Digital Library

Vol. 21, No. 2, Apr. 2025  Cover

1. Multi-Feature Fusion for E-Learning Based Student Concentration Analysis
   Changjian Zhou, He Jia, Jinge Xing, and Yunfu Liang
   Vol. 21, No. 2, pp. 114-124, Apr. 2025  
   https://doi.org/10.3745/JIPS.04.0339
   Details Full Text PubReader
   Keywords: Artificial intelligence, Concentration Analysis, educational technology, Multi-Feature Fusion, online learning
   Show / Hide Abstract

   Since the outbreak of COVID-19, the hybrid teaching style, combining online and offline methods, has evolved into a normal pedagogical approach. In offline classrooms, teachers may pay attention to the state of students and observe whether they are listening attentively, to adjust the teaching process in time. However, in the Elearning environment, teachers are hindered by their inability to access students’ states in time. Particularly, it is challenging to find out whether students are distracted in class. Although there are various existing student concentration analysis models, the metrics, such as convenience and accuracy, of these models often fail to meet the expected requirements. To address these obstacles, a multi-feature fusion method is proposed for Elearning-based student concentration analysis in this work. In this study the 300 questionnaires were collected and seven factor features were summarized. To establish the experimental dataset, 2,000 video clips were acquired, and each one was labelled with one of the five-class concentration degree scores. Finally, the ResNet-50 deep learning model with multilayer perceptron layer was employed for training and fine-tuning. Experimental results demonstrated that the proposed method achieves 0.88 accuracy, outperforming the existing state-of-theart concentration analysis methods. The proposed method is designed to detect distracted students and to provide reference for teachers to adjust E-learning arrangements, which is of great application value.


2. Predicting Chinese Stocks Using XGBoost-LSTMAttention Model
   Zhiyong Yang, Yuxi Ye, and Yu Zhou
   Vol. 21, No. 2, pp. 125-138, Apr. 2025  
   https://doi.org/10.3745/JIPS.04.0340
   Details Full Text PubReader
   Keywords: LSTM, Attention Mechanism, Extreme Gradient Progression Tree, Stock Forecasting
   Show / Hide Abstract

   Forecasting is a popular topic in the stock market. In recent years, many scholars have utilized machine- and deep-learning models in this field. However, many stock forecasting models suffer from problems of information overlap in stock trading data and a relatively simple structure of the prediction model. To overcome these issues, we built a stock forecasting model based on extreme gradient boosting (XGBoost), long shortterm memory (LSTM), and attention (XGBoost-LSTM-Attention). XGBoost is used to extract important information from stock data, and the LSTM combined with the attention mechanism can enhance stock prediction performance. To verify the feasibility and effectiveness of XGBoost-LSTM-Attention, we selected 14 Chinese stocks from different industries for the prediction experiments and compared their performance with those of existing models. The experimental results showed that the average root-mean-square error value of the XGBoost-LSTM-Attention model for the different stock datasets was the smallest (0.012); the average R2 value (0.96) and average accuracy (66.1%) were the highest.


3. A Multiple Receptive Field Anti-occlusionSelf-supervision Network with CBAM for Light FieldAngular Super-resolution
   Wan Liao, Qian Zhang, Jiaqi Hou, Bin Wang, Yan Zhang, and Tao Yan
   Vol. 21, No. 2, pp. 139-151, Apr. 2025  
   https://doi.org/10.3745/JIPS.01.0112
   Details Full Text PubReader
   Keywords: Angular Resolution, CBAM, Light Field, Self-supervision Learning
   Show / Hide Abstract

   By recording high-dimensional light data, a light field (LF) can accurately perceive a dynamic environment, thus supporting the understanding and decision-making of intelligent systems. However, with the discrete sampling of high-dimensional signals, LF faces have insufficient efficacious acquisition of LF information. This study tackles this problem by introducing a self-supervised learning approach that uses convolutional neural networks with varying receptive fields for processing sparse view inputs and subsequently generating a dense view through warping. The primary basis relies on the fact that the inherent correlation of the LF data and the convolutional block attention module (CBAM) are applied to process the LF data and wrap the operation into a layer to construct a deep network. The proposed method eliminates occlusions and achieves super-resolution LF angle reconstruction. Extensive experiments on an HCI dataset demonstrated that the proposed model outperforms recent state-of-the-art models.


