection shows how F0, MFCC and voicing-related features are the most characterizing for this SR task. The high amount of training samples and the emotional content of the DEMoS dataset better reflect a real case scenario for speaker recognition, and account for the generalization power of the models. [ABSTRACT FROM AUTHOR]

The tremendous growth in online activity and the Internet of Things (IoT) led to an increase in cyberattacks. Malware infiltrated at least one device in almost every household. Various malware detection methods that use shallow or deep IoT techniques were discovered in recent years. Deep learning models with a visualization method are the most commonly and popularly used strategy in most works. This method has the benefit of automatically extracting features, requiring less technical expertise, and using fewer resources during data processing. Training deep learning models that generalize effectively without overfitting is not feasible or appropriate with large datasets and complex architectures. In this paper, a novel ensemble model, Stacked Ensemble—autoencoder, GRU, and MLP or SE-AGM, composed of three light-weight neural network models—autoencoder, GRU, and MLP—that is trained on the 25 essential and encoded extracted features of the benchmark MalImg dataset for classification was proposed. The GRU model was tested for its suitability in malware detection due to its lesser usage in this domain. The proposed model used a concise set of malware features for training and classifying the malware classes, which reduced the time and resource consumption in comparison to other existing models. The novelty lies in the stacked ensemble method where the output of one intermediate model works as input for the next model, thereby refining the features as compared to the general notion of an ensemble approach. Inspiration was drawn from earlier image-based malware detection works and transfer learning ideas. To extract features from the MalImg dataset, a CNN-based transfer learning model that was trained from scratch on domain data was used. Data augmentation was an important step in the image processing stage to investigate its effect on classifying grayscale malware images in the MalImg dataset. SE-AGM outperformed existing approaches on the benchmark MalImg dataset with an average accuracy of 99.43%, demonstrating that our method was on par with or even surpassed them. [ABSTRACT FROM AUTHOR]

Convolutional neural network (CNN) is widely deployed on edge devices, performing tasks such as objective detection, image recognition and acoustic recognition. However, the limited resources and strict power constraints of edge devices pose a great challenge to applying the computationally intensive CNN models. In addition, for the edge applications with real-time requirements, such as real-time computing (RTC) systems, the computations need to be completed considering the required timing constraint, so it is more difficult to trade off between computational latency and power consumption. In this paper, we propose a low-power CNN accelerator for edge inference of RTC systems, where the computations are operated in a column-wise manner, to realize an immediate computation for the currently available input data. We observe that most computations of some CNN kernels in deep layers can be completed in multiple cycles, while not affecting the overall computational latency. Thus, we present a multi-cycle scheme to conduct the column-wise convolutional operations to reduce the hardware resource and power consumption. We present hardware architecture for the multi-cycle scheme as a domain-specific CNN architecture, which is then implemented in a 65 nm technology. We prove our proposed approach realizes up to 8.45%, 49.41% and 50.64% power reductions for LeNet, AlexNet and VGG16, respectively. The experimental results show that our approach tends to cause a larger power reduction for the CNN models with greater depth, larger kernels and more channels. [ABSTRACT FROM AUTHOR]

The Internet of Medical Things (IoMT) has revolutionized Ambient Assisted Living (AAL) by interconnecting smart medical devices. These devices generate a large amount of data without human intervention. Learning-based sophisticated models are required to extract meaningful information from this massive surge of data. In this context, Deep Neural Network (DNN) has been proven to be a powerful tool for disease detection. Pulmonary Embolism (PE) is considered the leading cause of death disease, with a death toll of 180,000 per year in the US alone. It appears due to a blood clot in pulmonary arteries, which blocks the blood supply to the lungs or a part of the lung. An early diagnosis and treatment of PE could reduce the mortality rate. Doctors and radiologists prefer Computed Tomography (CT) scans as a first-hand tool, which contain 200 to 300 images of a single study for diagnosis. Most of the time, it becomes difficult for a doctor and radiologist to maintain concentration going through all the scans and giving the correct diagnosis, resulting in a misdiagnosis or false diagnosis. Given this, there is a need for an automatic Computer-Aided Diagnosis (CAD) system to assist doctors and radiologists in decision-making. To develop such a system, in this paper, we proposed a deep learning framework based on DenseNet201 to classify PE into nine classes in CT scans. We utilized DenseNet201 as a feature extractor and customized fully connected decision-making layers. The model was trained on the Radiological Society of North America (RSNA)-Pulmonary Embolism Detection Challenge (2020) Kaggle dataset and achieved promising results of 88%, 88%, 89%, and 90% in terms of the accuracy, sensitivity, specificity, and Area Under the Curve (AUC), respectively. [ABSTRACT FROM AUTHOR]

