Maximum Average Accuracy Research Articles

Synthetic Data generation using DCGAN for improved traffic sign recognition

Traffic sign detection and recognition perform a vital function in real-world driver guidance applications, including driver assistance systems. Research into vision-based traffic sign detection (TSD) and traffic sign recognition (TSR) has gained considerable attention in the scientific community, led mainly by three variables: identification, monitoring, and classification. In addition, TSR provides valuable details and alerts for smart cars including advanced driving assistance (ADAS) and cooperative intelligent transport systems (CITS). Our work will generate high-quality synthetic prohibitory sign images using deep convolutional generative adversarial networks (DCGAN). This paper analyzes and discusses CNN models incorporating different backbone architectures and feature extractors, focusing on Resnet 50 and Densenet for object detection. Assessment of the models provides important information, including mean average accuracy (mAP), workspace capacity, detection period, and the amount of billion floating-point operations (BFLOPS). The maximum average accuracy is 92% (Densenet DCGAN), led by 91% (Resnet 50 DCGAN), 88% (Densenet), and 63% (Resnet 50). We find when using the original image and a synthetic image, accuracy increases, while detection time falls. Our findings show that combining original images and synthetic images in the dataset for training can improve intersection over union (IoU) and traffic sign recognition performance.

Time domain analysis of electroencephalogram (EEG) signals for word level comprehension in deaf graduates with congenital and acquired hearing loss

Deafness can be classified on the basis of onset as congenital and acquired hearing loss. The brain is a sensitive part of our body, electrical pulses from the neurons interact with each other, generating brain signals. EEG signals are extensively used for clinical diagnosis for any brain anomalies, language comprehension and performance measurement studies. This study mainly focuses on analysing the word level comprehension in deaf adults in the age group (21 -25 years) using EEG signals. The raw EEG signals were pre-processed and the relevant time domain linear and nonlinear features were extracted and classified using machine learning algorithms. The approximate entropy feature was found to be best suited for finding the comprehension of both congenital and acquired deaf adults. This feature of ISL was observed to be achieving better classification rate with a maximum average accuracy of 96% in both congenital and acquired deaf adults using SVM classifier.

Energy efficient compressive sensing with predictive model for IoT based medical data transmission

Internet of Things (IoT) systems tends to produce massive and diverse kinds of data which needs to be processes and responds in a smallperiod. A most important challenge exists in IoT devices is the amount of energy utilization while transmitting data into cloud. This paper presents a new energy efficient compressive technique with predictive model for IoT based medical data collection and transmission. The proposed model make use of Sensor-Lempel Ziv Welch (SLZW) technique is utilized to perform compressive sensing earlier to data transmission followed by particle swarm optimization (PSO) based deep neural network (DNN) based prediction. The PSO algorithm is applied for optimizing the node count of hidden layer in DNN due to the issue that the classical DNN got trapped into local minima and the node count in hidden layer have to select manually. The performance of the presented SLZW-PSO-DNN algorithm has been validated and the results are investigated under distinct scenarios. The obtained experimental outcome indicated that the SLZW-PSO-DNN algorithm is found to be effective under several aspects over the existing method. The experimental results stated that the PSO-DNN model has resulted in a maximum predictive average accuracy of 98.5% and 98.4% under original and compressed data respectively.

Automatic identification and hardware implementation of a resource-constrained power model for embedded systems

In modern embedded systems, the use of hardware-level online power monitors is crucial to support the run-time power optimizations required to meet the ever increasing demand for energy efficiency. To be effective and to deal with the time-to-market pressure, the presence of such requirements must be considered even during the design of the power monitoring infrastructure. This paper presents a power model identification and implementation strategy with two main advantages over the state-of-the-art. First, our solution trades the accuracy of the power model with the amount of resources allocated to the power monitoring infrastructure. Second, the use of an automatic power model instrumentation strategy ensures a timely implementation of the power monitor regardless the complexity of the target computing platforms. Our methodology has been validated against 8 accelerators generated through a High-Level-Synthesis flow and by considering a more complex RISC-V embedded computing platform. Depending on the imposed user-defined constraints and with respect to the unconstrained power monitoring state-of-the-art solutions, our methodology shows a resource saving between 37.3% and 81% while the maximum average accuracy loss stays within 5%, i.e., using the aggressive 20us temporal resolution. However, by varying the temporal resolution closer to the value proposed in the state of the art, i.e., in the range of hundreds of microseconds, the average accuracy loss of our power monitors is lower than 1% with almost the same overheads. In addition, our solution demonstrated the possibility of delivering a resource constrained power monitor employing a 20us temporal resolution, i.e., far higher the one used by current state-of-the-art solutions.

