Manually-labeled Data Research Articles

Analysis of the Financial Information Contained in the Texts of Current Reports: A Deep Learning Approach

An important role in the fundamental analysis is played by the acquisition and analysis of various types of information about the company. Text documents are an increasingly important source of this information. Their accurate and quick analysis is an increasingly important challenge for financial analysts. Research in the area of financial text analysis is based on sentiment analysis. The deep neural networks and the stocks’ cumulative abnormal return are used in this article to analyze the sentiment of financial texts. The proposed approach, unlike those used so far, does not require manual labeling of data or the creation of dictionaries and is free from the subjective assessment of the researcher. Taking into account the broad context of words and their meaning in financial texts, it also eliminates the problem of ambiguity of words in various contexts. The sentiment of financial texts presented in this paper is directly related to the market reaction to the information contained in these texts. For texts belonging to one of the two classes (positive or negative) with the highest probability, the deep learning model gives predictions with a precision of 62% for the positive class and 55% for the negative class. The event study results show that the sentiment calculated under the proposed method can be successfully used to determine the probable direction of the market reaction to the information contained in current reports with a 1 percent significance level. The results can be used in market efficiency research, investment strategy development or support of investment analysts using fundamental analysis.

The State of Mind of Health Care Professionals in Light of the COVID-19 Pandemic: Text Analysis Study of Twitter Discourses.

BackgroundThe COVID-19 pandemic has affected populations worldwide, with extreme health, economic, social, and political implications. Health care professionals (HCPs) are at the core of pandemic response and are among the most crucial factors in maintaining coping capacities. Yet, they are also vulnerable to mental health effects caused by managing a long-lasting emergency with a lack of resources and under complicated personal concerns. However, there are a lack of longitudinal studies that investigate the HCP population.ObjectiveThe aim of this study was to analyze the state of mind of HCPs as expressed in online discussions published on Twitter in light of the COVID-19 pandemic, from the onset of the pandemic until the end of 2020.MethodsThe population for this study was selected from followers of a few hundred Twitter accounts of health care organizations and common HCP points of interest. We used active learning, a process that iteratively uses machine learning and manual data labeling, to select the large-scale population of Twitter accounts maintained by English-speaking HCPs, focusing on individuals rather than official organizations. We analyzed the topics and emotions in their discourses during 2020. The topic distributions were obtained using the latent Dirichlet allocation algorithm. We defined a measure of topic cohesion and described the most cohesive topics. The emotions expressed in tweets during 2020 were compared to those in 2019. Finally, the emotion intensities were cross-correlated with the pandemic waves to explore possible associations between the pandemic development and emotional response.ResultsWe analyzed the timelines of 53,063 Twitter profiles, 90% of which were maintained by individual HCPs. Professional topics accounted for 44.5% of tweets by HCPs from January 1, 2019, to December 6, 2020. Events such as the pandemic waves, US elections, or the George Floyd case affected the HCPs’ discourse. The levels of joy and sadness exceeded their minimal and maximal values from 2019, respectively, 80% of the time (P=.001). Most interestingly, fear preceded the pandemic waves, in terms of the differences in confirmed cases, by 2 weeks with a Spearman correlation coefficient of ρ(47 pairs)=0.340 (P=.03).ConclusionsAnalyses of longitudinal data over the year 2020 revealed that a large fraction of HCP discourse is directly related to professional content, including the increase in the volume of discussions following the pandemic waves. The changes in emotional patterns (ie, decrease in joy and increase in sadness, fear, and disgust) during the year 2020 may indicate the utmost importance in providing emotional support for HCPs to prevent fatigue, burnout, and mental health disorders during the postpandemic period. The increase in fear 2 weeks in advance of pandemic waves indicates that HCPs are in a position, and with adequate qualifications, to anticipate pandemic development, and could serve as a bottom-up pathway for expressing morbidity and clinical situations to health agencies.