Speech reflects people's mental state and using a microphone sensor is a potential method for human–computer interaction. Speech recognition using this sensor is conducive to the diagnosis of mental illnesses. The gender difference of speakers affects the process of speech emotion recognition based on specific acoustic features, resulting in the decline of emotion recognition accuracy. Therefore, we believe that the accuracy of speech emotion recognition can be effectively improved by selecting different features of speech for emotion recognition based on the speech representations of different genders. In this paper, we propose a speech emotion recognition method based on gender classification. First, we use MLP to classify the original speech by gender. Second, based on the different acoustic features of male and female speech, we analyze the influence weights of multiple speech emotion features in male and female speech, and establish the optimal feature sets for male and female emotion recognition, respectively. Finally, we train and test CNN and BiLSTM, respectively, by using the male and the female speech emotion feature sets. The results show that the proposed emotion recognition models have an advantage in terms of average recognition accuracy compared with gender-mixed recognition models. [ABSTRACT FROM AUTHOR]

The integration of solar energy with a power system brings great economic and environmental benefits. However, the high penetration of solar power is challenging due to the operation and planning of the existing power system owing to the intermittence and randomicity of solar power generation. Achieving accurate predictions for power generation is important to provide high-quality electric energy for end-users. Therefore, in this paper, we introduce a deep learning-based dual-stream convolutional neural network (CNN) and long short-term nemory (LSTM) network followed by a self-attention mechanism network (DSCLANet). Here, CNN is used to learn spatial patterns and LSTM is incorporated for temporal feature extraction. The output spatial and temporal feature vectors are then fused, followed by a self-attention mechanism to select optimal features for further processing. Finally, fully connected layers are incorporated for short-term solar power prediction. The performance of DSCLANet is evaluated on DKASC Alice Spring solar datasets, and it reduces the error rate up to 0.0136 MSE, 0.0304 MAE, and 0.0458 RMSE compared to recent state-of-the-art methods. [ABSTRACT FROM AUTHOR]

This paper describes the process of developing a classification model for the effective detection of malignant melanoma, an aggressive type of cancer in skin lesions. Primary focus is given on fine-tuning and improving a state-of-the-art convolutional neural network (CNN) to obtain the optimal ROC-AUC score. The study investigates a variety of artificial intelligence (AI) clustering techniques to train the developed models on a combined dataset of images across data from the 2019 and 2020 IIM-ISIC Melanoma Classification Challenges. The models were evaluated using varying cross-fold validations, with the highest ROC-AUC reaching a score of 99.48%. [ABSTRACT FROM AUTHOR]

The key to autonomous navigation in unmanned systems is the ability to recognize static and moving objects in the environment and to support the task of predicting the future state of the environment, avoiding collisions, and planning. However, because the existing 3D LiDAR point-cloud moving object segmentation (MOS) convolutional neural network (CNN) models are very complex and have large computation burden, it is difficult to perform real-time processing on embedded platforms. In this paper, we propose a lightweight MOS network structure based on LiDAR point-cloud sequence range images with only 2.3 M parameters, which is 66% less than the state-of-the-art network. When running on RTX 3090 GPU, the processing time is 35.82 ms per frame and it achieves an intersection-over-union(IoU) score of 51.3% on the SemanticKITTI dataset. In addition, the proposed CNN successfully runs the FPGA platform using an NVDLA-like hardware architecture, and the system achieves efficient and accurate moving-object segmentation of LiDAR point clouds at a speed of 32 fps, meeting the real-time requirements of autonomous vehicles. [ABSTRACT FROM AUTHOR]

