Volume 26 (2021): Issue 2 (December 2021)

MEMS gyroscopes are widely used as an alternative to the more expensive industrial IMUs. The instability of the lower cost MEMS gyroscopes creates a large demand for calibration algorithms. This paper provides an overview of existing calibration methods and describes the various types of errors found in gyroscope data. The proposed calibration method for gyroscope constants provides higher accuracy than datasheet constants. Furthermore, we show that using a different constant for each direction provides even higher accuracy.

Rail track breakages represent broken structures consisting of rail track on the railroad. The traditional methods for detecting this problem have proven unproductive. The safe operation of rail transportation needs to be frequently monitored because of the level of trust people have in it and to ensure adequate maintenance strategy and protection of human lives and properties. This paper presents an automatic deep learning method using an improved fully Convolutional Neural Network (FCN) model based on U-Net architecture to detect and segment cracks on rail track images. An approach to evaluating the extent of damage on rail tracks is also proposed to aid efficient rail track maintenance. The model performance is evaluated using precision, recall, F1-Score, and Mean Intersection over Union (MIoU). The results obtained from the extensive analysis show U-Net capability to extract meaningful features for accurate crack detection and segmentation.

Day-to-day working activities have been heavily altered by COVID-19 pandemic, forcing a transition from traditional on-site work to on-line telework across the whole world. It has become much harder to efficiently organise, guide and evaluate employee’s work. There are different factors that can influence “work from home” quality, and many of these affect such work negatively. A set of relevant methods and tools should be developed which could improve this situation. The goal of the study is to summarise related background of this problem and to propose an approach to overcoming this problem. To achieve the goal, design engineer’s work is evaluated in an appropriate environment (e.g., AutoCAD, etc.) using automated analysis and visualization of IS auditing data.

In recent years, various domains have been influenced by the rapid growth of machine learning. Autonomous driving is an area that has tremendously developed in parallel with the advancement of machine learning. In autonomous vehicles, various machine learning components are used such as traffic lights recognition, traffic sign recognition, limiting speed and pathfinding. For most of these components, computer vision technologies with deep learning such as object detection, semantic segmentation and image classification are used. However, these machine learning models are vulnerable to targeted tensor perturbations called adversarial attacks, which limit the performance of the applications. Therefore, implementing defense models against adversarial attacks has become an increasingly critical research area. The paper aims at summarising the latest adversarial attacks and defense models introduced in the field of autonomous driving with machine learning technologies up until mid-2021.

Insufficient user involvement, lack of user feedback, incomplete and changing user requirements are some of the critical reasons for the difficulty of IS usage, which could potentially reduce the number of customers. Under the previous authors’ research, the method for analysing the behaviour of IT system users was developed, which was intended to improve the usability of the system and thus could increase the efficiency of business processes. The developed method is based on the use of graph searching algo rithms, Markov chains and Machine Learning approach. This paper focuses on detailing of method output data in the context of definition of their importance based on expert evaluation and demonstration of visual presentation of different UX analysis situations. The paper briefly reminds the essence of the method, including both the input and output data sets, and, with the help of experts, evaluates the expected result in the context of their importance in UX analysis. It also introduces visualization prototype developed to obtain the output data, which allows verifying the input/output data transformation possibilities and expected data acquisition potential.

Age estimation from brain MRI has proved to be considerably helpful in early diagnosis of diseases such as Alzheimer’s and Parkinson’s. In this study, curriculum learning effect on age estimation models was measured using a brain MRI dataset consisting of normal and anomaly data. Three different strategies were selected and compared using 3D Convolutional Neural Networks as the Deep Learning architecture. The strategies were as follows: (1) model training performed only on normal data, (2) model training performed on the entire dataset, (3) model training performed on normal data first and then further training on the entire dataset as per curriculum learning. The results showed that curriculum learning improved results by 20 % compared to traditional training strategies. These results suggested that in age estimation tasks datasets consisting of anomaly data could also be utilized to improve performance.