Toward Robotic Weed Control: Detection of Nutsedge Weed in Bermudagrass Turf Using Inaccurate and Insufficient Training Data

To enable robotic weed control, we develop algorithms to detect nutsedge weed from bermudagrass turf. Due to the similarity between the weed and the background turf, manual data labeling is expensive and error-prone. Consequently, directly applying deep learning methods for object detection cannot generate satisfactory results. Building on an instance detection approach (i.e. Mask R-CNN), we combine synthetic data with raw data to train the network. We propose an algorithm to generate high fidelity synthetic data, adopting different levels of annotations to reduce labeling cost. Moreover, we construct a nutsedge skeleton-based probabilistic map (NSPM) as the neural network input to reduce the reliance on pixel-wise precise labeling. We also modify loss function from cross entropy to Kullback-Leibler divergence which accommodates uncertainty in the labeling process. We implement the proposed algorithm and compare it with both Faster R-CNN and Mask R-CNN. The results show that our design can effectively overcome the impact of imprecise and insufficient training sample issues and significantly outperform the Faster R-CNN counterpart with a false negative rate of only 0.4%. In particular, our approach also reduces labeling time by 95% while achieving better performance if comparing with the original Mask R-CNN approach.

Tracking social media during the COVID-19 pandemic: The case study of lockdown in New York State

Facing the COVID-19 pandemic, governments have implemented a wide range of policies to contain the spread of the virus. During the pandemic, large amounts of COVID-19-related tweets emerge every day. Real-time processing of daily tweets may offer insights for monitoring public opinion about intervention measures implemented. In this work, lockdown policy in New York State has been set as a target of public opinion research. This task includes two stages, stance detection and opinion monitoring. For the stance detection stage, we explored several combinations of different text representations and classification algorithms, finding that the combination of Long Short-Term Memory (LSTM) with Global Vectors for Word Representation (GloVe) outperforms others. Due to the shortage of labeled data, we adopted the data distillation method for the training data augmentation. The augmentation of the training data allows to improve the performance of the model with a very small amount of manually-labeled data. After applying the distillation method, the accuracy of the model has been significantly improved. Utilizing the enhanced model, automatically classified tweets are analyzed over time to monitor the public opinion. By exploring the tweets in New York from January 22nd until September 30th, 2020, we show the correlation of public opinion with COVID-19 cases and mortality data, and the effect of government responses on the opinion shift. These results demonstrate the capability of the presented method to effectively and efficiently monitor public opinion during a pandemic.

Automatic Detection of Impervious Surfaces from Remotely Sensed Data Using Deep Learning

The large scale quantification of impervious surfaces provides valuable information for urban planning and socioeconomic development. Remote sensing and GIS techniques provide spatial and temporal information of land surfaces and are widely used for modeling impervious surfaces. Traditionally, these surfaces are predicted by computing statistical indices derived from different bands available in remotely sensed data, such as the Landsat and Sentinel series. More recently, researchers have explored classification and regression techniques to model impervious surfaces. However, these modeling efforts are limited due to lack of labeled data for training and evaluation. This in turn requires significant effort for manual labeling of data and visual interpretation of results. In this paper, we train deep learning neural networks using TensorFlow to predict impervious surfaces from Landsat 8 images. We used OpenStreetMap (OSM), a crowd-sourced map of the world with manually interpreted impervious surfaces such as roads and buildings, to programmatically generate large amounts of training and evaluation data, thus overcoming the need for manual labeling. We conducted extensive experimentation to compare the performance of different deep learning neural network architectures, optimization methods, and the set of features used to train the networks. The four model configurations labeled U-Net_SGD_Bands, U-Net_Adam_Bands, U-Net_Adam_Bands+SI, and VGG-19_Adam_Bands+SI resulted in a root mean squared error (RMSE) of 0.1582, 0.1358, 0.1375, and 0.1582 and an accuracy of 90.87%, 92.28%, 92.46%, and 90.11%, respectively, on the test set. The U-Net_Adam_Bands+SI Model, similar to the others mentioned above, is a deep learning neural network that combines Landsat 8 bands with statistical indices. This model performs the best among all four on statistical accuracy and produces qualitatively sharper and brighter predictions of impervious surfaces as compared to the other models.

