Search

We present a deep learning-based framework to design and quantify point-of-care sensors. As its proof-of-concept and use-case, we demonstrated a low-cost and rapid paper-based vertical flow assay (VFA) for high sensitivity C-Reactive Protein (hsCRP) testing, a common medical test used for quantifying the degree of inflammation in patients at risk of cardio-vascular disease (CVD). A machine learning-based sensor design framework was developed for two key tasks: (1) to determine an optimal configuration of immunoreaction spots and conditions, spatially-multiplexed on a paper-based sensing membrane, and (2) to accurately infer the target analyte concentration based on the signals of the optimal VFA configuration. Using a custom-designed mobile-phone based VFA reader, a clinical study was performed with 85 human serum samples to characterize the quantification accuracy around the clinically defined cutoffs for CVD risk stratification. Results from blindly-tested VFAs indicate a competitive coefficient of variation of 11.2% with a linearity of R2 = 0.95; in addition to the success in the high-sensitivity CRP range (i.e., 0-10 mg/L), our results further demonstrate a mitigation of the hook-effect at higher CRP concentrations due to the incorporation of antigen capture spots within the multiplexed sensing membrane of the VFA. This paper-based computational VFA that is powered by deep learning could expand access to CVD health screening, and the presented machine learning-enabled sensing framework can be broadly used to design cost-effective and mobile sensors for various point-of-care diagnostics applications.

The paper structure of historical prints is sort of a unique fingerprint. Paper with the same origin shows similar chain line distances. As the manual measurement of chain line distances is time consuming, the automatic detection of chain lines is beneficial. We propose an end-to-end trainable deep learning method for segmentation and parameterization of chain lines in transmitted light images of German prints from the 16th Century. We trained a conditional generative adversarial network with a multitask loss for line segmentation and line parameterization. We formulated a fully differentiable pipeline for line coordinates’ estimation that consists of line segmentation, horizontal line alignment, and 2D Fourier filtering of line segments, line region proposals, and differentiable line fitting. We created a dataset of high-resolution transmitted light images of historical prints with manual line coordinate annotations. Our method shows superior qualitative and quantitative chain line detection results with high accuracy and reliability on our historical dataset in comparison to competing methods. Further, we demonstrated that our method achieves a low error of less than 0.7 mm in comparison to manually measured chain line distances.

Background Bone age assessment (BAA) is an important tool for diagnosis and in determining the time of treatment in a number of pediatric clinical scenarios, as well as in legal settings where it is used to estimate the chronological age of an individual where valid documents are lacking. Traditional methods for BAA suffer from drawbacks, such as exposing juveniles to radiation, intra- and interrater variability, and the time spent on the assessment. The employment of automated methods such as deep learning and the use of magnetic resonance imaging (MRI) can address these drawbacks and improve the assessment of age. Objective The aim of this paper is to propose an automated approach for age assessment of youth and young adults in the age range when the length growth ceases and growth zones are closed (14-21 years of age) by employing deep learning using MRI of the knee. Methods This study carried out MRI examinations of the knee of 402 volunteer subjects—221 males (55.0%) and 181 (45.0%) females—aged 14-21 years. The method comprised two convolutional neural network (CNN) models: the first one selected the most informative images of an MRI sequence, concerning age-assessment purposes; these were then used in the second module, which was responsible for the age estimation. Different CNN architectures were tested, both training from scratch and employing transfer learning. Results The CNN architecture that provided the best results was GoogLeNet pretrained on the ImageNet database. The proposed method was able to assess the age of male subjects in the range of 14-20.5 years, with a mean absolute error (MAE) of 0.793 years, and of female subjects in the range of 14-19.5 years, with an MAE of 0.988 years. Regarding the classification of minors—with the threshold of 18 years of age—an accuracy of 98.1% for male subjects and 95.0% for female subjects was achieved. Conclusions The proposed method was able to assess the age of youth and young adults from 14 to 20.5 years of age for male subjects and 14 to 19.5 years of age for female subjects in a fully automated manner, without the use of ionizing radiation, addressing the drawbacks of traditional methods.

Graphical abstract, Highlights • COVID-19 reveals radiological signatures that can be detected using chest X-rays. • The evaluation of radiological signatures is a time-consuming and error-prone task. • Therefore, there is a need to automate the analysis of chest X-rays. • An automatic analysis of chest X-rays is achieved using deep learning models., The most widely used novel coronavirus (COVID-19) detection technique is a real-time polymerase chain reaction (RT-PCR). However, RT-PCR kits are costly and take 6-9 hours to confirm infection in the patient. Due to less sensitivity of RT-PCR, it provides high false-negative results. To resolve this problem, radiological imaging techniques such as chest X-rays and computed tomography (CT) are used to detect and diagnose COVID-19. In this paper, chest X-rays is preferred over CT scan. The reason behind this is that X-rays machines are available in most of the hospitals. X-rays machines are cheaper than the CT scan machine. Besides this, X-rays has low ionizing radiations than CT scan. COVID-19 reveals some radiological signatures that can be easily detected through chest X-rays. For this, radiologists are required to analyze these signatures. However, it is a time-consuming and error-prone task. Hence, there is a need to automate the analysis of chest X-rays. The automatic analysis of chest X-rays can be done through deep learning-based approaches, which may accelerate the analysis time. These approaches can train the weights of networks on large datasets as well as fine-tuning the weights of pre-trained networks on small datasets. However, these approaches applied to chest X-rays are very limited. Hence, the main objective of this paper is to develop an automated deep transfer learning-based approach for detection of COVID-19 infection in chest X-rays by using the extreme version of the Inception (Xception) model. Extensive comparative analyses show that the proposed model performs significantly better as compared to the existing models.

Presently, Supervisory Control and Data Acquisition (SCADA) systems are broadly adopted in remote monitoring large-scale production systems and modern power grids. However, SCADA systems are continuously exposed to various heterogeneous cyberattacks, making the detection task using the conventional intrusion detection systems (IDSs) very challenging. Furthermore, conventional security solutions, such as firewalls, and antivirus software, are not appropriate for fully protecting SCADA systems because they have distinct specifications. Thus, accurately detecting cyber-attacks in critical SCADA systems is undoubtedly indispensable to enhance their resilience, ensure safe operations, and avoid costly maintenance. The overarching goal of this paper is to detect malicious intrusions that already detoured traditional IDS and firewalls. In this paper, a stacked deep learning method is introduced to identify malicious attacks targeting SCADA systems. Specifically, we investigate the feasibility of a deep learning approach for intrusion detection in SCADA systems. Real data sets from two laboratory-scale SCADA systems, a two-line three-bus power transmission system and a gas pipeline are used to evaluate the proposed method’s performance. The results of this investigation show the satisfying detection performance of the proposed stacked deep learning approach. This study also showed that the proposed approach outperformed the standalone deep learning models and the state-of-the-art algorithms, including Nearest neighbor, Random forests, Naive Bayes, Adaboost, Support Vector Machine, and oneR. Besides detecting the malicious attacks, we also investigate the feature importance of the cyber-attacks detection process using the Random Forest procedure, which helps design more parsimonious models.