The academic performance of students is essential for academic progression at all levels of education. However, the availability of several cognitive and non-cognitive factors that influence students’ academic performance makes it challenging for academic authorities to use conventional analytical tools to extract hidden knowledge in educational data. Therefore, Educational Data Mining (EDM) requires computational techniques to simplify planning and determining students who might be at risk of failing or dropping from school due to academic performance, thus helping resolve student retention. The paper studies several cognitive and non-cognitive factors such as academic, demographic, social and behavioural and their effect on student academic performance using machine learning algorithms. Heterogenous lazy and eager machine learning classifiers, including Decision Tree (DT), K-Nearest-Neighbour (KNN), Artificial Neural Network (ANN), Logistic Regression (LR), Random Forest (RF), AdaBoost and Support Vector Machine (SVM) were adopted and training was performed based on k-fold (k = 10) and leave-one-out cross-validation. We evaluated their predictive performance using well-known evaluation metrics like Area under Curve (AUC), F-1 score, Precision, Accuracy, Kappa, Matthew’s correlation coefficient (MCC) and Recall. The study outcome shows that Student Absence Days (SAD) are the most significant predictor of students’ academic performance. In terms of prediction accuracy and AUC, the RF (Acc = 0.771, AUC = 0.903), LR (Acc = 0.779, AUC = 0.90) and ANN (Acc = 0.760, AUC = 0.895) outperformed all other algorithms (KNN (Acc = 0.638, AUC = 0.826), SVM (Acc = 0.727, AUC = 0.80), DT (Acc = 0.733, AUC = 0.876) and AdaBoost (Acc = 0.748, AUC = 0.808)), making them more suitable for predicting students’ academic performance.

Nowadays, natural language processing (NLP) is increasingly relaying on pre-trained word embeddings for use in various tasks. However, there is little research devoted to Latvian – a language that is much more morphologically complex than English. In this study, several experiments were carried out in three NLP tasks on four different methods of creating word embeddings: word2vec, fastText, Structured Skip-Gram and ngram2vec. The obtained results can serve as a baseline for future research on the Latvian language in NLP. The main conclusions are the following: First, in the part-of-speech task, using a training corpus 46 times smaller than in a previous study, the accuracy was 91.4 % (versus 98.3 % in the previous study). Second, fastText demonstrated the overall best effectiveness. Third, the best results for all methods were observed for embeddings with a dimension size of 200. Finally, word lemmatization generally did not improve results.

The classic Internet of Things-Cloud Computing model faces challenges like high response latency, high bandwidth consumption, and high storage requirement with increasing velocity and volume of generated data. Fog computing offers better services to end users by bringing processing, storage, and networking closer to them. Recently, there has been significant research addressing architectural and algorithmic aspects of fog computing. In the existing literature, a systematic study of architectural designs is widely conducted for various applications. Algorithms are seldom examined. Algorithms play a crucial role in fog computing. This survey aims to performing a comparative study of existing algorithms. The study also presents a systematic classification of the current fog computing algorithms and highlights the key challenges and research issues associated with them.

Deep neural networks are a tool for acquiring an approximation of the robot mathematical model without available information about its parameters. This paper compares the LSTM, stacked LSTM and phased LSTM architectures for time series forecasting. In this paper, motion sensor data from mobile robot driving episodes are used as the experimental data. From the experiment, the models show better results for short-term prediction, where the LSTM stacked model slightly outperforms the other two models. Finally, the predicted and actual trajectories of the robot are compared.

The study proposes the text tone detection system based on sentiment dictionaries and fuzzy rules. Computer analysis of texts from different sources has been performed in emotional categories: anger, anticipation, disgust, fear, joy, sadness, surprise and trust. A synonym dictionary has been used to expand the vocabulary. To increase the accuracy and validity of sentiment analysis, the authors of the study have used coefficients that take into account different emotional loads of words of various parts of speech and the action of intensifying or softening adverbs. A quantitative value of the text tone has been obtained as a result of an aggregation of normalized data on all emotional categories by the fuzzy inference methods. It has been found that emotional words have a greater impact on the text tone value in the case of analysis of short messages. The proposed approach makes it possible to contribute to all emotional categories in the final text evaluation.

