February 28 · Article · Eastern-European Journal of Enterprise Technologies · Civil and Structural Engineering
The object of the research is in-pipe defect detection and classification. The primary problem to be solved is the inefficiency, high cost, and inaccuracy of traditional manual inspection methods, which are often time-consuming and prone to human error. The results obtained include the creation of a multi-modal platform that integrates Red-Green-Blue (RGB) imaging and depth data with advanced artificial intelligence algorithms, Canny edge detection, and Density-Based Spatial Clustering of Applications with Noise (DBSCAN) clustering, achieving a 93 % mean Average Precision (mAP) in detecting and classifying various defects such as cracks, corrosion, and debris. A brief interpretation of the findings reveals that the high performance is due to the synergy between multi-modal sensing, artificial intelligence pattern recognition, and robust robotic navigation. This integrated approach ensures that the system not only detects defects accurately but does so in real time. Features and characteristics of the obtained results that directly address the identified problem include real-time high-precision defect identification, and reduced inspection downtime. As a result, inspection time is shortened, costs are lowered, and the safety of the pipeline system is increased, leading to accurate measurement of indicators (93 % mAP) and a reduction in occupational safety risks. The developed system is designed for use in traditional industrial environments, especially in large pipeline networks and in conditions where traditional methods are ineffective
LKLazzat KurmangaliyevaOpen access
February 28 · Article · Eastern-European Journal of Enterprise Technologies · Artificial Intelligence
This paper considers the efficiency of neural networks for human voice recognition. The objects of the study are artificial neural networks used for human voice recognition. Their ability to effectively recognize a human voice regardless of language, trained on a small number of speakers in noisy conditions, has been considered. The task being solved is to enhance the accuracy of speech activity detection, which plays a significant role in improving the functioning of automatic speech recognition systems, especially under conditions of a low signal-to-noise ratio. The findings showed that the accuracy of human voice recognition in languages of different phonetic proximity could vary greatly. As a result of the study, it was found that the recurrent neural network (RNN) demonstrates high accuracy in voice recognition – 95 %, which exceeds the results of the convolutional neural network (CNN), reaching an accuracy of 94 %. Special features of the results are the adaptation of neural networks to multilingual features, which made it possible to increase the efficiency of their work. An important conclusion was that training neural networks on data with different languages and types of speakers significantly improves recognition accuracy. The study confirmed that training neural networks on different languages and speaker types could significantly affect recognition accuracy. The results are an important contribution to the development of speech recognition technologies and have the potential for application in various fields where high accuracy in human voice recognition is required
AAAinur AkhmediyarovaOpen access
February 27 · Article · Eastern-European Journal of Enterprise Technologies · Artificial Intelligence
The object of the study is the text classification and semantic search tailored to the unique linguistic features of the Kazakh language. The research addresses the challenge of improving the accuracy, relevance, and efficiency of semantic search. This study focuses on improving semantic search for the Kazakh language by analyzing computational models tailored to its unique linguistic features, such as agglutinative morphology and rich inflectional systems. The research compares traditional rule-based approaches and advanced transformer architectures, including fine-tuned models like RoBERTa, for their ability to handle semantic nuances, contextual relationships, and user intent. The results reveal that fine-tuned transformer models achieved significant advancements, with the RoBERTa model attaining a Precision@10 of 89.4 %, a Mean Reciprocal Rank (MRR) of 85.6 %, and an F1-Score of 88.0 %. Additionally, the semantic search system developed in this study demonstrated a precision of 88.4 %, recall of 87.6 %, and an F1-score of 88.0 % on a domain-specific Kazakh dataset. Key to these improvements were innovations in preprocessing pipelines, including custom tokenization and lemmatization tailored to Kazakh's agglutinative morphology, and the integration of contextual embeddings to resolve issues such as synonymy and homonymy. Computational efficiency was enhanced through resource optimization techniques, enabling the deployment of these advanced models in constrained environments. These findings underscore the potential of tailored transformer models to bridge the gap in semantic search capabilities for underrepresented languages like Kazakh, advancing the inclusivity of natural language processing technologies
AAAigerim AitimOpen access