Clinical studies from WHO have demonstrated that only 50–70% of patients adhere properly to prescribed drug therapy. Such adherence failure can impact therapeutic efficacy for the patients in question and compromises data quality around the population-level efficacy of the drug for the indications targeted. In this study, we applied various ensemble learning and deep learning models to predict medication adherence among patients. Our contribution to this endeavour involves targeting the problem of adherence prediction for a particularly challenging class of patients who self-administer injectable medication at home. Our prediction pipeline, based on event history, comprises a connected sharps bin which aims to help patients better manage their condition and improve outcomes. In other words, the efficiency of interventions can be significantly improved by prioritizing the patients who are most likely to be non-adherent. The collected data comprising a rich event feature set may be exploited for the purposes of predicting the status of the next adherence state for individual patients. This paper reports on how this concept can be realized through an investigation using a wide range of ensemble learning and deep learning models on a real-world dataset collected from such a system. The dataset investigated comprises 342,174 historic injection disposal records collected over the course of more than 5 years. A comprehensive comparison of different models is given in this paper. Moreover, we demonstrate that the selected best performer, long short-term memory (LSTM), generalizes well by deploying it in a true future testing dataset. The proposed end-to-end pipeline is capable of predicting patient failure in adhering to their therapeutic regimen with 77.35 % accuracy (Specificity: 78.28 %, Sensitivity: 76.42%, Precision: 77.87%,F1 score: 0.7714, ROC AUC: 0.8390).

The Coronavirus disease (COVID-19) is an infectious disease caused by the newly discovered Severe Acute Respiratory Syndrome Coronavirus two (SARS-CoV-2). Most of the people do not have the acquired immunity to fight this virus. There is no specific treatment or medicine to cure the disease. The effects of this disease appear to vary from individual to individual, right from mild cough, fever to respiratory disease. It also leads to mortality in many people. As the virus has a very rapid transmission rate, the entire world is in distress. The control and prevention of this disease has evolved as an urgent and critical issue to be addressed through technological solutions. The Healthcare industry therefore needs support from the domain of artificial intelligence (AI). AI has the inherent capability of imitating the human brain and assisting in decision-making support by automatically learning from input data. It can process huge amounts of data quickly without getting tiresome and making errors. AI technologies and tools significantly relieve the burden of healthcare professionals. In this paper, we review the critical role of AI in responding to different research challenges around the COVID-19 crisis. A sample implementation of a powerful probabilistic machine learning (ML) algorithm for assessment of risk levels of individuals is incorporated in this paper. Other pertinent application areas such as surveillance of people and hotspots, mortality prediction, diagnosis, prognostic assistance, drug repurposing and discovery of protein structure, and vaccine are presented. The paper also describes various challenges that are associated with the implementation of AI-based tools and solutions for practical use.

Background and aims Artificial Intelligence (AI) shows extensive capabilities to impact different healthcare areas during the COVID-19 pandemic positively. This paper tries to assess the capabilities of AI in the field of cardiology during the COVID-19 pandemic. This technology is useful to provide advanced technology-based treatment in cardiology as it can help analyse and measure the functioning of the human heart. Methods We have studied a good number of research papers on Artificial Intelligence on cardiology during the COVID-19 pandemic to identify its significant benefits, applications, and future scope. AI uses artificial neuronal networks (ANN) to predict. In cardiology, it is used to predict the survival of a COVID-19 patient from heart failure. Results AI involves complex algorithms for predicting somewhat successful diagnosis and treatments. This technology uses different techniques, such as cognitive computing, deep learning, and machine learning. It is incorporated to make a decision and resolve complex challenges. It can focus on a large number of diseases, their causes, interactions, and prevention during the COVID-19 pandemic. This paper introduces AI-based care and studies its need in the field of cardiology. Finally, eleven major applications of AI in cardiology during the COVID-19 pandemic are identified and discussed. Conclusions Cardiovascular diseases are one of the major causes of death in human beings, and it is increasing for the last few years. Cardiology patients' treatment is expensive, so this technology is introduced to provide a new pathway and visualise cardiac anomalies. AI is used to identify novel drug therapies and improve the efficiency of a physician. It is precise to predict the outcome of the COVID-19 patient from cardiac-based algorithms. Artificial Intelligence is becoming a popular feature of various engineering and healthcare sectors, is thought for providing a sustainable treatment platform. During the COVID-19 pandemic, this technology digitally controls some processes of treatments.

The purpose of current paper is to create a smart and effective tool for telemedicine to early detect and diagnose COVID-19 disease and therefore help to manage Pandemic Crisis (MCPC) in Sultanate of Oman, as a tool for future pandemic containment. In this paper, we used tools to create robust models in real-time to support Telemedicine, it is Machine Learning (ML), Deep Learning (DL), Convolutional Neural Networks using Tensorflow (CNN-TF), and CNN Deployment. These models will assist telemedicine, 1) developing Automated Medical Immediate Diagnosis service (AMID). 2) Analysis of Chest X-rays image (CXRs). 3) Simplifying Classification of confirmed cases according to its severity. 4) Overcoming the lack of experience, by improving the performance of medical diagnostics and providing recommendations to the medical staff. The results show that the best Regression among the five Regression models is Random Forest Regression. while the best classification among the eight classification models and Recurrent Neural Network using Tensorflow (RNNTF) is Random Forest classification, and the best Clustering model among two Clustering models is K-Means++. Furthermore, CNN-TF model was able to discriminate between those with positive cases Covid-19 and those with negative cases.

A number of recent papers have shown experimental evidence that suggests it is possible to build highly accurate deep neural network models to detect COVID-19 from chest X-ray images. In this paper, we show that good generalization to unseen sources has not been achieved. Experiments with richer data sets than have previously been used show models have high accuracy on seen sources, but poor accuracy on unseen sources. The reason for the disparity is that the convolutional neural network model, which learns features, can focus on differences in X-ray machines or in positioning within the machines, for example. Any feature that a person would clearly rule out is called a confounding feature. Some of the models were trained on COVID-19 image data taken from publications, which may be different than raw images. Some data sets were of pediatric cases with pneumonia where COVID-19 chest X-rays are almost exclusively from adults, so lung size becomes a spurious feature that can be exploited. In this work, we have eliminated many confounding features by working with as close to raw data as possible. Still, deep learned models may leverage source specific confounders to differentiate COVID-19 from pneumonia preventing generalizing to new data sources (i.e. external sites). Our models have achieved an AUC of 1.00 on seen data sources but in the worst case only scored an AUC of 0.38 on unseen ones. This indicates that such models need further assessment/development before they can be broadly clinically deployed. An example of fine-tuning to improve performance at a new site is given.

Novel coronavirus (COVID-19) is started from Wuhan (City in China), and is rapidly spreading among people living in other countries. Today, around 215 countries are affected by COVID-19 disease. WHO announced approximately number of cases 11,274,600 worldwide. Due to rapidly rising cases daily in the hospitals, there are a limited number of resources available to control COVID-19 disease. Therefore, it is essential to develop an accurate diagnosis of COVID-19 disease. Early diagnosis of COVID-19 patients is important for preventing the disease from spreading to others. In this paper, we proposed a deep learning based approach that can differentiate COVID- 19 disease patients from viral pneumonia, bacterial pneumonia, and healthy (normal) cases. In this approach, deep transfer learning is adopted. We used binary and multi-class dataset which is categorized in four types for experimentation: (i) Collection of 728 X-ray images including 224 images with confirmed COVID-19 disease and 504 normal condition images (ii) Collection of 1428 X-ray images including 224 images with confirmed COVID-19 disease, 700 images with confirmed common bacterial pneumonia, and 504 normal condition images. (iii) Collections of 1442 X- ray images including 224 images with confirmed COVID-19 disease, 714 images with confirmed bacterial and viral pneumonia, and 504 images of normal conditions (iv) Collections of 5232 X- ray images including 2358 images with confirmed bacterial and 1345 with viral pneumonia, and 1346 images of normal conditions. In this paper, we have used nine convolutional neural network based architecture (AlexNet, GoogleNet, ResNet-50, Se-ResNet-50, DenseNet121, Inception V4, Inception ResNet V2, ResNeXt-50, and Se-ResNeXt-50). Experimental results indicate that the pre trained model Se-ResNeXt-50 achieves the highest classification accuracy of 99.32% for binary class and 97.55% for multi-class among all pre-trained models.