Naive Bayes for Thesis Labeling

The thesis preparation in the Department of Informatics Universitas Ahmad Dahlan is divided into two areas of interest, namely Intelligent Systems and Software and Data Engineering. Existing thesis title data is only used as an archive and has never been processed or classified to determine the trend of thesis topics based on student interest each year. The stages include data collection, the data is divided into two parts (training data and test data), manual labeling of training data, text preprocessing, and classification using Naive Bayes. The results show the trend of thesis title taking from 2013 to 2018 shows the thesis trend in the field of Intelligent Systems and Software. Accuracy testing uses Confusion Matrix and K-Fold Cross Validation with a k value is 10, has a value of 94.60%, precision of 97.30%, and a recall of 85.70%.

Countering the Cognitive, Linguistic, and Psychological Underpinnings Behind Susceptibility to Fake News: A Review of Current Literature With Special Focus on the Role of Age and Digital Literacy

Fake news poses one of the greatest threats to democracy, journalism, and freedom of expression. In recent cases, fake news’ designs are to create confusion and lower trust among the general public—as seen in the 2016 United States presidential campaign and the Brexit referendum. The spread of information without formal verification increased since the introduction of social media and online news channels. After the popularization of fake news, researchers have tried to evaluate and understand the effects of false information from multiple different perspectives. However, it is evident that to tackle the problem of fake news, interdisciplinary collaboration is needed. This article evaluates the main findings of recent literature from an integrated psychological, linguistic, cognitive, and societal perspective, with a particular focus on digital and age-related aspects of fake news. From a psychosociological standpoint, the article provides a synthesized profile of the fake news believer. This profile generally denotes overconfidence in one’s ability to assess falsehoods due to a human need for causal explanations. The fake news believer can be described as well-intentioned and critical, yet driven by a basis of distrust and false foundational knowledge. Within linguistics, manual analytical tools exist to understand the persuasive tactics in fake news. The article takes analytical techniques from both the humanities and the social sciences, such as transitivity analysis, Hugh Rank’s language persuasive framework, and others that can be used to analyze the language used in the news. However, in the age of big data perhaps only computational techniques can adequately address the issue at the root. While this proves successful, there are hurdles like the ambiguity of satire and sarcasm, manual labeling of data, and the supple nature of language. Reading comprehension differences between digital versus paper reading seem inconclusive. There are, however, notable behavioral and cognitive differences in reading behavior for the digital medium such as more scanning, less sustained attention, cognitive retreat, and shallower processing. Interestingly, when metacognitive strategies were probed by, for example, having participants independently allocate reading time, a difference in comprehension scores started to emerge. Researchers have also found accounts of differences due to medium preference; and on average older people seem to prefer paper reading. Cognitive retreat, shallow processing, and overconfidence associated with digital reading and the digital medium, in general, might make readers less likely to engage in the cognitive effort fake news detection requires. Considering that there are clear cognitive differences between older generations and younger generations (in terms of decreased processing speed, metacognition, and ability to multitask) differences in how these generations process fake news is plausible. Regrettably, most current research into psychological factors influencing susceptibility to fake news does not take into account age differences. Our meta-analysis showed that 74% of behavioral studies looking at fake news largely ignore age (N= 62), even though voter turnout was far higher among older generations for both the 2016 United States presidential election and the 2016 United Kingdom European Union membership referendum. Many provisional programs set up in the past few years aimed at training digital literacy, reading comprehension, and asking critical questions as virtual skills to detect fake news. These training programs are, however, mostly aimed at younger – digitally native – groups. As a result, these efforts might not be as efficacious as intended and could be improved upon significantly. This article argues that age must become a larger focus in fake news research and efforts in educating people against fake news must expand outside of the universities and isolated areas and include older generations.

