Abstract: Concrete crack detection is essential for maintaining aging infrastructure. However, conventional deep learning methods rely heavily on pixel-level annotations, which are costly and time-consuming to obtain. To overcome this limitation, we proposed a weakly supervised crack extraction framework that eliminates the need for pixel-level labels in our previous work. A ResNet-50 model with a Convolutional Block Attention Module (CBAM) is trained using image-level or patch-level annotations, and Score-CAM is employed to visualize crack regions. In this article, to enhance detection accuracy, a Multi-Scale Patch Fusion (MSPF) module is introduced, which processes subdivided image patches and fuses their outputs, thereby emphasizing local features and improving the detection of fine cracks. Although MSPF enhances crack detection, it can occasionally generate false positives due to background textures. To further address this issue, we introduce a False-Positive Patch Filtering (FPPF) module, an auxiliary CNN that automatically suppresses erroneous detections. Experiments on a benchmark dataset show that the combination of MSPF and FPPF achieves higher recall and IoU than the baseline, enabling robust crack detection without pixel-level annotations. Overall, the proposed framework substantially reduces annotation costs while maintaining high extraction accuracy, providing a scalable and practical solution for…
Abstract: Crack detection on concrete surfaces is essential for maintaining the structural integrity of aging infrastructure. However, supervised deep learning models often suffer from performance degradation when the training and test data distributions differ. This paper addresses covariate shift caused by differences in image resolution—termed Multi-Resolution Adaptation (MRA)—by introducing a one-step framework for crack segmentation. The proposed method jointly trains a U-Net–based segmentation model and a lightweight importance-estimation model that assigns weights to training images according to their estimated likelihood of appearing in the test domain. The training in this method is done in one step, reducing effects of underfitting inherent in two-step approaches and only requires unlabeled images in target domains. Class Activation Mapping (CAM) is employed to visualize the regions contributing to importance estimation, revealing that resolution differences are captured mainly through background surface texture rather than crack structures. Experiments conducted on a public dataset with artificially generated multi-resolution images demonstrate that the proposed approach improves Intersection over Union (IoU), particularly when low-resolution images constitute as little as 1% of the training data. These results confirm that one-step covariate shift adaptation effectively enhances model robustness to multi-resolution …
Abstract: Writer verification is a form of biometrics. Among its various approaches, finger-writing
verification of a simple symbol is aimed at being the most convenient, in which a user is verified by
writing a simple symbol with a finger on a smartphone screen. In a previous study, an error rate of
approximately 10 % was achieved. Furthermore, by selecting and fusing individually high-performing features, an equivalent error rate was obtained, even though the number of features used was reduced. In another previous study, we selected features that were resistant to tracing. In this study, we propose new feature selection methods focusing on individuality and independence, and a fusion method that combines them with two conventional selection methods. By using the proposed method, we finally achieved the error rate of approximately 6-7 %.
Abstract: The grading of SQL query assignments is a time-consuming task for instructors, especially in large classes where ensuring both accuracy and consistency is difficult. In this paper, we develop and evaluate an automatic grading tool for SQL query design using generative AI. The system targets SPJ (Selection, Projection, Join) query tasks from the Information Systems Experiment course. By incorporating model answers, lecture materials, and common student errors into carefully designed prompts, the tool aimed to improve grading accuracy and the quality of feedback. Comparative experiments were conducted using different prompt configurations and an existing rubric-based tool developed by Ando et al. The results demonstrated that the optimized GPT-based prompt design achieved 96.8% accuracy (387 out of 400 items correctly graded) and provided detailed feedback specifying “what was wrong,” “why,” and “how to fix it.” This feedback was found to support student learning and reduce grading inconsistencies for instructors. The findings confirm the potential of GPT-based systems as effective complements to existing rubric-based methods. This research highlights the promise of generative AI in delivering scalable, fair, and pedagogically meaningful assessments in database education.
Abstract: Background: Adaptive structural changes in biological systems have inspired a range of
network optimization algorithms, particularly for problems where global reconfiguration is impractical. In network design, integrating multiple independently constructed networks requires the effective selection of fusion edges to ensure stable and efficient connectivity. Objective: This study examines the adaptive network formation behavior of the slime mold Physarum polycephalum and proposes a heuristic fusion edge selection method based on flow-driven conductivity dynamics. Methods: The proposed method enables localized network integration by selecting candidate fusion edges based on the characteristic transport properties that emerge within the network. Results: Numerical simulations on random mesh networks demonstrate that the method effectively forms fusion paths between isolated networks and accelerates flow concentration along the resulting paths, even in heterogeneous network environments. The results further indicate that appropriate control of both the selection and the number of added edges is crucial for efficient network fusion under flow-conserved dynamics. Conclusions:
Although the proposed method does not explicitly optimize shortest path length and involves several hyperparameters, it provides a flexible framework for network fusion based on local interactions. These findings suggest that biologically inspired, flow-based heurist…
Abstract: In modern university education, managing diverse student inquiries poses a significant burden on instructors, highlighting the increasing importance of personalized support. To address this challenge, we developed an AI chatbot that provides responses tailored to specific course content, using lecture materials and video transcripts. The system is built on the OpenAI API (GPT-5) and integrates Retrieval-Augmented Generation (RAG) technology. The developed chatbot is deployed in two courses, "Requirements Engineering" and "Career Design for IT Engineers," and evaluated with 80 university students. Evaluation results indicate that approximately 80% of the students positively rated the chatbot's responses as "helpful", confirming its high utility in aspects such as organizing and re-presenting lecture content, suggesting future actions, and offering diverse perspectives. Furthermore, the use of past student comments and instructor responses as learning data proved highly valuable as a supplementary resource for enhancing student understanding. This research shows AI chatbots are practical tools for personalized support and reducing instructor workload. Future work aims to build a more personalized and effective learning support environment by incorporating features such as tracking individual student learning progress and enabling instructors to analyze conversation logs.