Designing a medical rule model system by using rough–grey modeling

PurposeIn this study, a hybrid rough and grey set-based rule model is designed for diagnosis of one type of blood cancer called multiple myeloma (MM). The grey clustering method is used to combine the same condition attributes and to improve the validity of the final model.Design/methodology/approachSome tools of the rough set theory (RST) and grey incidence analysis (GIA) are used in this research to analyze the serum protein electrophoresis (SPE) test results. An RST-based rule model is extracted based on the laboratory SPE test results of patients. Also, one decision attribute and 15 condition attributes are used to extract the rules. About four rule models are constructed due to the different algorithms of data complement, discretization, reduction and rule generation. In the following phases, the condition attributes are clustered into seven clusters by using a grey clustering method, the value set of the decision attribute is decreased by using manual discretizing and the number of observations is increased in order to improve the accuracy of the model. Cross-validation is used for evaluation of the model results and finally, the best model is chosen with 5,216 rules and 98% accuracy.FindingsIn this paper, a new rule model with high accuracy is extracted based on the combination of the grey clustering method and RST modeling for diagnosis of the MM disease. Also, four primary rule models and four improved rule models have been extracted from different decision tables in order to define the result of SPE test of patients. The maximum average accuracy of improved models is equal to 95% and related to the gamma globulins percentage attribute/object-related reducts (GA/ORR) model.Research limitations/implicationsThe total number of observations for rule extraction is 115 and the results can be improved by further samples. To make the designed expert system handy in the laboratory, new computer software is under construction to import data automatically from the electrophoresis machine into the resultant rule model system.Originality/valueThe main originality of this paper is to use the RST and GST together to design and create a hybrid rule model to diagnose MM. Although many studies have been carried out on designing expert systems in medicine and cancer diagnosis, no studies have been found in designing systems to diagnose MM. On the other hand, using the grey clustering method for combining the condition attributes is a novel solution for improving the accuracy of the rule model.

In-Ear EEG Based Attention State Classification Using Echo State Network.

It is important to maintain attention when carrying out significant daily-life tasks that require high levels of safety and efficiency. Since degradation of attention can sometimes have dire consequences, various brain activity measurement devices such as electroencephalography (EEG) systems have been used to monitor attention states in individuals. However, conventional EEG instruments have limited utility in daily life because they are uncomfortable to wear. Thus, this study was designed to investigate the possibility of discriminating between the attentive and resting states using in-ear EEG signals for potential application via portable, convenient earphone-shaped EEG instruments. We recorded both on-scalp and in-ear EEG signals from 6 subjects in a state of attentiveness during the performance of a visual vigilance task. We have designed and developed in-ear EEG electrodes customized by modelling both the left and right ear canals of the subjects. We use an echo state network (ESN), a powerful type of machine learning algorithm, to discriminate attention states on the basis of in-ear EEGs. We have found that the maximum average accuracy of the ESN method in discriminating between attentive and resting states is approximately 81.16% with optimal network parameters. This study suggests that portable in-ear EEG devices and an ESN can be used to monitor attention states during significant tasks to enhance safety and efficiency.

An efficient Bayesian network for differential diagnosis using experts' knowledge

PurposeThis study aims to differential diagnosis of some diseases using classification methods to support effective medical treatment. For this purpose, different classification methods based on data, experts’ knowledge and both are considered in some cases. Besides, feature reduction and some clustering methods are used to improve their performance.Design/methodology/approachFirst, the performances of classification methods are evaluated for differential diagnosis of different diseases. Then, experts' knowledge is utilized to modify the Bayesian networks' structures. Analyses of the results show that using experts' knowledge is more effective than other algorithms for increasing the accuracy of Bayesian network classification. A total of ten different diseases are used for testing, taken from the Machine Learning Repository datasets of the University of California at Irvine (UCI).FindingsThe proposed method improves both the computation time and accuracy of the classification methods used in this paper. Bayesian networks based on experts' knowledge achieve a maximum average accuracy of 87 percent, with a minimum standard deviation average of 0.04 over the sample datasets among all classification methods.Practical implicationsThe proposed methodology can be applied to perform disease differential diagnosis analysis.Originality/valueThis study presents the usefulness of experts' knowledge in the diagnosis while proposing an adopted improvement method for classifications. Besides, the Bayesian network based on experts' knowledge is useful for different diseases neglected by previous papers.

