|
Affiliation |
Information Technology center |
|
Title |
Associate Professor |
|
External Link |
|
AOKI Kenji
|
|
|
Degree 【 display / non-display 】
-
Doctor (Engineering) ( 2010.3 Kagoshima University )
-
修士（情報工学） ( 2000.3 九州工業大学 )
Research Areas 【 display / non-display 】
-
Informatics / Database
-
Informatics / Theory of informatics
-
Humanities & Social Sciences / Educational technology
Papers 【 display / non-display 】
-
Unsupervised Defect Detection for Automatic Shiitake Sorting Reviewed
Nokura S., Kimura L., Ishimaru T., Oshikawa Y., Ikeda S., Aoki K., Ohe K., Takei A., Kawamura R., Sakamoto M., Sugimoto K.
Proceedings of International Conference on Artificial Life and Robotics 430 - 433 2026
Language：English Publishing type：Research paper (international conference proceedings) Publisher：Proceedings of International Conference on Artificial Life and Robotics
Automated shiitake sorting faces severe data imbalance (1,731 good vs. 241 defective images) and annotation difficulties. We compared data augmentation and unsupervised detection. First, augmenting scarce defective data with GANs failed; models either lost subtle defect features (e.g., discoloration) during pre-processing or overfit to augmentation patterns, proving label-less augmentation difficult. We then shifted to unsupervised anomaly detection (VAE+OC-SVM) trained only on good data. This model achieved perfect Recall (100%) for the defective class, identifying all bad items without a single miss. This Recall 100% capability demonstrates its high practical utility as a primary screening tool for quality control, particularly where preventing defective product leakage is prioritized (albeit with low precision). This approach highlights a path for automated sorting in data-scarce agricultural settings and suggests high transferability to other products (e.g., tomato, potato) with similar data constraints.
-
Broadening Access-to Creative Experience with MR 3D Painting Reviewed
Ishimaru T., Oshikawa Y., Nokura S., Ikeda S., Aoki K., Ohe K., Takei A., Kawamura R., Sakamoto M.
Proceedings of International Conference on Artificial Life and Robotics 421 - 425 2026
Publishing type：Research paper (scientific journal) Publisher：Proceedings of International Conference on Artificial Life and Robotics
Access to arts experiences varies by income and locality, creating an experience divide. We present a low-barrier mixed-reality (MR) 3D-painting system on Meta Quest 3 that renders strokes in the user's surrounding space in passthrough MR. We conducted two complementary studies: an in-the-wild festival study with local children (MR 3D free creation) and a controlled study with university students enabling within-participant comparisons across free 2D drawing, free 3D drawing (with depth), and prompted 3D drawing. After each condition, participants rated immersion, accomplishment, perceived creativity, self-efficacy, intention to continue, and usability (plus two free-creation items) on 5-point Likert scales, and collaboration was assessed in paired tasks. We report condition-wise medians and within-participant median differences. Results suggest that MR can deliver meaningful creative experiences with minimal setup in everyday spaces and inform how task framing supports depth-oriented 3D creation.
-
Proposal of a Muscle Training Method using EMG Visualization via Machine Learning Reviewed
Oshikawa Y., Ishimaru T., Nokura S., Ikeda S., Aoki K., Ohe K., Takei A., Kawamura R., Sakamoto M.
Proceedings of International Conference on Artificial Life and Robotics 426 - 429 2026
Publishing type：Research paper (scientific journal) Publisher：Proceedings of International Conference on Artificial Life and Robotics
Strength training is essential for maintaining health and building an attractive physique, yet many people struggle to stick with it. One reason for this is that they fail to feel the effects of strength training. To maximize the effects of strength training, mastering proper form is essential. Therefore, I embarked on this research to reduce the number of people who quit strength training by visualizing muscle load in real time during workouts.We are developing a system that uses machine learning to visualize muscle load from user form, enabling muscle load visualization without requiring electromyography. At present, it is possible to estimate muscle load, but the accuracy of this estimation is low. Therefore, we are currently experimenting to improve the accuracy of muscle load estimation.
-
Development of a Crisis-Avoidance Simulator Based on the Boids Model Reviewed
Hidaka T., Sakamoto M., Aoki K.