In recent times, a phishing attack has become one of the most prominent attacks faced by internet users, governments, and service-providing organizations. In a phishing attack, the attacker(s) collects the client's sensitive data (i.e., user account login details, credit/debit card numbers, etc.) by using spoofed emails or fake websites. Phishing websites are common entry points of online social engineering attacks, including numerous frauds on the websites. In such types of attacks, the attacker(s) create website pages by copying the behavior of legitimate websites and sends URL(s) to the targeted victims through spam messages, texts, or social networking. To provide a thorough understanding of phishing attack(s), this paper provides a literature review of Artificial Intelligence (AI) techniques: Machine Learning, Deep Learning, Hybrid Learning, and Scenario-based techniques for phishing attack detection. This paper also presents the comparison of different studies detecting the phishing attack for each AI technique and examines the qualities and shortcomings of these methodologies. Furthermore, this paper provides a comprehensive set of current challenges of phishing attacks and future research direction in this domain.

Objective: This paper proposes a multiclass deep learning method for the classification of dementia using an informant-based questionnaire. Methods: A deep neural network classification model based on Keras framework is proposed in this paper. To evaluate the advantages of our proposed method, we compared the performance of our model with industry-standard machine learning approaches. We enrolled 6,701 individuals, which were randomly divided into training data sets (6030 participants) and test data sets (671 participants). We evaluated each diagnostic model in the test set using accuracy, precision, recall, and F1-Score. Results: Compared with the seven conventional machine learning algorithms, the DNN showed higher stability and achieved the best accuracy with 0.88, which also showed good results for identifying normal (F1-score = 0.88), mild cognitive impairment (MCI) (F1-score = 0.87), very mild dementia (VMD) (F1-score = 0.77) and Severe dementia (F1-score = 0.94). Conclusion: The deep neural network (DNN) classification model can effectively help doctors accurately screen patients who have normal cognitive function, mild cognitive impairment (MCI), very mild dementia (VMD), mild dementia (Mild), moderate dementia (Moderate), and severe dementia (Severe)., We present a deep neural network (DNN) classification model that can help clinicians accurately screen patients who have normal cognitive function, mild cognitive impairment (MCI), very mild dementia (VMD), mild dementia (Mild), moderate dementia (Moderate), and severe dementia (Severe).

This paper attempts to bring new ideas regarding the use of artificial intelligence in language learning apps. Artificial intelligence has seen a dramatic increase and its utilization in various aspects of data science, information science and various platforms is unprecedented. Its ubiquitousness is evident in smart cities, in online marketing, and data mining, however, its utilization in language learning apps is still somehow neglected. There are a few questions, such as why it is so, etc., this paper attempts to address. The research focuses on several most used language learning apps and the presence of artificial intelligence in them. The findings of the paper are as follows: basically none of the analysed apps uses any kind of machine learning, artificial intelligence or deep learning, and they are mostly based on predefined algorithms that do not utilize the full potential of the computational power we have currently available. The paper also suggests possible solutions and brings practical advice on how to implement artificial intelligence in these apps. The paper is important for any education innovation at the beginning of the 21 century. Without his innovative approach, education will lack sustainability and competitiveness, which will present a serious threat to the whole educational system.

Space information networks is network systems that can receive, transmit, and process spatial information lively. It uses satellites, stratosphere airships, Unmanned Aerial Vehicles, and other platforms as the carrier. It supports high-dynamic, real-time broadband transmission of earth observations and ultra-long-distance, long-delay reliable transmission of deep space exploration. The deeper the network integration, the higher the system’s security concerns and the more likely SINs will be controlled and destroyed in terms of cybersecurity. How to integrate new IT technologies such as artificial intelligence, digital twins, and blockchain to diverse application scenarios of SINs while maintaining SIN cybersecurity will be a long-term critical technical issue. This study is a review of the security issues for space information networks. First, this paper examines space information networks’ security issues and figures out the relationship between the main security threats, services, and mechanisms. Then, this article selects secure routing and anomaly detection from many security technologies to conduct a detailed overview from two perspectives of traditional methods and artificial intelligence. Subsequently, this paper investigates anomaly detection schemes for spatial information networks and proposes a deep learning-based anomaly detection scheme. Finally, we suggest the potential research directions and opening problems of space information network security. Overall, this paper aims to give readers an overview of the newly emerging technologies in space information networks’ security issues and provide inspiration for future exploration.

Objective:Elevated noise levels in positron emission tomography (PET) images lower image quality and quantitative accuracy and are a confounding factor for clinical interpretation. The objective of this paper is to develop a PET image denoising technique based on unsupervised deep learning.Significance:Recent advances in deep learning have ushered in a wide array of novel denoising techniques, several of which have been successfully adapted for PET image reconstruction and post-processing. The bulk of the deep learning research so far has focused on supervised learning schemes, which, for the image denoising problem, require paired noisy and noiseless/low-noise images. This requirement tends to limit the utility of these methods for medical applications as paired training datasets are not always available. Furthermore, to achieve the best-case performance of these methods, it is essential that the datasets for training and subsequent real-world application have consistent image characteristics (e.g. noise, resolution, etc), which is rarely the case for clinical data. To circumvent these challenges, it is critical to develop unsupervised techniques that obviate the need for paired training data.Approach:In this paper, we have adapted Noise2Void, a technique that relies on corrupt images alone for model training, for PET image denoising and assessed its performance using PET neuroimaging data. Noise2Void is an unsupervised approach that uses a blind-spot network design. It requires only a single noisy image as its input, and, therefore, is well-suited for clinical settings. During the training phase, a single noisy PET image serves as both the input and the target. Here we present a modified version of Noise2Void based on a transfer learning paradigm that involves group-level pretraining followed by individual fine-tuning. Furthermore, we investigate the impact of incorporating an anatomical image as a second input to the network.Main Results:We validated our denoising technique using simulation data based on the BrainWeb digital phantom. We show that Noise2Void with pretraining and/or anatomical guidance leads to higher peak signal-to-noise ratios than traditional denoising schemes such as Gaussian filtering, anatomically guided non-local means filtering, and block-matching and 4D filtering. We used the Noise2Noise denoising technique as an additional benchmark. For clinical validation, we applied this method to human brain imaging datasets. The clinical findings were consistent with the simulation results confirming the translational value of Noise2Void as a denoising tool.

As an alternative approach, viseme-based lipreading systems have demonstrated promising performance results in decoding videos of people uttering entire sentences. However, the overall performance of such systems has been significantly affected by the efficiency of the conversion of visemes to words during the lipreading process. As shown in the literature, the issue has become a bottleneck of such systems where the system’s performance can decrease dramatically from a high classification accuracy of visemes (e.g., over 90%) to a comparatively very low classification accuracy of words (e.g., only just over 60%). The underlying cause of this phenomenon is that roughly half of the words in the English language are homophemes, i.e., a set of visemes can map to multiple words, e.g., “time” and “some”. In this paper, aiming to tackle this issue, a deep learning network model with an Attention based Gated Recurrent Unit is proposed for efficient viseme-to-word conversion and compared against three other approaches. The proposed approach features strong robustness, high efficiency, and short execution time. The approach has been verified with analysis and practical experiments of predicting sentences from benchmark LRS2 and LRS3 datasets. The main contributions of the paper are as follows: (1) A model is developed, which is effective in converting visemes to words, discriminating between homopheme words, and is robust to incorrectly classified visemes; (2) the model proposed uses a few parameters and, therefore, little overhead and time are required to train and execute; and (3) an improved performance in predicting spoken sentences from the LRS2 dataset with an attained word accuracy rate of 79.6%—an improvement of 15.0% compared with the state-of-the-art approaches.