Computer-aided diagnosis for burnt skin images using deep convolutional neural network

Numerous patients died every year due to the leading causes of deaths all over the world and burn injuries are one of them. Burn injury cases are most viewed in low and middle-income countries (LMIC). Researchers show great interest to classify the burn into different depths through digital means. In Pakistan, at provisional level, it’s really a significant issue to categorize the burn and its depths due to the non-availability of expert doctors and surgeons; hence the decision for the correct first treatment can't be made, so this may cause a serious issue later on. The main objectives of this research work are to segment the burn wounds and classification of burn depths into 1st, 2nd and 3rd degrees respectively. A real-time dataset of burnt patients has been collected from the burn unit of Allied Hospital Faisalabad, Pakistan. The dataset used for this research task contains 450 images of all the three levels of burn depths. Segmentation of the burnt area was done by the use of Otsu's method of thresholding and feature vector was obtained through the use of statistical methods. We have used the Deep Convolutional Neural Network (DCNN) to estimate the burn depths. The network was trained by 65 percent of the images and the remaining 35 percent images were used for testing the accuracy of the classifier. The maximum average accuracy obtained by using the Deep Convolutional Neural Network (DCNN) classifier is reported round about 79.4% and these results are the best if we compare them with previous results. From the obtained results of this research work, non-expert doctors will be able to apply the correct first treatment for the quality evaluation of burn depths.

A Study of Multispectral Technology and Two-Dimension Autoencoder for Coal and Gangue Recognition

Coal is one of the main sources of human energy. In the process of coal mining, separating gangue from coal has great significance for environmental protection and energy conservation. The core problem in the separation of gangue is the recognition of coal and gangue. In this study, multispectral technology was used to identify gangue. First, set up a data acquisition system in the laboratory, and then collect spectral data. Coal and gangue spectral data were collected in 202 and 201 groups, respectively. Secondly, design a spectral data dimension reduction model called two-dimension autoencoder(2D-AE). Finally, Random Forest was used to recognize coal and gangue. Meanwhile, CART, KNN, SVM, and AdaBoost were also used for gangue identification. The experimental results show that the maximum average accuracy by 2D-AE combined with RF was the largest, which was 98.89%. Also, the accuracy of gangue recognition is different for spectral images of different wavelengths. This paper mainly studies the recognition of coal and gangue based on multispectral technology, which is of great significance for the next step of detecting gangue based on the technology.

Executive Function Evaluation System Based on Wiping Desk Behavior in Virtual Space

In this study, we report the development of a virtual reality (VR) system that evaluates executive function based on cleaning behaviors in real-time. It is known that patients with acquired brain injuries present adverse symptoms such as attention, memory, and functional disorders, as well as aphasia. Current methods to evaluate acquired brain injuries include tests such as the behavioral assessment of the dysexecutive syndrome (BADS) and digital clinical assessment for attention (D-CAT) tests; however, these tests require special/specific toolkits, and the methods are rigorous when applied for real-time dynamic evaluation. Moreover, patients are often burdened by the need to undergo paper tests as required by the evaluation methods. In this context, we propose and verify the efficacy of a method that offers real-time dynamic evaluation of acquired brain injury based on daily-living activities such as cooking, cleaning, and shopping. In this study, we focus on executive function relating to acquired brain injury, and we propose a real-time dynamic evaluation method for executive functioning using VR, which enables the automatic evaluation of subjects’ table-cleaning behavior. Our results indicate that our system can automatically assess subjects’ table cleaning behavior based on BADS test, while affording a cleaning behavior maximum average accuracy of 75.5%.

