International Information Technology University

Road surface deterioration poses significant challenges for transportation safety, infrastructure maintenance, and intelligent transportation management. Automated road damage analysis based on computer vision has therefore become an important research topic in recent years. This study proposes a novel deep learning framework for multi-task road damage analysis that integrates detection, classification, and segmentation within a unified architecture. The proposed model incorporates multi-scale feature extraction, transformer-based contextual refinement, ROI-based classification enhancement, and a boundary-aware segmentation module to improve the identification of diverse pavement defects. Multi-scale alignment enables the network to capture road damage patterns at different spatial resolutions, while contextual attention mechanisms strengthen feature relationships across complex road scenes. The boundary-aware segmentation branch further improves structural understanding of pavement defects by enabling precise contour delineation. Extensive experiments demonstrate that the proposed architecture achieves superior performance compared with conventional detection models, improving both detection accuracy and segmentation quality while maintaining practical inference speed. Ablation studies further confirm the effectiveness of each architectural component and highlight the contribution of contextual feature refinement and boundary-aware learning to overall model performance. The results indicate that the proposed framework provides a reliable and efficient solution for automated road condition monitoring and intelligent infrastructure inspection. This approach can support real-time road maintenance systems and contribute to safer and more efficient transportation networks.

Automatic processing of morphologically rich, agglutinative, and low-resource languages remains challenging because productive affixation increases lexical sparsity, weakens statistical generalization, and often produces inconsistent predictions across related linguistic annotation tasks. This study presents QazNLP, a constraint-aware multi-task framework for Kazakh that jointly performs morphological tagging, part-of-speech tagging, and named entity recognition using a shared transformer encoder with task-specific prediction heads. To improve structural reliability, the framework introduces differentiable cross-task compatibility penalties that discourage linguistically invalid label combinations during training and constrained decoding. The study further provides a reproducible evaluation setting based on a cleaned Kazakh news corpus with fixed data splits and robustness diagnostics for out-of-vocabulary tokens, long agglutinative word forms, reduced-data regimes, and cross-task contradiction analysis. In addition, the manuscript explicitly documents the encoder and tokenization setup, justifies the choice of pretrained encoder, reports decoding complexity, and situates the NER component with respect to the KazNERD benchmark. Experimental results show that the proposed model consistently outperforms competitive single-task and shared multi-task baselines in joint average F1, robustness under sparse-data conditions, and structural consistency of predicted labels. The findings indicate that explicit compatibility-aware optimization offers a practical and extensible direction for sequence labeling in morphologically rich low-resource languages.

Conventional methods, including quantification of evidence and fuzzy logic, are not sufficient to eliminate the complex nonlinear correlations present in geological data sets obtained from various sources The integration of satellite data into the automation of mineral exploration is important for modern techniques that optimise the detection of promising ore deposits in conditions of resource depletion, growing demand for materials and the need to ensure economic stability. This paper presents deep learning models such as convolutional neural networks (CNNs) for automated detection of copper porphyry deposits based on satellite imagery. The framework employs Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) imagery with eight mineral alteration bands, integrated with 108 known deposit sites from Australian porphyry copper regions. Our baseline CNN achieved 91.0% classification accuracy and 0.93 area under the receiver operating characteristic curve (AUC-ROC) on test data. A comparative analysis of seven deep learning architectures showed that the Attention-CNN model outperformed others, achieving 93.0% accuracy and 0.95 AUC-ROC. This demonstrates that deep learning can effectively capture spatial-spectral patterns, offering a scalable solution for prospectivity mapping while reducing exploration costs and environmental impact.

This paper presents an extended study of a phoneme-based approach to speaker identification in the Kazakh language using ultra-short utterances. Unlike the previous study conducted on a corpus of 20 speakers, the present experiment was conducted on 145 speakers, which made it possible to verify the reproducibility of the identified patterns under conditions close to real biometric authentication systems. The material consisted of short words with a Consonant–Vowel–Consonant (CVC) structure. Four models based on the ECAPA-TDNN architecture were developed: three phoneme-based models (for the vowels "A," "O," and "E") and one generalized model (AEO). The data were normalized in terms of loudness and cleaned of leading and trailing pauses using VAD. The training and test corpora did not overlap in terms of speakers. For test utterances ranging from 3 to 24 words, all models exhibited an increase in accuracy following a power-law trend as the utterance length increased. For short utterances, phoneme-based models retained an advantage, particularly the model based on the vowel "E," which is associated with phonetic stability and consistent formant structure. Expanding the corpus to 145 speakers reduced the variability of the results and confirmed the robustness of the observed effects. The practical value of this study lies in the fact that using words dominated by the vowel "E" in short voice passwords can improve identification accuracy without increasing command length or the amount of training data, which is especially important for voice authentication systems in low-resource languages.

This paper introduces a polynomially focused methodology for addressing the subset sum problem and evaluates its relevance to unstructured information retrieval. The research addresses the challenge of determining a subset of positive integers that totals a certain target value and formulates an accurate solution framework utilizing combinatorial mappings, symmetry characteristics of binomial coefficients, and a merge-based building method. The core concept of the proposed methodology is to condense the original combinatorial parameter into a more compact search structure, thereby reducing the number of potential combinations and enhancing computational feasibility. Based on this, generalized methodologies for the subset sum issue and associated k-SUM formulations are established. The results indicate that the suggested method can streamline the exact search process relative to traditional exponential or pseudo-polynomial approaches, while maintaining solution accuracy. The practical relevance of the approach pertains to search activities inside extensive and unstructured data environments, where effective filtering and combination matching are crucial. In the context of sustainable development, the proposed method contributes to the design of more resource-efficient digital intelligence systems by reducing redundant search operations, memory usage, and computational overhead in large-scale information processing tasks. Such efficiency is important for sustainable decision-support systems, including public administration, environmental monitoring, smart-city services, and policy-oriented information retrieval.