https://doi.org/10.5573/IEIESPC.2025.14.2.141

(Ying Liu)

With the Internet’s wide application, the mixed teaching mode (MTM) is gradually popularized in education. As an important part of digital media technology, animation design (AD) teaching is facing a new round of teaching reform. Based on the analysis of the characteristics of the mixed teaching mode, the research establishes the learning path of AD under this mode. Then the global optimal solution and adaptive inertia factor are introduced to improve the artificial bee colony (ABC) for feature selection of AD teaching resources (TRs). Meanwhile, the optimized Apriori is combined to apply to the mining and analysis of TR. The results showed that in the mining and classification of audio, graphics, images and other animation TR, the accuracy of the combined method was mostly in the range of 90% to 95%, with the highest of 97%. Among the scores of students majoring in digital media technology, most of the teaching models under the application of this method scored more than 80 points, and 90 points accounted for about 30%. It shows that the model can provide students with better learning experience, effectively improve the learning effect, which has certain practical application value in the development of digital media technology.

https://doi.org/10.5573/IEIESPC.2025.14.2.152

(Hang Yang) ; (Ren Wu) ; (Mitsuru Nakata) ; (Qi-Wei Ge)

This paper discusses a machine learning-based approach to optimize acupuncture and moxibustion treatment (AMT). The goal is to develop a model that can offer personalized acupoints prescriptions for patients based on their symptoms, enhancing both the efficiency and effectiveness of treatment. A database comprising symptoms and acupoints prescriptions for 3,000 disease cases was used, and 11 machine learning algorithms were applied to learn from this data. The training process utilized 90% of the data for 5-fold cross-validation and 10% for testing to assess generalization ability. Intersection over Union (IoU) was chosen as the key evaluation metric for the models. The Seq2seq model with attention mechanism emerged as the best-performing algorithm, achieving an IoU of 95.72% on cross-validation and 95.33% on the test set. These results suggest that using Seq2seq with attention can significantly reduce subjectivity in acupoint selection and increase the efficiency of AMT. This approach provides a promising data-driven method for improving treatment precision and saving time in clinical settings.

https://doi.org/10.5573/IEIESPC.2025.14.2.165

(Qianwen Zhang)

Modern educational concepts are student-centered, emphasizing personalized and comprehensive development of students. However, traditional English teaching methods have problems such as single teaching content and insufficient teacher resources. Based on the dynamic time warping, this study introduces clustering algorithm to construct an automatic intelligent platform for English teaching. On the basis of data analysis and clustering, the constructed English automatic teaching intelligent platform is evaluated. Experiments showed that the proposed algorithm could effectively improve the weaknesses of traditional algorithms. There was no crossing between data points. The clustering effect of data points was significant. The accuracy is the highest at 0.91 under different classification point splits. The research system platform had the highest correlation coefficient for speech recognition, ranging from 0.7 to 0.8. Its center frequency was within 1500. The center frequency variance was within 200. When the number of knowledge points was 50, the average teaching time using the proposed teaching system was reduced by nearly 300 minutes. In summary, the English automatic teaching intelligent platform can solve the problems existing in traditional English teaching methods, improving the effectiveness of English teaching.

https://doi.org/10.5573/IEIESPC.2025.14.2.178

(Bo Zhang) ; (Meichen Tao)

The multi factor mining system for stocks can provide favorable support for the analysis of financial markets. Due to the large number of influencing factors, the design of such systems generally faces significant deviation in prediction results. The continuous advancement of intelligent algorithm technology provides more diversified support methods for stock multi factor combination analysis, and also provides more technical support for risk assessment and problem decision-making in financial markets. Therefore, in order to achieve a more efficient stock multi factor analysis model, this paper constructs a stock multi factor intelligent mining system for financial markets based on the GA-TGCN intelligent algorithm. The GA-TGCN stock multi factor intelligent mining system based on TGCN technology and integrating GA algorithm provides more support for the prediction of multi factor combination analysis. By predicting and analyzing factors and strategy returns, excellent multi factor strategies can be obtained by combining simulated trading. Combining the application of different models in the process of multi factor mining combinations, it was found that the GA-TGCN system can achieve higher accuracy and lower loss values while achieving smaller prediction errors, laying the foundation for improving the efficiency of stock multi factor combination analysis.

https://doi.org/10.5573/IEIESPC.2025.14.2.191

(Cheolsu Kwag) ; (Sung Soo Hwang)

