Kyutacar : Kyushu Institute of Technology Academic Repository
Not a member yet
9746 research outputs found
Sort by
Environment-specific knowledge acquisition system for home service robots via brain-inspired memory and commonsense knowledge
Full text linkHome-service robots are expected to enhance the quality of daily life. Two types of knowledge, commonsense and environment-specific knowledge, are required for home-service robots. Current robots can handle audio-visual-based common knowledge, task planning, and action generation. Additionally, several systems that handle environment-specific knowledge have been developed. Previously proposed brain-inspired models can integrate visual and location information to represent episodes; however, it is difficult to obtain novel environment-specific knowledge. In this study, we propose a system that efficiently acquires novel environment-specific knowledge by combining environment-specific knowledge stored in a brain-inspired memory system with commonsense knowledge inferred by large language models (LLMs). We verified the performance of common-sense retrieval from LLMs and evaluated the effectiveness of combining environment-specific knowledge and commonsense knowledge in the home environment.journal articl
PCA-Based Classification of Traditional Japanese Stencil Images
No full textIn this study, we propose an efficient method for classifying images of Ise-Katagami, a traditional Japanese paper stencil used for dyeing kimono fabric, by leveraging principal component analysis (PCA). Ise-Katagami has been produced in the Ise region of Japan for centuries, and a vast number of digital images have been preserved. However, issues related to copyright and ownership have hindered the application of conventional deep learning or transfer learning methods. To address this challenge, we introduce a simple yet effective classification approach based solely on PCA. Unlike deep learning models, our method does not require pre-training, eliminating the need for a large dataset while enabling rapid classification. Furthermore, compared to existing techniques that rely on image distance calculations, our approach offers both higher efficiency and improved classification accuracy.journal articl
Experimental Study on Permeability of Highway Embankments Subjected to Post-Earthquake Rainfall
Full text linkIn this study, the effects of rainfall after an earthquake on the infiltration characteristics and stability of highway embankment structures were experimentally evaluated. A scale model of a highway embankment was used in the test, and nine cases were set with the initial density of the embankment (degree of compaction, Dc) and the degree of damage caused by seismic motion (settlement of crest, ds) as variables. The rainfall intensity was set at 40 mm/h, and data such as water pressure head, hw during rainfall, and the time of occurrence of slope deformation, tf, were collected. As a result of the test, in the case of high-density embankment with Dc = 90%, the number of cracks increased as ds increased due to vibration, and the time when hw began to rise was shortened. On the other hand, in the low-density embankment with Dc = 85%, the density of the embankment increased after the earthquake, and the permeability coefficient decreased, which suppressed rainfall infiltration. The relationship between the number of cracks, nc, after the shaking and the rate of increase in hw per unit time, dhw/dt, showed that dhw/dt also increased as nc increased.journal articl
A novel soil stress estimation method of wheel-soil interaction using photoelasticity