4. Implementation of Extreme Programming Method in Developing Website of Agribusiness Harvest Transportation Order Data Management
   Ginanjar Wiro Sasmito, Slamet Wiyono, Edy Irwansyah, and Derwin Suhartono
   Vol. 21, No. 2, pp. 152-162, Apr. 2025  
   https://doi.org/10.3745/JIPS.04.0341
   Details Full Text PubReader
   Keywords: Agribusiness, eXtreme Programming, Transportation
   Show / Hide Abstract

   Agribusiness is the largest and most significant sector in the national economy of Indonesia. In running and sustaining a business of agriculture, plantation, forestry, fishery, and animal husbandry fields, or called agribusiness, a lot of problems may arise. One of the obstacles occurring during the process of transporting agribusiness crops is the management of harvest transportation data. The mismanagement of processing data for the transportation of agricultural crops is one of the major factors that hinders the distribution of crops after harvesting. Inaccessibility of the place to rent transportation means, unstable cost of the transportation rent, and insufficient number of transportation to load all the crops made it difficult to fix the recording of the data. Therefore, the website application of renting transportation for crops, which includes: the number and identity of business owners of transportation, the type and number of transportation, transportation order data, as well as the transaction data of post-harvest transportation, is developed by the extreme programming (XP) method. This method has been chosen since it can speed up the website development process compared to the prototyping method. This method only takes 84 days to release with better design quality and function, and it can also reduce the cost and optimize productivity. This website application is developed using the PHP programming language with the Laravel and Bootstrap frameworks to produce a UI/UX-responsive web design. Meanwhile, MariaDB is used as a relational database management system.


5. Research on the Tracking Registration Method of Markerless Augmented Reality for Product Assembly
   Pengxia Cao and Yibo Huang
   Vol. 21, No. 2, pp. 163-179, Apr. 2025  
   https://doi.org/10.3745/JIPS.04.0342
   Details Full Text PubReader
   Keywords: Augmented Reality (AR), Iterative Closest Point (ICP) Registration, LINEMOD Template Matching, Product Assembly, Tracking Registration Method
   Show / Hide Abstract

   To solve the problems of assembly object variety, volume difference, and texture features not being obvious in the intelligent products assembly process, augmented-reality technology was employed. To improve the feasibility and robustness of the augmented-reality tracking registration algorithm in the product assembly scenarios, combined with the weak texture of product assembly operation objects and the complex scenes, a real-time and robust unmarked tracking registration method was proposed. First, the multimodal approach of LINEMOD was improved to detect the weakly textured assembly target base so that the rough pose of the camera can be obtained and the problem of reinitialization can be solved when the process of tracking registration is interrupted. Then, the iterative closest point (ICP) registration algorithm was improved to accelerate the registration process while accurately positioning the camera pose to determine the benchmark for virtual-real fusion. Finally, based on the pose relationship between the assembly and the assembly guidance information established in the offline phase, visual guidance of the assembly process was realized. A performance analysis and comparison of the proposed tracking registration method show that the proposed method has better accuracy than the LINEMOD template matching method. The visual display of the assembly guidance process shows that the proposed method can be applied to the assembly scene of products lacking texture to realize intelligent assembly.


6. Weakly Supervised Aspect Category Detection Method Based on Label Hierarchical Filtering Mechanism
   Yan Xiang, Anlan Zhang, Jinlei Shi, and Yuxin Huang
   Vol. 21, No. 2, pp. 180-192, Apr. 2025  
   https://doi.org/10.3745/JIPS.04.0343
   Details Full Text PubReader
   Keywords: Aspect Category Detection, Aspect-Level Sentiment Analysis, Label Filtering, Topic Model, Weakly Supervised Method
   Show / Hide Abstract