This survey paper explores recent advancements in neural rendering, focusing on the development and impact of Neural Radiance Fields (NeRF). Initially, 3D reconstruction relied on methods such as Photogrammetry, Structure from Motion (SfM), and Image-Based Rendering (IBR). While these techniques provided foundational approaches for creating 3D models from 2D images, they were limited in resolution, texture fidelity, and computational efficiency. IBR, in particular, was crucial in producing photorealistic environments using actual photographs but faced challenges in terms of flexibility and rendering complex scenes. NeRF emerged as a novel solution, utilizing neural networks to render 3D scenes from 2D images with improved realism and efficiency. This paper examines four key areas of advancement in the post-NeRF era: 1) Training and Rendering Speed: Enhancements in the efficiency of neural rendering algorithms; 2) Quality: Improvements in image realism and texture representation. 3) 3D Geometry/Reconstruction: Advancements in achieving accurate and detailed 3D models; and 4) Neural Scene Editing: Innovations enabling dynamic modifications in neural-rendered scenes.

https://doi.org/10.5573/IEIESPC.2025.14.2.205

(Peng Gao) ; (Jinlin Ruan) ; (Yuan Sun) ; (Hao Li) ; (Cong Sang)

Belt conveyors are common transportation equipment, and optimizing their performance is an effective measure to achieve intelligence and efficient transportation. The application of new technologies has increased the controllable parameters and operational parameters to be optimized in belt transportation systems, making performance prediction and multi-objective optimization problems more challenging. Traditional response prediction and optimization methods have become increasingly inadequate to meet research requirements. The emergence and development of machine learning methods and modern intelligent optimization methods have provided new directions for the prediction and optimization research of belt transportation systems. Therefore, in order to improve the system prediction and performance optimization of belt transportation systems, this study proposes an interpretive machine learning approach based on improved neural networks and support vector machines, which combines the optimization of network initial weight threshold and training set test set, and establishes a high-performance predictive interpretive machine learning model. The results indicate that under different combinations of training and validation set partitioning, all evolutionary processes in the trajectory of the initial weight threshold optimization of the proposed comprehensive neural network optimization method reach their optimal state in the 17th generation. The optimization algorithm stably converged to the optimal value after only 14 generations of evolution, resulting in a mean square error of 0.011678 for the optimal network prediction, while the mean square error of all individuals in the initial population was 0.016845. If the average network prediction performance of all individuals in the initial population is taken as the basic standard, a comprehensive optimization algorithm is used to optimize the network’s prediction performance to 29.6%. After the 10th reinforcement training, the prediction error of the support vector machine decreases to 0.99%, and the gross of simulations is 65. The designed method helps to achieve intelligent transportation systems, unmanned operation and high performance, and can achieve sustainable development while ensuring transportation safety and efficiency.

https://doi.org/10.5573/IEIESPC.2025.14.2.218

(Yunhong Guo) ; (Shihao Zhang)

To effectively process and analyze these data, this article designs a reasonable storage architecture and data processing flow. This article studies real-time monitoring technology to detect and process network attacks in a timely manner. The platform designed in this article uses 500 CPU cores and 1TB of RAM. This study can deal with different types of network attacks such as DDoS attacks, malware infections, and phishing websites.The CPU utilization and memory utilization are 65% and 70%, respectively.In the case of intricate APT attacks, the utmost response time is set at 4 hours. Remarkably, 99% of identified threats are effectively addressed within just 1 hour of detection. The network security situational awareness platform, which leverages big data technology, has attained noteworthy achievements in practical applications. This platform can monitor network traffic in real-time, detect abnormal behavior, and provide warning for potential threats, providing strong support. This has practical significance for improving network security defense capabilities and ensuring network information security.

https://doi.org/10.5573/IEIESPC.2025.14.2.229

(Zhehua Fan) ; (Kun Ma)

In response to the shortcomings of low recognition accuracy and long recognition time in current sports action recognition models, this study combines extended Kalman filtering and microelectromechanical system sensors to build a new sports action recognition model. Firstly, the full angle pose calculation method is introduced to optimize the extended Kalman filtering algorithm. Then, the optimized pose estimation algorithm is combined with microelectromechanical system sensors to build the final sports motion recognition model. The research results indicated that the estimation error of the optimized attitude estimation algorithm was as low as 0.01. The motion recognition model constructed had high accuracy rates of 0.99, 0.98, and 0.98 for recognizing serve, drop, and spike movements in badminton, with a time consumption of 2.01 s, 1.88 s, and 1.96 s, respectively, demonstrating good recognition performance. The above results indicate that the attitude estimation algorithm and recognition model designed in this study have good performance and practical application effects, and can provide new reference methods for intelligent recognition of sports.

https://doi.org/10.5573/IEIESPC.2025.14.2.242

(Ying Jiang)