No full textThis paper proposes a new approach to understanding the wheel-soil interaction, which is an indirect estimation method of soil stress distributions beneath a traveling wheel soil using a photoelastic method. Thus far, several photoelastic methods have been discussed for the wheel-soil terramechanics, but it is difficult for the previous configuration to simulate the dynamic behaviors of natural soil, e.g., compaction, failure, or wheel ruts. Accordingly, achieving both the stress visualization and the dynamic behaviors of soil is a significant challenge to make the photoelastic method more practical. To cope with this challenging issue, we have developed a novel experimental setup consisting of a photoelastic wheel (top layer), soil (middle layer), and a photoelastic plate (bottom layer). By vertically sandwiching the soil between the photoelastic wheel and plate, the soil stresses can be indirectly estimated to satisfy the boundary stress conditions. To achieve this approach, we conducted calibration tests of the photoelastic wheel and plate, and then identified the force vector and contact patch corresponding to the visualized stresses. In this paper, we demonstrate that it is possible to indirectly estimate how the stress propagates and attenuates in the soil by the proposed method.journal articl
Suitable Aperture Shape of Fish Nest Blocks for Rest of Plecoglossus Altivelis, Opsariichthys Platypus and Nipponocypris Temminckii
Full text link魚がすめる川の条件の一つに避難場所の確保があり，その一手法に魚巣ブロックの設置が挙げられる．ところが，魚巣ブロックの推奨規格は未解明である．本研究では魚巣ブロックの開口部形状を，全面，上流側，下流側，水面側および底面側開放の5種類に変化させるとともに，主流の体長倍流速を3，6，9(1/s)に変化させてアユ，オイカワ，カワムツの挙動を観察した．その結果，魚巣ブロックの開口部下辺が主流底面と接する場合に全魚種で利用回数が高く，この場合オイカワおよびカワムツは速やかに進入する．アユは警戒心が強いため進入までに時間を要する．また，全面開放および底面側開放では全魚種で利用回数が増加する．One of the conditions for a fish-friendly river is the provision of shelter, and one method of securing shelter for fish is the installation of fish nest blocks. Various studies have been conducted on fish nest blocks, and it has been found that fish nest blocks are useful as resting places for fish when they become fatigued and as shelter from external factors. In this study, the effects of aperture shapes (full open, upstreamopen, downstreamopen, top open, and bottom open) and velocity divided by body length (3, 6 and 9 (1/s)) on the resting characteristics of Plecoglossus altivelis, Opsariichthys platypus and Nipponocypris temminckii were examined. As a result, in the case where the aperture was secured on the bottomside, it was found that fishes can enter the fish nest block easy, and quickly. Spending time in the full open and bottom open fish nest blocks is longer than that of another one.journal articl
Study On Optimum Geometry of Strip Roughness in Open-Channel Flows for Anguilla Japonica Migration
Full text link近年，ニホンウナギの個体数が激減している．その一因はダムや堰による遡上阻害であるため，魚道が併設される．ニホンウナギのような匍匐遊泳を行う底生魚が魚道を遡上するには，桟粗度のような遡上反力を支持する物体の設置が有効である．本研究では開水路底面に設置する桟粗度の断面形状を4種類，また，全長倍流速を1～4(1/s)に変化させ，平均全長200mmのニホンウナギの遡上に適した桟粗度の断面形状を探索した．その結果，横断方向に20mmおきに幅20mm，深さ5mmの凹部をくり抜いた辺長10mmの正方形桟粗度を，流下方向にニホンウナギの全長50%相当の間隔で設置する状態が最適と判明した．なぜならこの状況では，ニホンウナギは桟粗度天端および凹凸部の直角壁に躯幹を接触させて摩擦力を得られるため，流出しにくいからである．In recent years, Japanese eel population has been declining significantly, which is thought to be due in part to the discontinuity of the water level caused by dams and weirs. In order to remedy this situation, fishways are installed in river crossing structures. Fishway is also expected to support the upstream reaction force and improve fish migration rate by installing pier roughness. In this study, the wooden cleats installed on the bottom of an open channel were varied into four types of roughness shapes, and the total length times velocity was varied from 1-4 (1/s) to investigate the suitable cleat shape for migration of Japanese eels. The results showed that, under the experimental conditions, Japanese eels could easily adhere their trunks to cleats with a square cross section of 10 mm in height and 10 mm in width in downstream direction and with 20 mm wide and 5 mm deep recesses at 20 mm intervals in crosswise direction. Japanese eels were found to adhere to cleat roughness easily and to use it as a resting place and an object to support upstream reaction force.journal articl