As is well-known, defects precisely affect the lives and functions of the machines in which they occur, and even cause potentially catastrophic casualties. Therefore, quality assessment before mounting is an indispensable requirement for factories. Apart from the recognition accuracy, current networks suffer from excessive computing complexity, making it of great difficulty to deploy in the manufacturing process. To address these issues, this paper introduces binary networks into the area of surface defect detection for the first time, for the reason that binary networks prohibitively constrain weight and activation to +1 and −1. The proposed Bi-ShuffleNet and U-BiNet utilize binary convolution layers and activations in low bitwidth, in order to reach comparable performances while incurring much less computational cost. Extensive experiments are conducted on real-life NEU and Magnetic Tile datasets, revealing the least OPs required and little accuracy decline. When classifying the defects, Bi-ShuffleNet yields comparable results to counterpart networks, with at least 2× inference complexity reduction. Defect segmentation results indicate similar observations. Some network design rules in defect detection and binary networks are also summarized in this paper.

Human-robot interaction has received a lot of attention as collaborative robots became widely utilized in many industrial fields. Among techniques for human-robot interaction, collision identification is an indispensable element in collaborative robots to prevent fatal accidents. This paper proposes a deep learning method for identifying external collisions in 6-DoF articulated robots. The proposed method expands the idea of CollisionNet, which was previously proposed for collision detection, to identify the locations of external forces. The key contribution of this paper is uncertainty-aware knowledge distillation for improving the accuracy of a deep neural network. Sample-level uncertainties are estimated from a teacher network, and larger penalties are imposed for uncertain samples during the training of a student network. Experiments demonstrate that the proposed method is effective for improving the performance of collision identification.

This paper’s core objective is to develop and validate a new neurocomputing model to classify document images in particularly demanding hard conditions such as image distortions, image size variance and scale, a huge number of classes, etc. Document classification is a special machine vision task in which document images are categorized according to their likelihood. Document classification is by itself an important topic for the digital office and it has several usages. Additionally, different methods for solving this problem have been presented in various studies, their respectively reached performance is however not yet good enough. This task is very tough and challenging. Thus, a novel, more accurate and precise model is needed. Although the related works do reach acceptable accuracy values for less hard conditions, they generally fully fail in the face of those above-mentioned hard, real-world conditions, including, amongst others, distortions such as noise, blur, low contrast, and shadows. In this paper, a novel deep CNN model is developed, validated and benchmarked with a selection of the most relevant recent document classification models. Additionally, the model’s sensitivity was significantly improved by injecting different artifacts during the training process. In the benchmarking, it does clearly outperform all others by at least 4%, thus reaching more than 96% accuracy.

One of the most common methods for diagnosing coronary artery disease is the use of the coronary artery calcium score CT. However, the current diagnostic method using the coronary artery calcium score CT requires a considerable time, because the radiologist must manually check the CT images one-by-one, and check the exact range. In this paper, three CNN models are applied for 1200 normal cardiovascular CT images, and 1200 CT images in which calcium is present in the cardiovascular system. We conduct the experimental test by classifying the CT image data into the original coronary artery calcium score CT images containing the entire rib cage, the cardiac segmented images that cut out only the heart region, and cardiac cropped images that are created by using the cardiac images that are segmented into nine sub-parts and enlarged. As a result of the experimental test to determine the presence of calcium in a given CT image using Inception Resnet v2, VGG, and Resnet 50 models, the highest accuracy of 98.52% was obtained when cardiac cropped image data was applied using the Resnet 50 model. Therefore, in this paper, it is expected that through further research, both the simple presence of calcium and the automation of the calcium analysis score for each coronary artery calcium score CT will become possible.

With the automotive industry moving towards automated driving, sensing is increasingly important in enabling technology. The virtual sensors allow data fusion from various vehicle sensors and provide a prediction for measurement that is hard or too expensive to measure in another way or in the case of demand on continuous detection. In this paper, virtual sensing is discussed for the case of vehicle suspension control, where information about the relative velocity of the unsprung mass for each vehicle corner is required. The corresponding goal can be identified as a regression task with multi-input sequence input. The hypothesis is that the state-of-art method of Bidirectional Long–Short Term Memory (BiLSTM) can solve it. In this paper, a virtual sensor has been proposed and developed by training a neural network model. The simulations have been performed using an experimentally validated full vehicle model in IPG Carmaker. Simulations provided the reference data which were used for Neural Network (NN) training. The extensive dataset covering 26 scenarios has been used to obtain training, validation and testing data. The Bayesian Search was used to select the best neural network structure using root mean square error as a metric. The best network is made of 167 BiLSTM, 256 fully connected hidden units and 4 output units. Error histograms and spectral analysis of the predicted signal compared to the reference signal are presented. The results demonstrate the good applicability of neural network-based virtual sensors to estimate vehicle unsprung mass relative velocity.

Deep neural networks (DNNs) have introduced novel and useful tools to the machine learning community. Other types of classifiers can potentially make use of these tools as well to improve their performance and generality. This paper reviews the current state of the art for deep learning classifier technologies that are being used outside of deep neural networks. Non-neural network classifiers can employ many components found in DNN architectures. In this paper, we review the feature learning, optimization, and regularization methods that form a core of deep network technologies. We then survey non-neural network learning algorithms that make innovative use of these methods to improve classification performance. Because many opportunities and challenges still exist, we discuss directions that can be pursued to expand the area of deep learning for a variety of classification algorithms.

The literature on coronaviruses counts more than 300,000 publications. Finding relevant papers concerning arbitrary queries is essential to discovery helpful knowledge. Current best information retrieval (IR) use deep learning approaches and need supervised training sets with labeled data, namely to know a priori the queries and their corresponding relevant papers. Creating such labeled datasets is time-expensive and requires prominent experts’ efforts, resources insufficiently available under a pandemic time pressure. We present a new self-supervised solution, called SUBLIMER, that does not require labels to learn to search on corpora of scientific papers for most relevant against arbitrary queries. SUBLIMER is a novel efficient IR engine trained on the unsupervised COVID-19 Open Research Dataset (CORD19), using deep metric learning. The core point of our self-supervised approach is that it uses no labels, but exploits the bibliography citations from papers to create a latent space where their spatial proximity is a metric of semantic similarity, for this reason, it can also be applied to other domains of papers corpora. SUBLIMER, despite is self-supervised, outperforms the Precision@5 (P@5) and Bpref of the state-of-the-art competitors on CORD19, which, differently from our approach, require both labeled datasets and a number of trainable parameters that is an order of magnitude higher than our.

As a complex task, robot sorting has become a research hotspot. In order to enable robots to perform simple, efficient, stable and accurate sorting operations for stacked multi-objects in unstructured scenes, a robot multi-object sorting system is built in this paper. Firstly, the training model of rotating target detection is constructed, and the placement state of five common objects in unstructured scenes is collected as the training set for training. The trained model is used to obtain the position, rotation angle and category of the target object. Then, the instance segmentation model is constructed, and the same data set is made, and the instance segmentation network model is trained. Then, the optimized Mask R-CNN instance segmentation network is used to segment the object surface pixels, and the upper surface point cloud is extracted to calculate the normal vector. Then, the angle obtained by the normal vector of the upper surface and the rotation target detection network is fused with the normal vector to obtain the attitude of the object. At the same time, the grasping order is calculated according to the average depth of the surface. Finally, after the obtained object posture, category and grasping sequence are fused, the performance of the rotating target detection network, the instance segmentation network and the robot sorting system are tested on the established experimental platform. Based on this system, this paper carried out an experiment on the success rate of object capture in a single network and an integrated network. The experimental results show that the multi-object sorting system based on deep learning proposed in this paper can sort stacked objects efficiently, accurately and stably in unstructured scenes.