Improving the Efficiency of Dental Implantation Process Using Guided Local Search Models and Continuous Time Neural Networks With Robotic Assistance

Nowadays, robotics plays a vital role in medical applications, especially in dentistry, where robots can track oral hygiene and perform dental surgeries. Dental implant replacement is one of the most challenging issues in dental surgery; quality procedures and safety measures need to be considered during this process. Manual dental implant is usually incapable to reach the satisfactory levels of accuracy and safety. In addition, it requires well-trained dentists and consumes a long time. Therefore, robot-assisted surgery systems are of utmost importance for dental implant placement as they can maintain higher level of dental examination precision and safety. More specifically, robotic arms can be manufactured with intelligent models for drilling identified locations in teeth. These intelligent robots have a high degree of autonomy, can automatically adjust during intraoperative procedures, and can execute dental surgical tasks directly on patients without any apparent control by a surgeon. In this article, we propose a novel approach to develop a robot-assisted intelligent system that improves the efficiency of dental implant process based on Guided Local Search with Continuous Time Neural Network (GLCTNN). Firstly, dental facts are collected from PubMed articles and Maryland school children datasets. Secondly, using the collected facts, an intelligent robot-assisted model based on GLCTNN is developed. The second step comprises data preprocessing to remove unsolicited details, extracting useful features from the clean data, and utilizing the extracted features to train the GLCTNN model. The proposed system recognizes the implantation location with high accuracy and maximizes implantation rate. The efficiency of the system is evaluated using experimental analysis at lab scale. The proposed GLCTNN-based approach ensures maximum average accuracy (99.5%) and minimum average deviation error (0.323) compared to W-J48, Naïve Bayes (NB), Support Vector Machine (SVM), K-Nearest Neighboring (KNN), Nearest Neighbors with Structural Risk Minimization (NNSRM) and Generalized Regression Neural Network (GRNN) approaches.

A new 360° rotating type stimuli for improved SSVEP based brain computer interface

The authors investigated for first time clockwise and counter clockwise 360° rotating stimuli at two different speeds of rotation/phase for brain computer interface (BCI). The 360° rotating stimuli of frequencies 3 Hz, 7.5 Hz and 10 Hz are considered and compared to the performance with non-rotating flickering stimuli using signal-to-noise ratio at four channels Oz, Pz, O1 and O2. In addition, the authors have also attempted to study the effect of stimuli in varying combinations of four channels in discriminating stimuli frequencies. This paper investigated the discrimination of stimuli frequencies using power spectral density analysis (PSDA) and canonical correlation analysis (CCA) with information transfer rate (ITR). The results revealed the maximum average classification accuracy of 87.6 % is achieved from the single channel Oz, for clockwise rotating slow (CS) stimuli using CCA. The maximum average accuracy of 90.6 % is achieved from the channel combinations Oz, Pz, and O1 for CS rotating stimulus from CCA. It is noticed that, counter clockwise fast rotating stimulus is poor performing in single channel performance analysis. However, counter clockwise fast rotating stimulus has shown maximum average accuracy of 90 % considering all four channels with CCA. The maximum average ITR of 58.5 bit/min is achieved using signals of CS stimuli from the Oz channel and 65.1 bit/min is achieved with multichannel combination of Oz, Pz, O1 for clockwise slow rotating stimuli. The results indicate an enhancement in performance of SSVEP based BCI with clockwise rotating stimuli and may be considered for communication and or control applications.

FORECASTING OIL PALM PRODUCTION BASED ON A NONLINEAR AUTOREGRESSIVE EXOGENOUS (NARX) NEURAL NETWORK MODEL

time series data analysis and prediction tool that is able to predict the yield of oil palm is needed to ensure an acceptable forecasting quality. An attempt was made in this study to develop a Nonlinear Autoregressive Exogenous (NARX) neural network model of oil palm production using MATLAB. This NARX model was used to predict the yield of oil palm in the states of Kelantan, Johor, Sabah and Sarawak in Malaysia. The performance of the NARX model was tested and validated using the Levenberg-Marquardt (LM) training algorithm and was compared with the Autoregressive Integrated Moving Average (ARIMA) model. The best performance of the NARX model was achieved at 70 per cent: 15 per cent: 15 per cent, with 10 neurons in the hidden layers and a delay value of four for Sarawak. For Kelantan and Johor, the NARX model produced the best result using the parameters of 70 per cent:10 per cent:20 per cent, with 13 neurons in the hidden layers and a delay value of four. The best result for Sabah was achieved using the parameters of 70 per cent: 15 per cent: 15 per cent, with 13 neurons in the hidden layers and a delay value of four. The results demonstrated that the proposed NARX model was more effective in modeling and forecasting time series data than the ARIMA model. The NARX model registered a minimum mean square error and mean absolute percentage error with a maximum average accuracy percentage and correlation coefficient. Keywords: Oil palm cultivation, yield predictions, nonlinear autoregressive exogenous neural network, autoregressive integrated moving average.

