81690 research outputs found
Sort by
道路橋の建設年を推定するための改良された衛星画像解析手法の開発
No full textEffective management of bridge infrastructure is crucial for ensuring safety and longevity with limited resources. This study proposes an improved method for estimating construction years of road bridges using remote sensing and statistical analysis techniques. The use of bridge age and damage ratings is a cost-effective approach for evaluating the deterioration rates and predicting the future physical conditions of road bridges. Bridge authorities and organizations can benefit from this method, enabling them to prioritize maintenance and repairs effectively. The purpose of this study is to develop a clear and objective method for estimating construction years of road bridges, addressing the limitations of previous research.
The study aims to provide a reliable tool for bridge authorities and organizations to manage bridge infrastructure effectively. The method’s accuracy and reliability are crucial for informing maintenance and repair decisions. The study utilizes the remote sensing data, the Normalized Difference Water Index 2 (NDWI_ 2), from Landsat 5 Thematic Mapper and Landsat 7 Enhanced Thematic Mapper Plus. Google Earth Engine (GEE) processes and analyzes the NDWI_2 data. The Sequential T-test Analysis of Regime Shift (STARS) method identifies the estimated construction year from the analyzed NDWI_2 data. The NDWI_2 is a function of near-infrared (NIR), and shortwave infrared 2 (SWIR2) bands, mathematically expressed as: NDWI_2 = (NIR - SWIR2) / (NIR + SWIR2).
The performance of the method was tested in Nago City, Japan, Cambodia, and Zambia. The results show promising accuracy, with correctly estimated construction years for road bridges in Nago City, Cambodia, and Zambia. The regression analysis yielded R² values of 0.33, 0.06, and 0.64 for the bridges in Nago City, Japan, Cambodia, and Zambia, respectively. The overall absolute mean errors determined were 2.69, 4.93, and 13.79 for the bridges in Nago City, Cambodia, and Zambia, respectively. The estimated number of correctly identified bridges with estimated construction years in Nago City were 39 out of 44 road bridges, 423 out of 649 road bridges in Cambodia, and 14 out of 27 in Zambia. The study’s findings demonstrate the potential of remote sensing and statistical analysis techniques in accurately estimating the construction years of road bridges. The relatively low R² values may be attributed to insufficient reflectance due to atmospheric absorption and scattering effects. In addition, differences between the planned year of construction recorded in the database and the actual year of construction may have contributed to the discrepancy between the estimated construction years and actual construction years, resulting in lower R² values. Such discrepancies may have been caused by construction delays. The limited number of road bridges tested in Zambia was due to the insufficient availability of road bridge data with known years of construction.
The improved method offers significant time and cost savings, particularly for road bridges with unknown construction years. It also serves as a valuable planning tool for bridge maintenance and improvement. The methodology holds promise beyond road bridges, extending to the determination of construction years for other structures. The study’s contributions include the development of an objective and clear method for estimating the construction years, the demonstration of the method’s accuracy and reliability, and the potential for broader applications in infrastructure management.
The method’s objectivity and clarity make it a valuable tool for bridge authorities and organizations. Future studies can build on this research to improve the accuracy and reliability of construction year estimates. The study’s findings have implications for infrastructure management, highlighting the importance of accurate construction year estimates for effective maintenance and repair decisions
動的特徴選択による最良分子を推定する逐次実験のための多腕バンディットアルゴリズムの研究
No full textMy research topic is the acceleration of finding the target molecule whose property (such as drug efficacy, reaction yield, enantioselectivity, wavelength and solvation affinity etc.) is the highest/lowest using a reinforcement learning, linear bandit (Optimism in the Face of Uncertainty Linear Bandit-OFUL bandit) framework. Precisely, the objective of the research is to find the molecule having the highest/lowest quantity of the desired property such as hydration free energy, and reaction enantioselectivity, solvation affinity etc., by fewer experiments as much as possible via bridging chemistry and information science. Usually, chemist do the this job by sequentially testing the molecules which is very time consuming, costly in terms of both money and labor, sometimes it is quite impossible if the set of candidate molecule is very large. To solve this issue sequential optimization is one of the promising approaches in identifying the optimal candidate(s) (molecules, reactants, drugs etc.) with desired properties (such as drug efficacy, reaction yield, enantioselectivity, wavelength and solvation affinity etc.) from a large set of potential candidates, while minimizing the number of experiments is required.
Molecular properties are predicted using fragment-based features (In Silico Design and data Analysis- ISIDA) that are derived from molecular structures represented by Simplified Molecular-Input Line-Entry System (SMILES) codes. However, the high dimensionality of the feature space (e.g., molecular fragment
descriptors, Morgan fingerprint descriptors) makes it often difficult to utilize the relevant features during the process of updating the set of candidates to be examined. In this thesis work, a new sequential optimization algorithm is developed for molecular problems based on a reinforcement learning, multi-armed linear bandit framework, and on-line, dynamic feature selections in which relevant molecular descriptors are updated along with the experiments.
A stopping condition is also designed aimed to guarantee the reliability of the chosen candidate from the pool of data. The developed algorithm was examined by comparing with Bayesian optimization, using two synthetic datasets (termed as Syn-I which is a smaller dataset and Syn-II which is relatively larger dataset) and three real datasets such as a) hydration free energy of molecules, b) energy barrier difference between enantiomer products in chemical reaction and c) photophysical property of molecules e.g., E isomer π−π∗Transition wavelength. It is found that the dynamic feature selection in representing the desired properties along the experiments provides a better performance (e.g., time required to find the best candidate and stop the experiment) as the overall trend, and that our multi-armed linear bandit approach with dynamic feature selection scheme outperforms the standard Bayesian optimization (BO) with fixed feature variables. The comparison of our algorithm to BO with dynamic feature selection is also addressed