With the continuous development of artificial intelligence, embedding object detection algorithms into autonomous underwater detectors for marine garbage cleanup has become an emerging application area. Considering the complexity of the marine environment and the low resolution of the images taken by underwater detectors, this paper proposes an improved algorithm based on Mask R-CNN, with the aim of achieving high accuracy marine garbage detection and instance segmentation. First, the idea of dilated convolution is introduced in the Feature Pyramid Network to enhance feature extraction ability for small objects. Secondly, the spatial-channel attention mechanism is used to make features learn adaptively. It can effectively focus attention on detection objects. Third, the re-scoring branch is added to improve the accuracy of instance segmentation by scoring the predicted masks based on the method of Generalized Intersection over Union. Finally, we train the proposed algorithm in this paper on the Transcan dataset, evaluating its effectiveness by various metrics and comparing it with existing algorithms. The experimental results show that compared to the baseline provided by the Transcan dataset, the algorithm in this paper improves the mAP indexes on the two tasks of garbage detection and instance segmentation by 9.6 and 5.0, respectively, which significantly improves the algorithm performance. Thus, it can be better applied in the marine environment and achieve high precision object detection and instance segmentation.

The wide variety of crops in the image of agricultural products and the confusion with the surrounding environment information makes it difficult for traditional methods to extract crops accurately and efficiently. In this paper, an automatic extraction algorithm is proposed for crop images based on Mask RCNN. First, the Fruits 360 Dataset label is set with Labelme. Then, the Fruits 360 Dataset is preprocessed. Next, the data are divided into a training set and a test set. Additionally, an improved Mask RCNN network model structure is established using the PyTorch 1.8.1 deep learning framework, and path aggregation and features are added to the network design enhanced functions, optimized region extraction network, and feature pyramid network. The spatial information of the feature map is saved by the bilinear interpolation method in ROIAlign. Finally, the edge accuracy of the segmentation mask is further improved by adding a micro-fully connected layer to the mask branch of the ROI output, employing the Sobel operator to predict the target edge, and adding the edge loss to the loss function. Compared with FCN and Mask RCNN and other image extraction algorithms, the experimental results demonstrate that the improved Mask RCNN algorithm proposed in this paper is better in the precision, Recall, Average precision, Mean Average Precision, and F1 scores of crop image extraction results.

The real-time detection and counting of rice ears in fields is one of the most important methods for estimating rice yield. The traditional manual counting method has many disadvantages: it is time-consuming, inefficient and subjective. Therefore, the use of computer vision technology can improve the accuracy and efficiency of rice ear counting in the field. The contributions of this article are as follows. (1) This paper establishes a dataset containing 3300 rice ear samples, which represent various complex situations, including variable light and complex backgrounds, overlapping rice and overlapping leaves. The collected images were manually labeled, and a data enhancement method was used to increase the sample size. (2) This paper proposes a method that combines the LC-FCN (localization-based counting fully convolutional neural network) model based on transfer learning with the watershed algorithm for the recognition of dense rice images. The results show that the model is superior to traditional machine learning methods and the single-shot multibox detector (SSD) algorithm for target detection. Moreover, it is currently considered an advanced and innovative rice ear counting model. The mean absolute error (MAE) of the model on the 300-size test set is 2.99. The model can be used to calculate the number of rice ears in the field. In addition, it can provide reliable basic data for rice yield estimation and a rice dataset for research.

In the paper, an attempt was made to use methods of artificial neural networks (ANN) and Fourier transform infrared spectroscopy (FTIR) to identify raspberry powders that are different from each other in terms of the amount and the type of polysaccharide. Spectra in the absorbance function (FTIR) were prepared as well as training sets, taking into account the structure of microparticles acquired from microscopic images with Scanning Electron Microscopy (SEM). In addition to the above, Multi-Layer Perceptron Networks (MLPNs) with a set of texture descriptors (machine learning) and Convolution Neural Network (CNN) with bitmap (deep learning) were devised, which is an innovative attitude to solving this issue. The aim of the paper was to create MLPN and CNN neural models, which are characterized by a high efficiency of classification. It translates into recognizing microparticles (obtaining their homogeneity) of raspberry powders on the basis of the texture of the image pixel.

Corona virus disease also acknowledged as COVID-19 outbreak, a worldwide pandemic is one of the most acute and severe viruses in recent time. The rate of COVID cases rise rapidly around the world. Although vaccines have been developed, deep learning (DL) techniques shown as a useful method for clinical diagnosis and other fields. Deep structured learning also known as Deep learning is method based on artificial neural network with interpretation learning. This paper aims to do a comparative analysis on medical images like computer tomography scans (CT scan) and X-ray by means of different deep learning systems. This analysis discusses about structures developed for COVID-19 analysis via deep learning performances on Inception, VGG, Xception, Resnet models and provide insights and on data sets to train these neural networks. A comparative analysis is done for considering the better deep learning model for detection. The main aim of this paper is to ease medical experts and help them to understand the ways of deep learning techniques and how they can be prospective used to combat COVID- 19.

This paper presents an automatic classification of plastic material’s inorganic pigment using terahertz spectroscopy and convolutional neural networks (CNN). The plastic materials were placed between the THz transmitter and receiver, and the acquired THz signals were classified using a supervised learning approach. A THz frequency band between 0.1–1.2 THz produced a one-dimensional (1D) vector that is almost impossible to classify directly using supervised learning. This paper proposes a novel pre-processing of 1D THz data that transforms 1D data into 2D data, which are processed efficiently using a convolutional neural network. The proposed pre-processing algorithm consists of four steps: peak detection, envelope extraction, and a down-sampling procedure. The last main step introduces the windowing with spectrum dilatation that reorders 1D data into 2D data that can be considered as an image. The spectrum dilation techniques ensure the classifier’s robustness by suppressing measurement bias, reducing the complexity of the THz dataset with negligible loss of accuracy, and speeding up the network classification. The experimental results showed that the proposed approach achieved high accuracy using a CNN classifier, and outperforms 1D classification of THz data using support vector machine, naive Bayes, and other popular classification algorithms.

Respiratory infections corona virus 2-caused inflammatory disorders are CORONAVIRUS DISEASE 2019 (COVID-19) (SARS-CoV-2). A serious corona virus acute disease arose in 2019. Wuhan, China, was the first location to find the virus in December 2019, which has now been spreading all over the world. Recurrent neural networks, together with the use of LSTMs, fail to provide solutions to numerous issues (RNNs). So this paper has proposed RNN with Gated Recurrent Units for the COVID-19 prediction. This paper utilizes system, which was developed to assist nations (the Czech Republic, the United States, India, and Russia) combat the early stages of a newly emerging infection. For instance, the system tracks confirmed and reported cases, and monitors cures and deaths on a daily basis. This was done to allow the relevant parties to have an early grasp of the disastrous damage the lethal virus will bring. The implemented is an ensemble approach of RNN and GRU that work has computed the RMSE value for the different cases such as infected, cure and death across the four different countries.