Attempted Arm and Hand Movements can be Decoded from Low-Frequency EEG from Persons with Spinal Cord Injury

We show that persons with spinal cord injury (SCI) retain decodable neural correlates of attempted arm and hand movements. We investigated hand open, palmar grasp, lateral grasp, pronation, and supination in 10 persons with cervical SCI. Discriminative movement information was provided by the time-domain of low-frequency electroencephalography (EEG) signals. Based on these signals, we obtained a maximum average classification accuracy of 45% (chance level was 20%) with respect to the five investigated classes. Pattern analysis indicates central motor areas as the origin of the discriminative signals. Furthermore, we introduce a proof-of-concept to classify movement attempts online in a closed loop, and tested it on a person with cervical SCI. We achieved here a modest classification performance of 68.4% with respect to palmar grasp vs hand open (chance level 50%).

Automated recognition and classification of adulteration levels from bulk paddy grain samples

Fraudulent labeling and adulteration are the major concerns in the global rice industry. Almost all the paddy varieties being sold in the market are prone to adulteration. It is very difficult to differentiate paddy grains of various varieties in the mixed bulk sample based on visual observation. Currently, there is no sophisticated appearance-based commercial scale technology to reliably detect and quantify adulteration in bulk paddy grain samples. The paper presents a cost-effective image processing technique for the recognition of adulteration and classification of adulteration levels (%) from the images of adulterated bulk paddy samples using state-of-the-art color and texture features. In this work, seven adulterated bulk paddy samples are considered and each of the samples is prepared by mixing a premium paddy variety with the identical looking and commercially inferior paddy variety at five different adulteration levels (weight ratios) of 10%, 15%, 20%, 25% and 30%. The study compares the performances of three different classification models, namely, Multilayer Back Propagation Neural Network (BPNN), Support Vector Machine (SVM) and k-Nearest Neighbor (k-NN). The Principal Component Analysis (PCA) and Sequential Forward Floating Selection (SFFS) methods have been employed separately for the automatic selection of optimal feature subsets from the combined color and texture features. The maximum average adulteration level classification accuracy of 93.31% is obtained using the BPNN classification model trained with PCA-based reduced features. The proposed technique can be used as an economic, rapid, non-destructive and quantitative technique for testing adulteration, authenticity, and quality of bulk paddy grain samples.

Adaptive Data Structure Based Oversampling Algorithm for Ordinal Classification

<p>The main objective of this research is to improve the predictive accuracy of classification in ordinal multiclass imbalanced scenario. The methodology attempts to uplift the classifier performance through synthesizing sophisticated objects of immature classes. A novel Adaptive Data Structure based oversampling algorithm is proposed to create synthetic objects and Extreme Learning Machine for Ordinal Regression (ELMOP) classifier is adopted to validate our work. The proposed method generating new objects by analyzing the characteristics and intricacy of immature class objects. On the whole, the data set is divided into training and test data. Training data set is updated with new synthetic objects. The experimental analysis is performed on testing data set to check the efficiency of the proposed methodology by comparing it with the existing work. The performance evaluation is conducted in terms of the parameters called Mean Absolute Error, Maximum Mean Absolute Error, Geometric Mean, Kappa and Average Accuracy. The measures prove that the proposed methodology can produce authentic synthetic objects than the existing techniques. The Proposed technique can synthesize the new effective objects through evaluating the structure of immature class. It boosts the global precision and class wise precision especially preserves rank order of the classes.</p>

RETRACTED ARTICLE: Noise-estimation-based anisotropic diffusion approach for retinal blood vessel segmentation