Due to their robustness, versatility and performance, induction motors (IMs) have been widely used in many industrial applications. Despite their characteristics, these machines are not immune to failures. In this sense, breakage of the rotor bars (BRB) is a common fault, which is mainly related to the high currents flowing along those bars during start-up. In order to reduce the stresses that could lead to the appearance of these faults, the use of soft starters is becoming usual. However, these devices introduce additional components in the current and flux signals, affecting the evolution of the fault-related patterns and so making the fault diagnosis process more difficult. This paper proposes a new method to automatically classify the rotor health state in IMs driven by soft starters. The proposed method relies on obtaining the Persistence Spectrum (PS) of the start-up stray-flux signals. To obtain a proper dataset, Data Augmentation Techniques (DAT) are applied, adding Gaussian noise to the original signals. Then, these PS images are used to train a Convolutional Neural Network (CNN), in order to automatically classify the rotor health state, depending on the severity of the fault, namely: healthy motor, one broken bar and two broken bars. This method has been validated by means of a test bench consisting of a 1.1 kW IM driven by four different soft starters coupled to a DC motor. The results confirm the reliability of the proposed method, obtaining a classification rate of 100.00% when analyzing each model separately and 99.89% when all the models are analyzed at a time. [ABSTRACT FROM AUTHOR]

It is of great significance to accurately detect ships on the ocean. To obtain higher detection performance, many researchers use deep learning to identify ships from images instead of traditional detection methods. Nevertheless, the marine environment is relatively complex, making it quite difficult to determine features of ship targets. In addition, many detection models contain a large amount of parameters, which is not suitable to deploy in devices with limited computing resources. The two problems restrict the application of ship detection. In this paper, firstly, an SAR ship detection dataset is built based on several databases, solving the problem of a small number of ship samples. Then, we integrate the SPP, ASFF, and DIOU-NMS module into original YOLOv3 to improve the ship detection performance. SPP and ASFF help enrich semantic information of ship targets. DIOU-NMS can lower the false alarm. The improved YOLOv3 has 93.37% mAP, 4.11% higher than YOLOv3 on the self-built dataset. Then, we use the MCP method to compress the improved YOLOv3. Under the pruning ratio of 80%, the obtained compressed model has only 6.7 M parameters. Experiments show that MCP outperforms NS and ThiNet. With the size of 26.8 MB, the compact model can run at 15 FPS on an NVIDIA TX2 embedded development board, 4.3 times faster than the baseline model. Our work will contribute to the development and application of ship detection on the sea. [ABSTRACT FROM AUTHOR]

Emotion recognition, or the ability of computers to interpret people's emotional states, is a very active research area with vast applications to improve people's lives. However, most image-based emotion recognition techniques are flawed, as humans can intentionally hide their emotions by changing facial expressions. Consequently, brain signals are being used to detect human emotions with improved accuracy, but most proposed systems demonstrate poor performance as EEG signals are difficult to classify using standard machine learning and deep learning techniques. This paper proposes two convolutional neural network (CNN) models (M1: heavily parameterized CNN model and M2: lightly parameterized CNN model) coupled with elegant feature extraction methods for effective recognition. In this study, the most popular EEG benchmark dataset, the DEAP, is utilized with two of its labels, valence, and arousal, for binary classification. We use Fast Fourier Transformation to extract the frequency domain features, convolutional layers for deep features, and complementary features to represent the dataset. The M1 and M2 CNN models achieve nearly perfect accuracy of 99.89% and 99.22%, respectively, which outperform every previous state-of-the-art model. We empirically demonstrate that the M2 model requires only 2 seconds of EEG signal for 99.22% accuracy, and it can achieve over 96% accuracy with only 125 milliseconds of EEG data for valence classification. Moreover, the proposed M2 model achieves 96.8% accuracy on valence using only 10% of the training dataset, demonstrating our proposed system's effectiveness. Documented implementation codes for every experiment are published for reproducibility. [ABSTRACT FROM AUTHOR]