Large-scale fading models play an important role in estimating radio coverage, optimizing base station deployments and characterizing the radio environment to quantify the performance of wireless networks. In recent times, multi-frequency path loss models are attracting much interest due to their expected support for both sub-6 GHz and higher frequency bands in future wireless networks. Traditionally, linear multi-frequency path loss models like the ABG model have been considered, however such models lack accuracy. The path loss model based on a deep learning approach is an alternative method to traditional linear path loss models to overcome the time-consuming path loss parameters predictions based on the large dataset at new frequencies and new scenarios. In this paper, we proposed a feed-forward deep neural network (DNN) model to predict path loss of 13 different frequencies from 0.8 GHz to 70 GHz simultaneously in an urban and suburban environment in a non-line-of-sight (NLOS) scenario. We investigated a broad range of possible values for hyperparameters to search for the best set of ones to obtain the optimal architecture of the proposed DNN model. The results show that the proposed DNN-based path loss model improved mean square error (MSE) by about 6 dB and achieved higher prediction accuracy R2 compared to the multi-frequency ABG path loss model. The paper applies the XGBoost algorithm to evaluate the importance of the features for the proposed model and the related impact on the path loss prediction. In addition, the effect of hyperparameters, including activation function, number of hidden neurons in each layer, optimization algorithm, regularization factor, batch size, learning rate, and momentum, on the performance of the proposed model in terms of prediction error and prediction accuracy are also investigated.

Edge-cloud collaborative inference can significantly reduce the delay of a deep neural network (DNN) by dividing the network between mobile edge and cloud. However, the in-layer data size of DNN is usually larger than the original data, so the communication time to send intermediate data to the cloud will also increase end-to-end latency. To cope with these challenges, this paper proposes a novel convolutional neural network structure—BBNet—that accelerates collaborative inference from two levels: (1) through channel-pruning: reducing the number of calculations and parameters of the original network, (2) through compressing the feature map at the split point to further reduce the size of the data transmitted. In addition, This paper implemented the BBNet structure based on NVIDIA Nano and the server. Compared with the original network, BBNet’s FLOPs and parameter achieve up to 5.67× and 11.57× on the compression rate, respectively. In the best case, the feature compression layer can reach a bit-compression rate of 512×. Compared with the better bandwidth conditions, BBNet has a more obvious inference delay when the network conditions are poor. For example, when the upload bandwidth is only 20 kb/s, the end-to-end latency of BBNet is increased by 38.89× compared with the cloud-only approach.

Loop Closure Detection (LCD) is an important technique to improve the accuracy of Simultaneous Localization and Mapping (SLAM). In this paper, we propose an LCD algorithm based on binary classification for feature matching between similar images with deep learning, which greatly improves the accuracy of LCD algorithm. Meanwhile, a novel lightweight convolutional neural network (CNN) is proposed and applied to the target detection task of key frames. On this basis, the key frames are binary classified according to their labels. Finally, similar frames are input into the improved lightweight feature matching network based on Transformer to judge whether the current position is loop closure. The experimental results show that, compared with the traditional method, LFM-LCD has higher accuracy and recall rate in the LCD task of indoor SLAM while ensuring the number of parameters and calculation amount. The research in this paper provides a new direction for LCD of robotic SLAM, which will be further improved with the development of deep learning.

Crop variety identification is an essential link in seed detection, phenotype collection and scientific breeding. This paper takes peanut as an example to explore a new method for crop variety identification. Peanut is a crucial oil crop and cash crop. The yield and quality of different peanut varieties are different, so it is necessary to identify and classify different peanut varieties. The traditional image processing method of peanut variety identification needs to extract many features, which has defects such as intense subjectivity and insufficient generalization ability. Based on the deep learning technology, this paper improved the deep convolutional neural network VGG16 and applied the improved VGG16 to the identification and classification task of 12 varieties of peanuts. Firstly, the peanut pod images of 12 varieties obtained by the scanner were preprocessed with gray-scale, binarization, and ROI extraction to form a peanut pod data set with a total of 3365 images of 12 varieties. A series of improvements have been made to VGG16. Remove the F6 and F7 fully connected layers of VGG16. Add Conv6 and Global Average Pooling Layer. The three convolutional layers of conv5 have changed into Depth Concatenation and add the Batch Normalization(BN) layers to the model. Besides, fine-tuning is carried out based on the improved VGG16. We adjusted the location of the BN layers. Adjust the number of filters for Conv6. Finally, the improved VGG16 model's training test results were compared with the other classic models, AlexNet, VGG16, GoogLeNet, ResNet18, ResNet50, SqueezeNet, DenseNet201 and MobileNetv2 verify its superiority. The average accuracy of the improved VGG16 model on the peanut pods test set was 96.7%, which was 8.9% higher than that of VGG16, and 1.6–12.3% higher than that of other classical models. Besides, supplementary experiments were carried out to prove the robustness and generality of the improved VGG16. The improved VGG16 was applied to the identification and classification of seven corn grain varieties with the same method and an average accuracy of 90.1% was achieved. The experimental results show that the improved VGG16 proposed in this paper can identify and classify peanut pods of different varieties, proving the feasibility of a convolutional neural network in variety identification and classification. The model proposed in this experiment has a positive significance for exploring other Crop variety identification and classification.

Recent years have seen the rapid growth of new approaches to optical imaging, with an emphasis on extracting three-dimensional (3D) information from what is normally a two-dimensional (2D) image capture. Perhaps most importantly, the rise of computational imaging enables both new physical layouts of optical components and new algorithms to be implemented. This paper concerns the convergence of two advances: the development of a transparent focal stack imaging system using graphene photodetector arrays, and the rapid expansion of the capabilities of machine learning including the development of powerful neural networks. This paper demonstrates 3D tracking of point-like objects with multilayer feedforward neural networks and the extension to tracking positions of multi-point objects. Computer simulations further demonstrate how this optical system can track extended objects in 3D, highlighting the promise of combining nanophotonic devices, new optical system designs, and machine learning for new frontiers in 3D imaging., Transparent photodetectors based on graphene stacked vertically along the optical axis have shown promising potential for light field reconstruction. Here, the authors develop transparent photodetector arrays and implement a neural network for real-time 3D optical imaging and object tracking.

With COVID-19 cases rising rapidly, deep learning has emerged as a promising diagnosis technique. However, identifying the most accurate models to characterize COVID-19 patients is challenging because comparing results obtained with different types of data and acquisition processes is non-trivial. In this paper we designed, evaluated, and compared the performance of 20 convolutional neutral networks in classifying patients as COVID-19 positive, healthy, or suffering from other pulmonary lung infections based on Chest CT scans, serving as the first to consider the EfficientNet family for COVID-19 diagnosis and employ intermediate activation maps for visualizing model performance. All models are trained and evaluated in Python using 4173 Chest CT images from the dataset entitled "A COVID multiclass dataset of CT scans," with 2168, 758, and 1247 images of patients that are COVID-19 positive, healthy, or suffering from other pulmonary infections, respectively. EfficientNet-B5 was identified as the best model with an F1 score of 0.9769+/-0.0046, accuracy of 0.9759+/-0.0048, sensitivity of 0.9788+/-0.0055, specificity of 0.9730+/-0.0057, and precision of 0.9751 +/- 0.0051. On an alternate 2-class dataset, EfficientNetB5 obtained an accuracy of 0.9845+/-0.0109, F1 score of 0.9599+/-0.0251, sensitivity of 0.9682+/-0.0099, specificity of 0.9883+/-0.0150, and precision of 0.9526 +/- 0.0523. Intermediate activation maps and Gradient-weighted Class Activation Mappings offered human-interpretable evidence of the model's perception of ground-class opacities and consolidations, hinting towards a promising use-case of artificial intelligence-assisted radiology tools. With a prediction speed of under 0.1 seconds on GPUs and 0.5 seconds on CPUs, our proposed model offers a rapid, scalable, and accurate diagnostic for COVID-19., 35 pages, 4 figures, 5 Tables. Paper in review by Expert Systems with Applications