自己教師あり学習に基づく事前学習モデルを用いたマルチインスタンス学習による脳血種マーカーの分類
Full text linkMultiple Instance Learning (MIL)は弱教師あり学習の1つであり，医療画像をはじめとして様々な分野に応用されている．MILではインスタンスの集合をバッグと定義し，インスタンスのラベルは未知のままバッグのラベルのみを参照することで学習を行う．我々の研究室では脳血種CT画像のマーカーの分類に取り組んでおり，このタスクにMILを適用したところ分類性能が低く，疑似相関に陥るという問題があった．そこで本研究では，自己教師あり学習に基づく事前学習モデルを構築し，ダウンストリームタスクとしてMILによるマーカーの分類を行う手法を提案する．提案手法により，ベースライン手法と比較してaccuracyにおいて6.5%，F1-scoreにおいて11.2%の性能向上を示し，血種領域をより効果的に捉え疑似相関問題に対処できることを確認した．conference pape
多次元センサデータ処理のためのTransformerを用いた自己教師あり学習手法
Full text linkセンサ信号を入力として, 人間行動認識を行う深層学習アルゴリズムを開発した. ここでは自然言語で用いられるTransformerに基づいた事前学習言語モデルを構築して, その事前学習言語モデルを用いて, 下流タスクである人間行動認識タスクを解く形を追求する. VanillaのTransformerでもこれは可能であるが, ここでは, 線形層によるn次元数値データの埋め込み、ビン化処理、出力層の線形処理層という３つの要素を特色とするｎ次元数値処理トランスフォーマーを提案する。5種類のデータセットに対して、このモデルの効果を確かめた. VanillaのTransformerと比較して, 精度で10%～15%程度, 向上させることができた.conference pape
小型水槽内における複数魚の3次元位置推定・追跡手法の提案およびラスボラ・エスペイの群れ形成の効果検証
Full text link九州工業大学博士（工学）九州⼯業⼤学博⼠学位論⽂ 学位記番号：生工博甲第508号 学位授与年⽉⽇: 令和7年3⽉25⽇令和6年度doctoral thesi
バイオセンシングおよびバイオイメージングアプリケーションのための近赤外高感度プローブの合理的設計
Full text link九州工業大学博士（学術）1 INTRODUCTION| 2 INSTRUMENTATION AND CHARACTERIZATION| 3 SQUARAINE DYE-PEPTIDE CONJUGATE FOR SENSITIVE TRYPSIN DETECTION| 4 WAVELENGTH-TUNABLE SQUARAINE DYES AND THEIR
APPLICATION IN ELASTASE DETECTION| 5 NIR SQUARAINE DYE-ANTIBODY CONJUGATE FOR LONGTERM IN VIVO FLUORESCENCE BIOIMAGING| 6 CONCLUSION AND FUTURE PROSPECTSThis thesis presents the design, synthesis, and application of advanced squaraine-based fluorescent probes for enzyme detection and long-term in vivo bioimaging. By leveraging the unique photophysical properties of squaraine dyes, including their strong absorption in the near-infrared (NIR) region and tunable fluorescence behavior, this work addresses critical challenges in biological imaging and molecular sensing, such as sensitivity, specificity, and deep tissue penetration.
The study begins with the development of squaraine dye-peptide conjugates tailored for the detection of trypsin, an enzyme with significant diagnostic and therapeutic implications. A novel dye-peptide probe (SQ-3 PC) was designed to exploit aggregation-induced quenching (AIQ) and Förster resonance energy transfer (FRET) mechanisms, enabling fluorescence ON/OFF sensing. This probe demonstrated exceptional sensitivity with a detection limit of 1.07 nM, providing a robust tool for enzyme-specific monitoring with potential applications in cancer diagnostics and proteomics research. It further extends to squaraine dyes with tunable optical properties for elastase detection. Structural modifications were employed to enhance their performance in aqueous environments, enabling precise and selective enzymatic sensing.
The study highlights the adaptability of squaraine dyes to different enzymatic targets,showcasing their versatility and potential for broader diagnostic applications.
In the context of bioimaging, the thesis explores the development of a squaraine dye-antibody conjugate for long-term NIR imaging. This conjugate exhibited superior photostability, deep tissue penetration, and minimal background interference, making it ideal for sustained in vivo imaging. These advancements position squaraine-based probes as powerful tools for visualizing dynamic biological processes, tracking disease progression, and monitoring therapeutic efficacy.
Collectively, this work introduces a general and adaptable platform for designing advanced fluorescent probes that integrate AIQ, FRET, and NIR fluorescence. By addressing current limitations in biological imaging, these findings pave the way for the next generation of enzyme-specific biosensors and theranostic agents, advancing both fundamental research and translational applications in medicine.九州⼯業⼤学博⼠学位論⽂ 学位記番号：生工博甲第510号 学位授与年⽉⽇: 令和7年3⽉25⽇令和6年度doctoral thesi