Lifelong and Continual Learning Dialogue Systems: Learning during Conversation

Dialogue systems, also called chatbots, are now used in a wide range of applications. However, they still have some major weaknesses. One key weakness is that they are typically trained from manually-labeled data and/or written with handcrafted rules, and their knowledge bases (KBs) are also compiled by human experts. Due to the huge amount of manual effort involved, they are difficult to scale and also tend to produce many errors ought to their limited ability to understand natural language and the limited knowledge in their KBs. Thus, the level of user satisfactory is often low. In this paper, we propose to dramatically improve the situation by endowing the chatbots the ability to continually learn (1) new world knowledge, (2) new language expressions to ground them to actions, and (3) new conversational skills, during conversation by themselves so that as they chat more and more with users, they become more and more knowledgeable and are better and better able to understand diverse natural language expressions and to improve their conversational skills.

Learning Prediction Intervals for Model Performance

Understanding model performance on unlabeled data is a fundamental challenge of developing, deploying, and maintaining AI systems. Model performance is typically evaluated using test sets or periodic manual quality assessments, both of which require laborious manual data labeling. Automated performance prediction techniques aim to mitigate this burden, but potential inaccuracy and a lack of trust in their predictions has prevented their widespread adoption. We address this core problem of performance prediction uncertainty with a method to compute prediction intervals for model performance. Our methodology uses transfer learning to train an uncertainty model to estimate the uncertainty of model performance predictions. We evaluate our approach across a wide range of drift conditions and show substantial improvement over competitive baselines. We believe this result makes prediction intervals, and performance prediction in general, significantly more practical for real-world use.

Automated Labeling for Robotic Autonomous Navigation Through Multi-Sensory Semi-Supervised Learning on Big Data

Imitation learning holds the promise to address challenging robotic tasks such as autonomous navigation. It however requires a human supervisor to oversee the training process and send correct control commands to robots without feedback, which is always prone to error and expensive. To minimize human involvement and avoid manual labeling of data in the robotic autonomous navigation with imitation learning, this paper proposes a novel semi-supervised imitation learning solution based on a multi-sensory design. This solution includes a suboptimal sensor policy based on sensor fusion to automatically label states encountered by a robot to avoid human supervision during training. In addition, a recording policy is developed to throttle the adversarial affect of learning too much from the suboptimal sensor policy. As a result, this solution allows the robot to learn a navigation policy in a self-supervised manner without human intervention after the initial data collection. With extensive experiments in indoor environments, this solution can achieve near human performance in most of the tasks and even surpasses human performance in case of unexpected events such as hardware failures or human operation errors. To best of our knowledge, this is the first work that synthesizes sensor fusion and imitation learning to enable robotic autonomous navigation in the real world without human supervision.

ReSFlow: A Remote Sensing Imagery Data-Flow for Improved Model Generalization

As satellite imagery collections continue to grow at an astonishing rate, so is the demand for automated and scalable object detection and segmentation. Scaling computational activities demand models that generalize well across various challenges that can hamper progress, including diverse imaging and geographic conditions, sampling bias in training data, manual ground truth collection, tooling for model reuse and accountability assessment, and poor model training strategies. A great deal of progress has been made on these challenges. We contribute to the improvement through further development of ReSFlow, a workflow that breaks the problem of model generalization into a collection of specialized exploitations. ReSFlow partitions imagery collections into homogeneous buckets equipped with exploitation models trained to perform well under each bucket.s specific context. Essentially, ReSFlow aims for generalization through stratification. Therefore, within a bucket, exploitation is a homogeneous process that mitigates heterogeneity challenges, including the number of training data and data biases that can occur over varied conditions. Furthermore, custom model architectures and rich training strategies effective for within-bucket conditions can be developed. Meanwhile, across buckets, performance metrics support systematic views of the workflow leading to optimal data labeling allocations and indications that further specialization is warranted. Herein, we discuss the formation of models during the framework's “Offline Initialization” stage. Lastly, we exploit the inherent parallelism due to bucketing to introduce model reuse and demonstrate efficacy by reducing an 89-day manual data labeling cost to zero-days in a new area of interest.