Abstract: Visual cues strongly influence postural control and fall risk, yet direct comparisons between Mixed Reality (MR) and Virtual Reality (VR) on center‑of‑gravity (CoG) sway are limited. This study investigated the effects of RE (real environment), MR, and VR on CoG responses during quiet stance using moving‑light gaze guidance visual stimuli. Ten healthy adult participants completed 12 experimental tasks, it consists of 3 environments (RE, VR, MR) and 4 frequency conditions (no‑light, low, medium, high frequency). MR produced largest CoG sway than RE and VR for both Total Track Length (mean ΔTTL: MR = 315.69 ± 288.92 mm; RE = 221.18 ± 280.15 mm; VR = 167.44 ± 177.86 mm, n = 30) and Trajectory Length per Unit (mean ΔTraj: MR = 2.59 ± 2.36 mm; RE = 1.79 ± 2.30 mm; VR = 1.38 ± 1.46 mm, n = 30). This study results show that MR tended to elicit greater sway, compared to RE and VR.
Keywords: balance control, body sway, postural stability, rehabilitation, virtual and mixed reality, visual stimuli.
Abstract: Fatigue is one of major contributing factor to traffic accidents worldwide. To save many human lives, research on detecting driver fatigue using various methods continues to be developed. Ease of implementation and low cost are key challenges in any research study. Another challenge is the ability to measure fatigue significantly and in more realistic scenarios. This study aims to measure fatigue levels using driver behavioral measurement methods with the addition of cognitive load, like real-world driver tasks. This study used embedded cameras to measure driver behavior. The parameters used were body and head posture, gaze, and changes in driver expression. While driving in a simulator, drivers are tasked with following the car in front of them while maintaining distance and speed, and applying the brake pedal when the car in front suddenly stops. By implementing a one-class support vector machine (OCSVM), it was found that, in general, although each parameter tends to vary between individuals, there are significant anomalies in facial position and rotation along with driving duration, indicating driver fatigue. Meanwhile, there was a slight anomaly in body position, although subjective measurements indicated that all participants felt a higher workload after using the driving simulator
Abstract: The number of isolated persons who suffer from an inability to seek help from others on their own is increasing, and this has become a social problem. To develop a solution to this problem, this study examines the effectiveness of an outreach agent in guiding and supporting isolated persons to a third place through a multi-agent simulation (MAS) using a social interactive swarm model. From the results of the MAS experiments, as a novel insight, we found that guiding isolated persons to a third place at specific intervals until their mental state calmed down, rather than guiding them to a third place one after another every time an outreach agent found them, was effective in reducing the state of isolation of those who were guided to the third place.
Keywords: Isolation, Third Place, Outreach Worker, Guiding Interval Coordination, Social Interactive Swarm Model, Multi-Agent Simulation, Iterated Prisoner's Dilemma Game.
Abstract: This study explores how people manage conflict when they belong to multiple groups and decide which groups to support at a given time. We introduced a computer simulation called the inter-role conflict game to mimic real-life situations in which individuals face competing demands. To represent the differences between adolescence and adulthood, we used Recurrent Q-Learning (RQL) agents that adjusted behavior based on past experiences. Our findings show that an agent’s ability to cooperate depends on how well it can predict future actions. When agents have a limited memory, they perform better by focusing on immediate rewards. In contrast, when agents can remember more information, they can plan ahead and work together more effectively over time. We also observed that gradual adjustments in the learning process using a low learning rate helped the agents become more stable and cooperative. This study provides insight into how short-term reactions and long-term planning can be balanced to improve cooperation in complex social settings.
Abstract: Digital Holographic Microscopy (DHM) can obtain three-dimensional (3D) information about the fine structure of an object by utilizing the phase information of coherent... Read More
Abstract: Combining with the register-transfer-level description, we aim to develop a generic hardware interface architecture for SPI devices using high-level synthesis that automatically converts... Read More
Abstract: Scattering noise reduction is a challenging project to remove noise under scattering media conditions such as fog or turbid water. In previous study,... Read More
Abstract: The recently discovered Static Electricity-induced Luminescence (SEL) material, SrAl₂O₄:Eu²⁺, exhibits the unique property of emitting luminescence in response to external energy generated by static electricity. This property addresses the long-standing challenge of visualizing static electricity, positioning the luminescence as a promising foundation for developing image-based static electricity sensing technologies. In this study, we developed a system for the quantitative evaluation of the SEL material using image analysis techniques. The experiment involved analyzing luminescence intensity and luminescence area of the SEL material over time under varying applied voltages. The results revealed a positive correlation between the applied voltage and both the luminescence intensity and the luminescence area. These findings highlight the effectiveness of the SEL material for quantitative assessment in static electricity sensing, offering significant potential for advancing static electricity visualization and measurement technologies.