There are several biometric methods for identification. These are generally classified under two main groups as physiological and behavioural biometric methods. Recently, methods using behavioural biometric features have gained popularity. Identification made using gait pattern is also one of these methods. The present study proposes a machine learning based system performing identification in real time via gait features using a Kinect device. The data set is composed of 23 individuals’ skeleton model data obtained by the authors. From these data, 147 handcrafted features have been extracted. Deep Neural Network (DNN), Random Forest (RF), Gradient Boosting (GB), XG-Boost (XGB) and K-Nearest Neighbour (KNN) classifiers have been trained with these features. Furthermore, the output of these five machine learning models has been combined with a voting approach. The highest classification has been obtained with 97.5 % accuracy via a voting approach. The classification accuracies of the RF, DNN, XGB, GB and KNN classifiers are 95 %, 87.5 %, 85 %, 80 % and 65 %, respectively. The classification accuracy obtained via a voting approach is higher than in the previous studies. The developed system successfully performs real-time identification.

In this paper, we describe our team’s data processing practice for an event-based camera dataset. In addition to the event-based camera data, the Agri-EBV dataset contains data from LIDAR, RGB, depth cameras, temperature, moisture, and atmospheric pressure sensors. We describe data transfer from a platform, automatic and manual validation of data quality, conversions to multiple formats, and structuring of the final data. Accurate time offset estimation between sensors achieved in the dataset uses IMU data generated by purposeful movements of the sensor platform. Therefore, we also outline partitioning of the data and time alignment calculation during post-processing.

Generally, the process of plagiarism detection can be divided into two main stages: source retrieval and text alignment. The paper evaluates and compares effectiveness of five fingerprint selection algorithms used during the source retrieval stage: Every p-th, 0 mod p, Winnowing, Frequency-biased Winnowing (FBW) and Modified FBW (MFBW). The algorithms are evaluated on a dataset containing plagiarism cases in Bachelor and Master Theses written in English in the field of computer science. The best performance is reached by 0 mod p, Winnowing and MFBW. For these algorithms, reduction of fingerprint size from 100 % to about 20 % kept the effectiveness at approximately the same level. Moreover, MFBW sends overall fewer document pairs to the text alignment stage, thus also reducing the computational cost of the process. The software developed for this study is freely available at the author’s website http://www.cs.rtu.lv/jekabsons/.

It is important to know how much the lungs are affected in the course of the disease in patients with COVID-19. Detecting infected tissues on CT lung images not only helps diagnose the disease but also helps measure the severity of the disease. In this paper, using the hybrid artificial intelligence-based segmentation method, which we call TA-Segnet, it has been revealed how the region with COVID-19 affects the lung on 2D CT images. A hybrid convolutional neural network-based segmentation method (TA-Segnet) has been developed for this process. We use “COVID-19 CT Lung and Infection Segmentation Dataset” and “COVID-19 CT Segmentation Dataset” to evaluate TA-SegNET. At first, the tissues with COVID-19 on each lung image are determined, then the measurements obtained are evaluated according to the parameters of Accuracy, Dice, Jaccard, Mean Square Error, Mutual Information and Cross-correlation. Accuracy, Dice, Jaccard, Mean Square Error, Mutual Information and Cross-correlation values for data set-1 are 98.63 %, 0.95, 0.919, 0.139, 0.51, and 0.904, respectively. For data set-2, these parameters are 98.57 %, 0.958, 0.992, 0.0088, 0.565 and 0.8995, respectively. Second, the ratio of COVID-19 regions relative to the lung region on CT images is determined. This ratio is compared with the values in the original data set. The results obtained show that such an artificial intelligence-based method during the pandemic period will help prioritize and automate the diagnosis of COVID-19 patients.