Supervised and Semi-Supervised Learning for Failure Identification in Microwave Networks

Automated failure-cause identification in communication networks allows operators to reduce service unavailability. Once the most likely failure root-cause is identified, appropriate countermeasures can be effectively put in place (e.g., by choosing an in-field intervention vs. a remote equipment reconfiguration). In this article, we describe a successful application of Machine Learning (ML) for automatic failure identification in microwave networks based on the real-field data. On microwave links, different heterogeneous causes (e.g., adverse atmospheric conditions, or obstacles) lead to service unavailability and produce not easily-distinguishable degradation effects on the transmission parameters. Hence, failure identification is traditionally accomplished by domain experts via direct inspection of transmission-parameter logs. As a first contribution, we identify six categories of failure causes in microwave networks and show that supervised ML enables very accurate failure identification, hence significantly simplifying failure troubleshooting. Comparing various ML algorithms, we find that up to 93% classification accuracy is obtained using real-field labeled datasets with 2513 points. One main hindrance to the application of supervised learning is that, in real network deployments, limited amount of labeled data is available for training, as manual labeling is performed by domain experts based on their knowledge and experience. On the other hand, collecting unlabeled data is relatively simple as network management systems retrieve large amounts of unlabeled information automatically. As a second contribution, we investigate an automated labeling procedure, based on autoencoders-like Artificial Neural Networks, to combine the knowledge of the few manually-labeled data with large unlabeled data. Results show that our data augmentation based on autoencoders can slightly improve failure-cause identification only when Artificial Neural Networks or Support Vector Machines are used, while accuracy slightly decreases when adopting Random Forest.

Deep learning for scanning electron microscopy: Synthetic data for the nanoparticles detection

Deep learning algorithms are one of most rapid developing fields into the modern computation technologies. One of the bottlenecks into the implementation of such advaced algorithms is their requirement for a large amount of manually-labelled data for training. For the general-purpose tasks, such as general purpose image classification/detection the huge images datasets are already labelled and collected. For more subject specific tasks (such as electron microscopy images treatment), no labelled data available. Here I demonstrate that a deep learning network can be successfully trained for nanoparticles detection using semi-synthetic data. The real SEM images were used as a textures for rendered nanoparticles at the surface. Training of RetinaNet architecture using transfer learning can be helpful for the large-scale particle distribution analysis. Beyond such applications, the presented approach might be applicable to other tasks, such as image segmentation.

ANALISIS SENTIMEN GOJEK PADA MEDIA SOSIAL TWITTER DENGAN KLASIFIKASI SUPPORT VECTOR MACHINE (SVM

Appearance of PT Aplikasi Karya Anak Bangsa or as known as Gojek since 2015 give a convenience facility to people in Indonesia especially in daily activities. Sentiment analysis on Twitter social media can be the option to see how Gojek users respond to the services that have been provided. The response was classified into positive sentiment and negative sentiment using Support Vector Machine method with model evaluation 10-fold cross validation. The kernel used is the linear kernel and the RBF kernel. Data labeling can be done with manually and sentiment scoring. The test results showed that the RBF kernel gets overall accuracy and the highest kappa accuracy on manual data labeling and sentiment scoring. On manual data labeling, the overall accuracy is 79.19% and kappa accuracy is 16.52%. While the labeling of data with sentiment scoring obtained overall accuracy of 79.19% and kappa accuracy of 21%. The greater overall accuracy value and kappa accuracy obtained, the better performance of the classification model. Keywords: Gojek, Twitter, Support Vector Machine, overall accuracy, kappa accuracy

Significance of spectral cues in automatic speech segmentation for Indian language speech synthesizers

