[00669] Mathematical Solutions of Industrial Applications
Session Time & Room : 3C (Aug.23, 13:20-15:00) @G404
Type : Proposal of Industrial Minisymposium
Abstract : Mathematics plays an important role in modern industry, for instance, as a tool for research & development and as algorithmic parts of products. This session presents success stories of industrial mathematics as a solution to various business challenges. Several domains of industry are considered: automotive, optics, manufacturing, medical imaging and agriculture. The following specific topics are included: 1. Data-driven development in industry, 2. Industrial applications of machine learning, 3. Inverse problems in medical X-ray imaging, 4. Machine learning in industry. Each talk will discuss the motivation, approaches and implementations based on mathematics.
[02716] Prediction of Leaf Area Index of Tomato Plants by Image Processing and Deep Learning
Format : Talk at Waseda University
Author(s) :
Mario Tsukassa Sato (University of Tsukuba)
Aiga Goto (University of Tsukuba)
Nanami Isoda (University of Tsukuba)
Ayu Kaise (University of Tsukuba)
Claus Aranha (University of Tsukuba)
Akira Imakura (University of Tsukuba)
Tetsuya Sakurai (University of Tsukuba)
Naomichi Fujiuchi (Ehime University)
Naoya Fukuda (University of Tsukuba)
Abstract : In this work, we study how to improve image recognition in greenhouse tomato plants using Machine Learning. We propose a method for non-destructively estimating the Leaf Area Index (LAI), as well as a method to improve the segmentation of different parts of the plant. Moreover, we discuss our investigation of three ways to estimate the relevant parts of the segmented image. Additionally, we investigated new methods to improve the task of semantic segmentation of the whole plant.
[02893] Frequency based graph estimation for multivariate time-series
Format : Talk at Waseda University
Author(s) :
Yuuya Takayama (Nikon Corporation)
Abstract : We propose a method to estimate an underlying graph structure from multivariate time-series data. This method is derived from the Dynamic Mode Decomposition (DMD) method so as to recover an exact graph from a solution of the graph wave equation. In our talk, we introduce its applications and discuss uncertainty of an estimated graph based on the mathematical equation.
[02900] Passive Gamma Emission Tomography for Spent Nuclear Fuel
Format : Talk at Waseda University
Author(s) :
Riina Virta (Radiation and Nuclear Safety Authority (STUK), and Helsinki Institute of Physics, University of Helsinki)
Tatiana Alessandra Bubba (University of Bath)
Mikael Moring (Radiation and Nuclear Safety Authority (STUK))
Samuli Siltanen (Department of Mathematics and Statistics of the University of Helsinki)
Tapani Honkamaa (Radiation and Nuclear Safety Authority (STUK))
Peter Dendooven (Helsinki Institute of Physics, University of Helsinki)
Abstract : Spent nuclear fuel needs to be verified prior to geological disposal in a deep underground repository. In Passive Gamma Emission Tomography (PGET), the gamma radiation emitted by the fuel is recorded with highly collimated semiconductor detectors and 2D slice images of activity and attenuation are simultaneously reconstructed to account for the high self-attenuation of the material. General information about the object geometry is used as a prior to regularize the ill-posed inverse problem.
[02901] Boundary estimation of the X-ray tomographic reconstruction using persistent homology
Format : Talk at Waseda University
Author(s) :
Elli Karvonen (University of Helsinki)
Abstract : In some applications, one can only use limited-angle X-ray tomography, which results in a much more difficult reconstruction problem than a full-angle case. Despite an algorithm, all parts of the smooth boundaries of the target object cannot be detected stably. In this talk the new boundary estimation method, which utilizes complex wavelets and persistent homology, is presented. The latest results are shown and discussed.