Recently, numerous research works in retinal-structure analysis have been performed to analyze retinal images for diagnosing and preventing ocular diseases such as diabetic retinopathy, which is the first most common causes of vision loss in the world. In this paper, an algorithm for vessel detection in fundus images is employed. First, a denoising process using the noise-estimation-based anisotropic diffusion technique is applied to restore connected vessel lines in a retinal image and eliminate noisy lines. Next, a multi-scale line-tracking algorithm is implemented to detect all the blood vessels having similar dimensions at a selected scale. An openly available dataset, called “the STARE Project’s dataset,” has been firstly utilized to evaluate the accuracy of the proposed method. Accordingly, our experimental results, performed on the STARE dataset, depict a maximum average accuracy of around 93.88%. Then, an experimental evaluation on another dataset, named DRIVE database, demonstrates a satisfactory performance of the proposed technique, where the maximum average accuracy rate of 93.89% is achieved.

RETRACTED ARTICLE: Automated screening of congestive heart failure using variational mode decomposition and texture features extracted from ultrasound images

Heart is an important and hardest working muscular organ of the human body. Inability of the heart to restore normal perfusion to the entire body refers to cardiac failure, which then with symptoms results in manifestation of congestive heart failure (CHF). Impairment in systolic function associated with chronic dilation of left ventricle is referred as dilated cardiomyopathy (DCM). The clinical examination, surface electrocardiogram (ECG), chest X-ray, blood markers and echocardiography play major role in the diagnosis of CHF. Though the ECG manifests chamber enlargement changes, it does not possess sensitive marker for the diagnosis of DCM, whereas echocardiographic assessment can effectively reveal the presence of asymptomatic DCM. This work proposes an automated screening method for classifying normal and CHF echocardiographic images affected due to DCM using variational mode decomposition technique. The texture features are extracted from variational mode decomposed image. These features are selected using particle swarm optimization and classified using support vector machine classifier with different kernel functions. We have validated our experiment using 300 four-chamber echocardiography images (150: normal, 150: CHF) obtained from 50 normal and 50 CHF patients. Our proposed approach yielded maximum average accuracy, sensitivity and specificity of 99.33%, 98.66% and 100%, respectively, using ten features. Thus, the developed diagnosis system can effectively detect CHF in its early stage using ultrasound images and aid the clinicians in their diagnosis.

Phase Segmentation Methods for an Automatic Surgical Workflow Analysis.

In this paper, we present robust methods for automatically segmenting phases in a specified surgical workflow by using latent Dirichlet allocation (LDA) and hidden Markov model (HMM) approaches. More specifically, our goal is to output an appropriate phase label for each given time point of a surgical workflow in an operating room. The fundamental idea behind our work lies in constructing an HMM based on observed values obtained via an LDA topic model covering optical flow motion features of general working contexts, including medical staff, equipment, and materials. We have an awareness of such working contexts by using multiple synchronized cameras to capture the surgical workflow. Further, we validate the robustness of our methods by conducting experiments involving up to 12 phases of surgical workflows with the average length of each surgical workflow being 12.8 minutes. The maximum average accuracy achieved after applying leave-one-out cross-validation was 84.4%, which we found to be a very promising result.

DIFFERENT DUTY CYCLE RATIO AND BRIGHTNESS OF VISUAL STIMULI CHANGE TO STEADY STATE VISUAL EVOKED POTENTIAL RESPONSE

Stimuli types are very crucial for the performance of electroencephalogram (EEG) based brain computer interface (BCI) systems. This study aims to investigate methods for obtaining higher information transfer rate (ITR) through duty cycle and brightness variation of visual stimuli which have high frequency for steady state visual evoked potential-based BCI. Although previous studies were concentrated on either duty cycle or brightness of stimuli separately, our study focused on the change of duty cycle ratio and brightness of stimuli at the same time. Duty cycle values of 40%, 50%, and 60% were used. During the experiment, 16 flickering stimuli were used on liquid crystal display. Participants gazed to the flicker which had frequency of 15 Hz. Canonical correlation analyses (CCA) was used for channel selection and frequency detection. According to the CCA, the maximum average accuracy of the experiment was 92.54% when the frequency of flicker was in beta band and its duty cycle was 40% with a brightness tuning wave. Under the same conditions stated above, average ITR was improved 16.1% according to the most commonly used flicker model which is square wave and has 50% duty cycle.