Building speech synthesis systems for Indian languages is challenging owing to the fact that digital resources for these languages are hardly available. Vocabulary independent speech synthesis requires that a given text is split at the level of the smallest sound unit, namely, phone. The waveforms or models of phones are concatenated to produce speech. The waveforms corresponding to that of the phones are obtained manual (listening and marking) when digital resources are scarce. But the manual labeling of speech data (also known as speech segmentation) can lead to inconsistencies as the duration of phones can be as short as 10ms.The most common approach to automatic segmentation of speech is to perform forced alignment using monophone hidden Markov models (HMMs) that have been obtained using embedded re-estimation after flat start initialization. These results are then used in neural network frameworks to build better acoustic models for speech synthesis/recognition. Segmentation using this approach requires large amounts of data and does not work very well for low resource languages. To address the issue of paucity of data, signal processing cues like short-term energy (STE) and sub-band spectral flux (SBSF) are used in tandem with HMM based forced alignment for automatic speech segmentation.STE and SBSF are computed on the speech waveforms. STE yields syllable boundaries, while SBSF provides locations of significant change in spectral flux that are indicative of fricatives, affricates, and nasals. STE and SBSF cannot be used directly to segment an utterance. Minimum phase group delay based smoothing is performed to preserve these landmarks, while at the same time reducing the local fluctuations. The boundaries obtained with HMMs are corrected at the syllable level, wherever it is known that the syllable boundaries are correct. Embedded re-estimation of monophone HMM models is again performed using the corrected alignment. Thus, using signal processing cues and HMM re-estimation in tandem, robust monophone HMM models are built. These models are then used in Gaussian mixture model (GMM), deep neural network (DNN) and convolutional neural network (CNN) frameworks to obtain state-level frame posteriors. The boundaries are again iteratively corrected and re-estimated.Text-to-speech (TTS) systems are built for different Indian languages using phone alignments obtained with and without the use of signal processing based boundary corrections. Unit selection based and statistical parametric based TTS systems are built. The result of the listening tests showed a significant improvement in the quality of synthesis with the use of signal processing based boundary correction.

Transfer Dictionary Learning Method for Cross-Domain Multimode Process Monitoring and Fault Isolation

Data-driven methods have shown its great latent capacity in the field of industrial process monitoring. However, the existing methods usually achieve good results under the assumption that the offline learning data and the online monitoring data are drawn from the same distribution. Unfortunately, in the industrial system, the assumption is often violated due to the harsh operating environment. Especially, with the increasing complexity and scale of industrial production, the supervisory control and data acquisition (SCADA) data of the industrial production process often collected from different machines, seasons, or operating modes. In addition, due to the cost of manual data labeling and real-time requirement of process monitoring, the offline learning data, which was used to build the model, often have abundant source-domain data and insufficient target-domain data. Consequently, these methods have bad performance on the online monitoring data collected from the target domain. In order to make full use of the knowledge from the abundant source-domain data, a transfer dictionary learning method is proposed to address the cross-domain problem in this article. The proposed method can learn an initial dictionary from the abundant source-domain data, and then, the final dictionary is updated by incorporating the feature of insufficient target-domain data in a smooth subspace interpolation way. The effectiveness of the proposed method is evaluated through a numerical simulation case, a continuous stirred tank heater (CSTH) case, and a wind turbine system case, from which we can see the proposed method has a better performance compared with some state-of-the-art methods.

Semi-supervised learning for ECG classification without patient-specific labeled data

In this paper, we propose a semi-supervised learning-based ECG classification system for detection of supraventricular ectopic beats (SVEB or S beats) and ventricular ectopic beats (VEB or V beats) which does not require manual labeling of the patient-specific ECG data. Owing to inter-subject variability in ECG signal, patient-specific data is usually required to achieve good performance in ECG classification system. However, manual labeling of patient-specific data requires expert intervention, which is costly and time consuming. Our proposed system is based on a 2D convolutional neural network (CNN) with inputs generated from heartbeat triplets. The system also consists of two auxiliary modules: a normal beat estimation module and an iterative beat label update algorithm. The normal beat estimation selects a small amount of patient-specific normal beats accurately from the testing ECG record in an unsupervised manner. These estimated normal beats are used, together with a common pool dataset, to train a preliminary patient-specific CNN classifier which provides initial labels for the testing data. These labels then undergo a semi-supervised iterative update process for improved performance. Our proposed system was evaluated on the MIT-BIH arrhythmia database. The training of our proposed system is fully automatic, and its performance is comparable with several state-of-art supervised methods which require extra manual labeling of patient-specific ECG data. Our proposed system can be a useful tool for batch processing a large amount of ECG data in clinical applications.