In the inference process of existing deep learning models, it is usually necessary to process the input data level-wise, and impose a corresponding relational inductive bias on each level. This kind of relational inductive bias determines the theoretical performance upper limit of the deep learning method. In the field of sleep stage classification, only a single relational inductive bias is adopted at the same level in the mainstream methods based on deep learning. This will make the feature extraction method of deep learning incomplete and limit the performance of the method. In view of the above problems, a novel deep learning model based on hybrid relational inductive biases is proposed in this paper. It is called CCRRSleepNet. The model divides the single channel Electroencephalogram (EEG) data into three levels: frame, epoch, and sequence. It applies hybrid relational inductive biases from many aspects based on three levels. Meanwhile, multiscale atrous convolution block (MSACB) is adopted in CCRRSleepNet to learn the features of different attributes. However, in practice, the actual performance of the deep learning model depends on the nonrelational inductive biases, so a variety of matching nonrelational inductive biases are adopted in this paper to optimize CCRRSleepNet. The CCRRSleepNet is tested on the Fpz-Cz and Pz-Oz channel data of the Sleep-EDF dataset. The experimental results show that the method proposed in this paper is superior to many existing methods.

In this paper, it is proposed to understand the opinion of the public regarding the policy of demonetization that is implemented recently in India through Aspect-based Sentiment Analysis (ABSA) that predicts the sentiment of specific aspects present in the text. The major aim is to identify the relevant contexts for various aspects. Most of the conventional techniques have adopted attention mechanisms and deep learning concepts that decrease the prediction accuracy and generate huge noise. Another major disadvantage with the attention mechanisms is that the sentiment related to few context words alters with various aspects, and hence it cannot be concluded from itself alone. This paper adopts the optimized deep learning concept for performing the ABSA for demonetization tweets. The proposed model involves various phases such as pre-processing, aspect extraction, polarity feature extraction, and sentiment classification. Initially, the different demonetization tweets collected from the Kaggle dataset are taken. Pre-processing is done with the help of four phases like stop words removal, punctuation removal, lower case conversion, and stemming from minimizing the data to its reduced format. This pre-processed data is further performed with aspect extraction to extract the opinion words. These extracted aspect words are converted to the features with the help of polarity score computation and Word2vec. The weight of the polarity scores is optimized using hybridization of two meta-heuristic algorithms like FireFly Algorithm (FF), and Multi-Verse Optimization (MVO), and the new algorithm is termed as Fire Fly-oriented Multi-Verse Optimizer (FF-MVO). Further, combined features are subjected to a deep learning algorithm called Recurrent Neural Network (RNN). As a modification to the existing RNN, the hidden neurons are optimized by the hybrid FF-MVO, FF-MVO-RNN classifies the positive and negative sentiments. Finally, the comparative analysis of different machine learning algorithms proves the competent performance of the proposed model.

Sentiment prediction remains a challenging and unresolved task in various research fields, including psychology, neuroscience, and computer science. This stems from its high degree of subjectivity and limited input sources that can effectively capture the actual sentiment. This can be even more challenging with only text-based input. Meanwhile, the rise of deep learning and an unprecedented large volume of data have paved the way for artificial intelligence to perform impressively accurate predictions or even human-level reasoning. Drawing inspiration from this, we propose a coverage-based sentiment and subsentence extraction system that estimates a span of input text and recursively feeds this information back to the networks. The predicted subsentence consists of auxiliary information expressing a sentiment. This is an important building block for enabling vivid and epic sentiment delivery (within the scope of this paper) and for other natural language processing tasks such as text summarisation and Q&A. Our approach outperforms the state-of-the-art approaches by a large margin in subsentence prediction (i.e., Average Jaccard scores from 0.72 to 0.89). For the evaluation, we designed rigorous experiments consisting of 24 ablation studies. Finally, our learned lessons are returned to the community by sharing software packages and a public dataset that can reproduce the results presented in this paper., 27 pages, 16 figures

Designing volume visualizations showing various structures of interest is critical to the exploratory analysis of volumetric data. The last few years have witnessed dramatic advances in the use of convolutional neural networks for identification of objects in large image collections. Whereas such machine learning methods have shown superior performance in a number of applications, their direct use in volume visualization has not yet been explored. In this paper, we present a deep-learning-assisted volume visualization to depict complex structures, which are otherwise challenging for conventional approaches. A significant challenge in designing volume visualizations based on the high-dimensional deep features lies in efficiently handling the immense amount of information that deep-learning methods provide. In this paper, we present a new technique that uses spectral methods to facilitate user interactions with high-dimensional features. We also present a new deep-learning-assisted technique for hierarchically exploring a volumetric dataset. We have validated our approach on two electron microscopy volumes and one magnetic resonance imaging dataset.

This paper presents a robust human posture and body parts detection method under a specific application scenario known as in-bed pose estimation. Although the human pose estimation for various computer vision (CV) applications has been studied extensively in the last few decades, the in-bed pose estimation using camera-based vision methods has been ignored by the CV community because it is assumed to be identical to the general purpose pose estimation problems. However, the in-bed pose estimation has its own specialized aspects and comes with specific challenges, including the notable differences in lighting conditions throughout the day and having pose distribution different from the common human surveillance viewpoint. In this paper, we demonstrate that these challenges significantly reduce the effectiveness of the existing general purpose pose estimation models. In order to address the lighting variation challenge, the infrared selective (IRS) image acquisition technique is proposed to provide uniform quality data under various lighting conditions. In addition, to deal with the unconventional pose perspective, a 2- end histogram of oriented gradient (HOG) rectification method is presented. The deep learning framework proves to be the most effective model in human pose estimation; however, the lack of large public dataset for in-bed poses prevents us from using a large network from scratch. In this paper, we explored the idea of employing a pre-trained convolutional neural network (CNN) model trained on large public datasets of general human poses and fine-tuning the model using our own shallow (limited in size and different in perspective and color) in-bed IRS dataset. We developed an IRS imaging system and collected IRS image data from several realistic life-size mannequins in a simulated hospital room environment. A pre-trained CNN called convolutional pose machine (CPM) was fine-tuned for in-bed pose estimation by re-training its specific intermediate layers. Using the HOG rectification method, the pose estimation performance of CPM improved significantly by 26.4% in the probability of correct key-point (PCK) criteria at PCK0.1 compared to the model without such rectification. Even testing with only well aligned in-bed pose images, our fine-tuned model still surpassed the traditionally tuned CNN by another 16.6% increase in pose estimation accuracy., This paper presents a robust human posture and body parts detection method under a specific application scenario, in-bed pose estimation.

Microorganisms such as bacteria and fungi play essential roles in many application fields, like biotechnique, medical technique and industrial domain. Microorganism counting techniques are crucial in microorganism analysis, helping biologists and related researchers quantitatively analyze the microorganisms and calculate their characteristics, such as biomass concentration and biological activity. However, traditional microorganism manual counting methods, such as plate counting method, hemocytometry and turbidimetry, are time-consuming, subjective and need complex operations, which are difficult to be applied in large-scale applications. In order to improve this situation, image analysis is applied for microorganism counting since the 1980s, which consists of digital image processing, image segmentation, image classification and suchlike. Image analysis-based microorganism counting methods are efficient comparing with traditional plate counting methods. In this article, we have studied the development of microorganism counting methods using digital image analysis. Firstly, the microorganisms are grouped as bacteria and other microorganisms. Then, the related articles are summarized based on image segmentation methods. Each part of the article is reviewed by methodologies. Moreover, commonly used image processing methods for microorganism counting are summarized and analyzed to find common technological points. More than 144 papers are outlined in this article. In conclusion, this paper provides new ideas for the future development trend of microorganism counting, and provides systematic suggestions for implementing integrated microorganism counting systems in the future. Researchers in other fields can refer to the techniques analyzed in this paper.

