This study proposed a new integrated decision framework that combined neutrosophic modeling, fuzzy representation, and multicriteria decision-making to produce an objective, transparent, and robust evaluation process for physical education teaching quality. The core idea was to represent each evaluation indicator using a neutrosophic triple that separated supportive evidence, indeterminate evidence, and opposing evidence, while fuzzy terms were used to capture gradual transitions in human judgments. Based on these representations, the study designed a two-stage weighting scheme that balanced evidence consistency across experts with data-driven dispersion across indicators, reducing the influence of biased or unstable opinions. Then, a robust ranking mechanism was introduced to compare teaching units against a clearly defined reference profile, producing an interpretable overall score and a final ranking. A complete case study was provided on university physical education teaching quality.
Traditional Chinese pleated clothing carries rich cultural meaning; however, its digital preservation has been challenged by labor-intensive manual annotation and limited generalization of existing computer-vision models. This study proposed a dual-visual-feature framework for pleat recognition by integrating global shape features and local pleat details. Based on a custom-built dataset of real-world Chinese traditional clothing images, the framework combined preprocessing, dual-feature extraction, dynamic fusion, and vision transformer base with 16 × 16 patch size and a multi-layer perceptron classification head for multilabel classification. Experimental results demonstrated effective recognition of traditional clothing pleats, including subtle hidden pleat structures. Visual attention heatmaps enhanced the model's interpretability, providing technical support for the digital documentation and preservation of traditional garments.
Font style transfer offers a unique window into human visual cognition, in which structural invariance and stylistic expression are processed in parallel. To emulate this dual mechanism, this paper proposes a cognitive-inspired hybrid architecture combining generative adversarial networks and vector quantization-variational autoencoders with multi-branch decoupling and perceptual feedback. The model achieves strong structural fidelity and style diversity under hard transformations, outperforming single-paradigm approaches in both objective metrics and preliminary aesthetic scoring. Designed with principles of natural intelligence such as few-shot generalization and sparse representation, the proposed system targets a modular deployment interface for real-world applications. Experiments on public font datasets validate its effectiveness and provide insights into the challenges of modeling human-like symbol generation.
This study aims to develop an artificial intelligence-powered augmented reality (AR) training system for real-time movement analysis and corrective feedback to enhance martial arts skill acquisition. A hybrid convolutional neural network-long short-term memory was adopted for high-precision recognition of complex martial arts techniques, with AR overlays delivering interactive guidance. The system leveraged red, green, blue cameras, depth sensors, and inertial measurement units for accurate motion tracking, trained on a 50,000-frame labeled martial arts dataset. Tested on smartphone-based AR platforms and evaluated via standard metrics, the convolutional neural network-long short-term memory model attained 94.3% accuracy—markedly higher than traditional training and existing artificial intelligence recognition systems. Real-time motion correction reduced technique errors by 28%, while AR-enhanced feedback boosted learning efficiency by 23%.
English for specific purposes (ESP) teaching resources are scarce and in demand. In traditional teaching, undergraduate student satisfaction is only 64.43%, and college teacher satisfaction is as low as 61.34%. This paper explores the practical path and application effect of corpus linguistics in ESP teaching. This study used big data modeling, independence detection of basic variables, and comparative experiments with 25 groups of teaching examples to build relevant big data models. After optimization, most samples reached over 95% authenticity, and errors were controlled within 1.5%. The results show that applying corpus linguistics can significantly improve ESP teaching scores, and hardware technical level and personalized sample demands have the strongest influence on teaching effectiveness. The conclusion shows that the model can effectively address ESP teaching pain points, support the development of personalized teaching programs in colleges and universities, and has strong promotional value.