Proceedings of International Conference on Artificial Life and Robotics 444 - 447 2026
Publishing type：Research paper (scientific journal) Publisher：Proceedings of International Conference on Artificial Life and Robotics
This study presents the development of a crisis-avoidance simulator based on Reynolds' Boids model, designed to simulate crowd escape behavior in two-dimensional environments during attack scenarios. The simulator incorporated structural elements such as openings, wall-induced reflection and repulsion, and line-of-sight occlusion. It featured a single attacker who pursued the nearest visible agent (boid) and multiple agents who attempted to flee. Simulation experiments under various room configurations revealed that spatial structures and inter-agent interactions significantly influenced escape dynamics. These findings suggest that the proposed simulator could serve as a valuable tool for optimizing evacuation route design and emergency behavior planning.
-
Pongpiyapaiboon S., Aoki K., Hashiguchi M., Akashi R., Kishima Y., Tanaka H.
Plant Phenome Journal 8 ( 1 ) 2025.12
Publishing type：Research paper (scientific journal) Publisher：Plant Phenome Journal
High-throughput and noninvasive phenotyping methods are promising technology for improving efficiency in plant research and breeding. In this study, we evaluated the performance of a digital phenotyping system (DPS) based on three-dimensional (3D) model reconstruction for quantifying key growth traits in rice (Oryza sativa). The DPS was used to estimate plant height, biomass, color, leaf morphology, and tiller angle in four rice varieties (Koshihikari, Nipponbare, PL9, and Tachiaoba). The results show high accuracy and correlation between manually measured and DPS-derived traits. Notably, the 3D volume analysis can quantify biomass accumulation and growth dynamics and revealed distinct differences among varieties. The strong correlation between the green-red normalized difference index (a red-green-blue-based index) and soil plant analysis development also demonstrated the viability of the system in monitoring leaf color without using a multispectral instrument. The analysis also captured growth patterns over time, including canopy development and senescence, which are often challenging to quantify through manual measurements alone. Furthermore, the tiller angle estimation derived from DPS provided an alternative method to plant architecture evaluation, demonstrating its potential for use in breeding programs aimed to optimize canopy structure. These findings establish DPS as a reliable and scalable tool for a digital phenotyping platform that enables comprehensive trait analysis with reduced labor and increased precision and the capability to continuously monitor plant growth and biomass accumulation. This study shows the potential of this novel digital tool for automating manual measurements, which can increase efficiency and expedite research and breeding in rice and other crops.
DOI： 10.1002/ppj2.70054
MISC 【 display / non-display 】
-
オンデマンド授業のための高画質コンテンツの作成と評価－高等教育における試み－
青木 謙二
メディア教育研究 Vol.1, No.2, p.91-101 2005
Language：Japanese Publishing type：Article, review, commentary, editorial, etc. (bulletin of university, research institution)
Presentations 【 display / non-display 】
-
An attempt to visualize device movement information using session logs of wireless LAN
Yusei Hayashida, Kenji Aoki
2024.9.19
Event date： 2024.9.19 - 2024.9.20
Presentation type：Oral presentation (general)
-
Analysis of self-diagnosis results using the information security measures self-diagnosis system
Kenji Aoki, Makoto Sonoda, Wataru Kurogi, Masashi Miyamoto, Isamu Hatsukade
2023.12.15
Event date： 2023.12.13 - 2023.12.15
Language：Japanese Presentation type：Oral presentation (general)
-
Analysis of self-diagnosis results using the information security measures self-diagnosis system
Kurogi Wataru, Hatsukade Isamu, Miyamoto Masashi, Aoki Kenji, Sonoda Makoto
Proceedings of the Annual Conference of Academic Exchange for Information Environment and Strategy 2023.12.6 Academic eXcange for Information Environment and Strategy
Event date： 2023.12.6
Language：Japanese Presentation type：Oral presentation (general)
-
CNNによる渦巻き錯視の認識
東郷拓弥、坂本眞人、青木謙二
2022年度 電気・情報関係学会九州支部連合大会 2022.9.17
Event date： 2022.9.16 - 2022.9.17
Language：Japanese Presentation type：Oral presentation (general)
-
コロナ禍におけるオンライン授業の実践
青木 謙二
第2９回 国公立大学情報システム研究会総会 2021.3.5