Despite growing relief 3D printing in arts and manufacturing, issues like inefficient path planning and inadequate accuracy persist. To tackle these problems, a Q-learning algorithm, coupled with an AI evaluation model, is employed to optimize path planning during 3D printing. Within a multi-objective optimization structure, the algorithm is trained to minimize print time, material usage, and enhance accuracy. Relying on a reward mechanism that simulates various path alternatives, the Q-learning algorithm iteratively fine-tunes path selection to strike the optimal cost-efficiency equilibrium. The results showed that after applying the Q-learning algorithm, the path planning efficiency of the relief process was significantly improved. In relief process 1, the optimization time of the Q-learning algorithm was 3.2 hours, with an accuracy of 96.13%. Compared with the 5.4 hours and 90.75% of genetic algorithm and the 6.2 hours and 91.44% of particle swarm algorithm, it showed significant advantages. After 40 iterations, the stability in printing time, material usage, and accuracy metrics indicated that the model reached a stable optimal solution. This research not only enhances relief 3D printing efficiency and accuracy but also presents a universally applicable path optimization approach for other 3D printing technologies, unlocking fresh opportunities for bespoke and customized relief art production.

https://doi.org/10.5573/IEIESPC.2025.14.2.257

(Lei Li) ; (Huimin Peng) ; (Xingxue Ren) ; (Qianqian Wang)

Drones have shown enormous potential in urban logistics due to their efficiency and flexibility. However, traditional path planning methods such as particle swarm optimization and A* algorithm often find it difficult to meet both efficiency and safety when used alone. Therefore, this study proposes a new unmanned aerial vehicle logistics distribution path planning method. By adjusting parameters and optimizing search strategies, the particle swarm optimization algorithm utilizes the efficient pathfinding ability of A* algorithm to ensure its security. The results show that the obstacle avoidance success rate of the model is 94.85%, which is the best performance compared to other comparative algorithms and provides the shortest and smoothest path selection. This method demonstrates good path planning efficiency and stability, improving logistics and distribution capabilities in urban environments. This provides valuable reference for intelligent path planning and intelligent transportation systems.

https://doi.org/10.5573/IEIESPC.2025.14.2.268

(Mingfeng Yan)

The design of B&B space presents challenges in extracting and matching Mazu pattern elements. This study introduces an efficient approach based on the fast robust feature algorithm to address this issue. Initially, an image-aware hash model is created, incorporating the fast robust feature algorithm. Subsequently, an enhanced Siamese network model is established, integrating the fast robust feature algorithm. Results indicate that the improved fast robust feature algorithm exhibits superior robustness compared to the traditional approach, achieving a matching ratio of 0.4 to 0.6. The proposed algorithm attains 93.53% and 93.91% image retrieval accuracy on self-built and Mnist datasets, surpassing other comparison algorithms. Through grayscale histogram and perceptual hash algorithm integration, the method enhances recognition accuracy during image deformation, especially under rotation and scale changes. Although encoding times are longer at 8.12 and 5.25 seconds, respectively, the proficient handling of rotation and scale invariance remains unaffected. This study offers an effective solution for precise feature extraction in intricate patterns within B&B space design, particularly in managing image rotation and scale alterations, presenting robust technical support for image processing and pattern recognition.

https://doi.org/10.5573/IEIESPC.2025.14.2.280

(Seong Jun Yoon) ; (Teresssa Talluri) ; (Amarnathvarma Angani) ; (Hee Tae Chung) ; (Kyoo Jae Shin)

The increase in battery temperatures results in the critical risks, including explosions, therefore need of efficient thermal management is increasing. In this point of view, we proposed a novel approach to battery thermal control, employing hot soaking and cold soaking experiments for the first time to identify phase change materials (PCMs) that enhance battery safety under temperature conditions. Machine learning methods such as Llng short-term memory (LSTM) and random forest (RF) models were applied and thermal performance was investigated in lithium polymer pouch batteries integrated with PCMs for fast and accurate prediction. Experiments were conducted at normal temperature of 25?C, hot temperature of 50?C, and cold temperature of ?10?C. Thermal performance metrics such as maximum temperature and thermal gradient ?T were measured during discharge of the battery. In this study we selected PCMs such as RT15, RT31, EG5, EG26, and EG28 to evaluate the performance with LSTM and RF are applied to predict temperature variations influencing thermal behavior. Results indicated that EG26 and EG28 PCMs, significantly improved thermal performance under extreme conditions. The LSTM model demonstrated high predictive accuracy of 99% compared to RF model with 97%. This integrated model approach provided both high predictive accuracy and valuable insights into battery thermal performance, underscoring the importance of PCM selection to ensure battery longevity and stability across diverse environments.