Demand for wind power has grown, and this has increased wind turbine blade (WTB) inspections and defect repairs. This paper empirically investigates the performance of state-of-the-art deep learning algorithms, namely, YOLOv3, YOLOv4, and Mask R-CNN for detecting and classifying defects by type. The paper proposes new performance evaluation measures suitable for defect detection tasks, and these are: Prediction Box Accuracy, Recognition Rate, and False Label Rate. Experiments were carried out using a dataset, provided by the industrial partner, that contains images from WTB inspections. Three variations of the dataset were constructed using different image augmentation settings. Results of the experiments revealed that on average, across all proposed evaluation measures, Mask R-CNN outperformed all other algorithms when transformation-based augmentations (i.e., rotation and flipping) were applied. In particular, when using the best dataset, the mean Weighted Average (mWA) values (i.e., mWA is the average of the proposed measures) achieved were: Mask R-CNN: 86.74%, YOLOv3: 70.08%, and YOLOv4: 78.28%. The paper also proposes a new defect detection pipeline, called Image Enhanced Mask R-CNN (IE Mask R-CNN), that includes the best combination of image enhancement and augmentation techniques for pre-processing the dataset, and a Mask R-CNN model tuned for the task of WTB defect detection and classification.

Automated segmentation methods are critical for early detection, prompt actions, and immediate treatments in reducing disability and death risks of brain infarction. This paper aims to develop a fully automated method to segment the infarct lesions from T1-weighted brain scans. As a key novelty, the proposed method combines variational mode decomposition and deep learning-based segmentation to take advantages of both methods and provide better results. There are three main technical contributions in this paper. First, variational mode decomposition is applied as a pre-processing to discriminate the infarct lesions from unwanted non-infarct tissues. Second, overlapped patches strategy is proposed to reduce the workload of the deep-learning-based segmentation task. Finally, a three-dimensional U-Net model is developed to perform patch-wise segmentation of infarct lesions. A total of 239 brain scans from a public dataset is utilized to develop and evaluate the proposed method. Empirical results reveal that the proposed automated segmentation can provide promising performances with an average dice similarity coefficient (DSC) of 0.6684, intersection over union (IoU) of 0.5022, and average symmetric surface distance (ASSD) of 0.3932, respectively.

Recent studies indicate that detecting radiographic patterns on CT chest scans can yield high sensitivity and specificity for COVID-19 identification. In this paper, we scrutinize the effectiveness of deep learning models for semantic segmentation of pneumonia-infected area segmentation in CT images for the detection of COVID-19. Traditional methods for CT scan segmentation exploit a supervised learning paradigm, so they (a) require large volumes of data for their training, and (b) assume fixed (static) network weights once the training procedure has been completed. Recently, to overcome these difficulties, few-shot learning (FSL) has been introduced as a general concept of network model training using a very small amount of samples. In this paper, we explore the efficacy of few-shot learning in U-Net architectures, allowing for a dynamic fine-tuning of the network weights as new few samples are being fed into the U-Net. Experimental results indicate improvement in the segmentation accuracy of identifying COVID-19 infected regions. In particular, using 4-fold cross-validation results of the different classifiers, we observed an improvement of 5.388 ± 3.046% for all test data regarding the IoU metric and a similar increment of 5.394 ± 3.015% for the F1 score. Moreover, the statistical significance of the improvement obtained using our proposed few-shot U-Net architecture compared with the traditional U-Net model was confirmed by applying the Kruskal-Wallis test (p-value = 0.026).

Fruit maturity is a critical factor in the supply chain, consumer preference, and agriculture industry. Most classification methods on fruit maturity identify only two classes: ripe and unripe, but this paper estimates six maturity stages of papaya fruit. Deep learning architectures have gained respect and brought breakthroughs in unimodal processing. This paper suggests a novel non-destructive and multimodal classification using deep convolutional neural networks that estimate fruit maturity by feature concatenation of data acquired from two imaging modes: visible-light and hyperspectral imaging systems. Morphological changes in the sample fruits can be easily measured with RGB images, while spectral signatures that provide high sensitivity and high correlation with the internal properties of fruits can be extracted from hyperspectral images with wavelength range in between 400 nm and 900 nm—factors that must be considered when building a model. This study further modified the architectures: AlexNet, VGG16, VGG19, ResNet50, ResNeXt50, MobileNet, and MobileNetV2 to utilize multimodal data cubes composed of RGB and hyperspectral data for sensitivity analyses. These multimodal variants can achieve up to 0.90 F1 scores and 1.45% top-2 error rate for the classification of six stages. Overall, taking advantage of multimodal input coupled with powerful deep convolutional neural network models can classify fruit maturity even at refined levels of six stages. This indicates that multimodal deep learning architectures and multimodal imaging have great potential for real-time in-field fruit maturity estimation that can help estimate optimal harvest time and other in-field industrial applications.

Accurate long-term streamflow and flood forecasting have always been an important research direction in hydrology research. Nowadays, climate change, floods, and other anomalies occurring more and more frequently and bringing great losses to society. The prediction of streamflow, especially flood prediction, is important for disaster prevention. Current hydrological models based on physical mechanisms can give accurate predictions of streamflow, but the effective prediction period is only about 1 month in advance, which is too short for decision making. The artificial neural network (ANN) has great potential for predicting runoff and is not only good at handling non-linear data but can also make long-period forecasts. However, most of ANN models are unstable in their predictions when faced with raw flow data, and have excessive errors in predicting extreme flows. Previous studies have shown a link between the El Niño–Southern Oscillation (ENSO) and the streamflow of the Yangtze River. In this paper, we use ENSO and the monthly streamflow data of the Yangtze River from 1952 to 2018 to predict the monthly streamflow of the Yangtze River in two extreme flood years and a small flood year by using deep neural networks. In this paper, three deep neural network frameworks are used: stacked long short-term memory, Conv long short-term memory encoder–decoder long short-term memory and Conv long short-term memory encoder–decoder gate recurrent unit. The results show that the use of ConvLSTM improves the stability of the model and increases the accuracy of the flood prediction. Besides, the introduction of ENSO to the experimental data resulted in a more accurate prediction of the time of the occurrence of flood peaks and flood flows. Furthermore, the best results were obtained on the convolutional long short-term memory + encoder–decoder gate recurrent unit model.

The quality of magnetic resonance images may influence the diagnosis and subsequent treatment. Therefore, in this paper, a novel no-reference (NR) magnetic resonance image quality assessment (MRIQA) method is proposed. In the approach, deep convolutional neural network architectures are fused and jointly trained to better capture the characteristics of MR images. Then, to improve the quality prediction performance, the support vector machine regression (SVR) technique is employed on the features generated by fused networks. In the paper, several promising network architectures are introduced, investigated, and experimentally compared with state-of-the-art NR-IQA methods on two representative MRIQA benchmark datasets. One of the datasets is introduced in this work. As the experimental validation reveals, the proposed fusion of networks outperforms related approaches in terms of correlation with subjective opinions of a large number of experienced radiologists.