Vehicle Type Recognition From Surveillance Data Based on Deep Active Learning

Vehicle type recognition (VTR) from surveillance data has recently received increasing attention in both intelligent transportation and computer vision field. The deep learning techniques, e.g., convolutional neural networks (CNNs), have provided great progress in tasks with the large-scale labeled data. However, in real-world applications, manually labeling from large-scale surveillance data is time-consuming and tedious, which strongly obstructs the application of a CNN-based VTR. The existing works on VTR studied samples selection for manual annotation. Here, a VTR framework based on deep active learning that releases the burden of large-scale labeling is presented. The framework selects the most worthy samples to the manually selected, and then retrains the network with the annotated samples incrementally. Besides, the framework simultaneously takes advantage of both auto-labeled and manually-labeled data using the strategy of bi-direction entropy threshold. The proposed framework is validated on the public dataset Comprehensive Cars, and experimental results demonstrated that, with the incremental query strategy, the proposed framework could reduce annotation cost dramatically compared with random selection.

Automated Activity Identification for Construction Equipment Using Motion Data From Articulated Members

Real-time monitoring and assessment of construction resources have always been a challenge due to the unique, dynamic, and complex nature of each construction site and operation. The ability to automatically classify activities performed by various equipment in real time can aid in making timely tactical operational decisions that can lead to increased fleet productivity, reduced time and cost of operations, and minimized idle times. Such endeavors have traditionally been performed manually through human observation, making it extremely labor and time-intensive. Meanwhile, the development of low-cost micro-electro-mechanical systems (MEMS) with rapidly evolving computing, networking, and storage capabilities, along with the advances in computational techniques such as machine learning present new opportunities in the real-time activity identification domain. Even though previous studies have shown promising results for equipment activity identification at limited levels of detail, they have a fundamental limitation in their reliance on the equipment vibration. Equipment vibration is highly dependent on factors that are extrinsic to the performance of an activity itself such as ground conditions, age and condition of equipment, and operator skill. This aspect of current methods necessitates the collection of training data from the specific equipment of interest and requires manual labeling of training data, thereby limiting its application across different types of equipment and operating conditions. This paper investigates the use of activity-specific equipment motions instead of vibration for activity identification. This approach is the first step towards the larger goal of generating training data automatically from virtual kinematic models of equipment in the future. This paper also provides an array of sensitivity analyses in order to determine the most appropriate parameters for implementing machine learning algorithms for equipment activity identification. A case study was performed using an excavator working on an earthmoving site that demonstrated a significant improvement in equipment activity identification results by utilizing IMU data of different articulated elements over previous efforts. The results of this paper indicate that more accurate results for activity identification can be obtained by using articulated equipment motion over vibration, which paves the way for the automatic generation of labeled training data in the future.

ПРОГРАММНЫЕ СРЕДСТВА ДЛЯ АНАЛИЗА ПРИЗНАКОВ ТРАЕКТОРИИ ДВИЖЕНИЯ МОБИЛЬНОГО РОБОТА С ИСПОЛЬЗОВАНИЕМ НЕЙРОННЫХ СЕТЕЙ С ВНИМАНИЕМ

The article describes an approach to the implementation of a software package for determining the probability of a collision of a mobile robot with obstacles. This approach is based on a neural network model with attention. Key feature is the method of the training dataset generation: the labeling of obstacles and the values of the probability of collision with them is performed not in manual mode, but using a deterministic algorithm that uses the result of semantic segmentation using another pre-trained neural network. This method allows to use a poorly detailed description of the external environment for training convolutional neural networks with attention on the example of recognizing obstacles when a mobile robot moves in simulation mode. At the same time, low detail allows to reduce the time-consuming process of manual data labeling due to automatically generated sampling in the NVIDIA Isaac environment, and the attention mechanism allows to increase the interpretability of the analysis results.