   Aspect category detection is crucial for aspect-level sentiment analysis. Traditionally, supervised methods require a large number of labeled samples to achieve effectiveness, whereas unsupervised methods rely heavily on human judgment, which can compromise identification accuracy. To address these limitations, we propose a weakly supervised aspect category detection method that uses a hierarchical label filtering mechanism. Our approach begins by assigning preliminary pseudo-labels to comments based on topic similarity. Subsequently, unreliable pseudo-labeled samples are filtered out using semantic similarity. Finally, high-confidence training samples are selected using both high and low thresholds. Using the hierarchical label filtering mechanism employed in these three stages, we constructed a high-confidence training dataset, which was used to train a topic representation-enhanced classifier for aspect category detection. The proposed method, which was evaluated on three publicly available datasets, outperformed existing baseline models, thereby reducing the need for human intervention.


7. Data-Driven Airflow Prediction for Wastewater Treatment Plant Aeration System
   Xuefei Li, Changqing Liu, Shuqi Liu, and Sheng Miao
   Vol. 21, No. 2, pp. 193-203, Apr. 2025  
   https://doi.org/10.3745/JIPS.04.0344
   Details Full Text PubReader
   Keywords: Airflow Prediction, Decision support, Machine Learning, Wastewater treatment plant
   Show / Hide Abstract

   A wastewater treatment plant is an intricate system with a wealth of information, where the aeration system of the active sludge process is designed to provide oxygen to microorganisms. Owing to the time delay in biochemical reactions, adjustments made by operational staff to the airflow often lead to delayed changes in dissolved oxygen concentration, frequently causing overaeration. The paper introduces a machine learning model that utilizes water quality indicators and air blower indicators to predict current airflow. By leveraging the airflow predicted by this model, the dissolved oxygen concentration for the next hour is successfully maintained within the optimal range of 2 mg/L to 4 mg/L. In the case of airflow prediction, the Transformer model proved more effective than the random forest and long short-term memory models, owing to its selfattention model architecture. In conclusion, the study demonstrates the successful applicability of machine learning models to predict airflow on the promise of maintaining dissolved oxygen stability. These findings present a data-driven approach to guarantee the steady operation of wastewater treatment plants.


8. Optimizing Energy Storage Systems Using Deep Reinforcement Learning in Smart Grids
   Hajin Noh and Yujin Lim
   Vol. 21, No. 2, pp. 204-215, Apr. 2025  
   https://doi.org/10.3745/JIPS.04.0345
   Details Full Text PubReader
   Keywords: Energy Storage System, Reinforcement Learning, Smart Grid
   Show / Hide Abstract

   With the progress of IT technology, it has become possible to reduce consumer costs in the power market by using energy storage systems (ESS) in smart grids. Traditional algorithms proposed to solve optimization of ESS problems are difficult to apply to dynamic situations, hence adaptable and relatively simple designs such as deep reinforcement learning (DRL) techniques have begun to be used instead. In this study, a Markov decision process is designed to determine the charging and discharging amounts within a certain range to extend the lifespan of the ESS. Furthermore, DRL techniques such as deep deterministic policy gradient (DDPG), twin delayed deep deterministic policy gradient (TD3), and soft actor-critic (SAC) were trained, and their performances were compared for analysis.


9. Constructing S-Box Using Fractional-Order Nonlinear Coupled Map and Elementary Cellular Automata
   Liyan Liu, Cheng Zhang, and Yingqian Zhang
   Vol. 21, No. 2, pp. 216-226, Apr. 2025  
   https://doi.org/10.3745/JIPS.03.0204
   Details Full Text PubReader
   Keywords: Elementary Cellular Automaton, Fractional-Order Nonlinear Coupled Map Lattices, S-box
   Show / Hide Abstract

   A novel substitution box (S-box) construction method is proposed by using a fractional-order nonlinear coupled map lattices chaotic system combined with an elementary cellular automaton. An original S-box is produced by utilizing chaos sequences from the fractional-order nonlinear coupled map lattices spatiotemporal chaotic system. The S-box elements are then rearranged by applying different state values of various cells in the elementary cellular automaton. These chaotic sequences with independent properties are subsequently used to perturb the S-box’s elements. Comparative results with previous schemes indicate that the constructed S-box demonstrates enhanced security performance and could be effectively utilized in the development of block encryption algorithms.