Selective laser melting (SLM) is a method of additive manufacturing that has become increasingly popular in recent years for fabricating complex components, especially in the medical and aerospace industries. By fabricating components in a layerwise fashion, SLM provides users the freedom to design components based on their desired functionality rather than their manufacturability. The current state-of-the-art for SLM is limited though, as defects induced by the SLM process have proven to greatly alter the material properties of fabricated parts. In addition, traditional post-process nondestructive inspection methods have experienced significant difficulty in accurately detecting these process-induced defects. Therefore, the objective of this study is to investigate methods of processing and analysis for optical in-situ monitoring data recorded during SLM fabrication of six test samples. Four of the samples were designed with seeded (i.e., intentional) defects located at their center to serve as a reference defect signatures in the resulting in-situ data. An off-axis optical tomography (OT) sensor was used to capture near-infrared (NIR) melt pool emissions during the fabrication of each layer. Image analysis was subsequently performed using a custom squared difference (SD) operator to enhance defect signatures in the OT data. Results from the SD operator were then used to perform k-means clustering to partition the data into k relevant clusters, where the optimal number of k clusters for each image is employed as metric for detecting the onset of defects in the samples. By employing OT image data from samples containing seeded intentional defects, the k-means clustering approach was investigated as a method of defect detection for the in-situ OT images. Results showed that the SD operator is capable of elucidating anomalous signatures in the in-situ data. However, variations within the SD distributions ultimately limited detection capabilities as the output from k-means clustering was unable to accurately distinguish the seeded defects from the fused regions of material.

From 2019 to 2020, greenhouse and field research was conducted at North Dakota State University to investigate the canopy temperature response of waterhemp (Amaranthus rudis), kochia (Kochia scoparia), common ragweed (Ambrosia artemisiifolia), horseweed (Conyza canadensis), Palmer amaranth (Amaranthus palmeri), and red root pigweed (Amaranthus retroflexus) after glyphosate application to identify glyphosate resistance. In these experiments, thermal images were captured of randomized glyphosate resistant populations and glyphosate susceptible populations of each weed species. The weed canopies' thermal values were extracted and submitted to statistical testing and various classifiers in an attempt to discriminate between resistant and susceptible populations. Glyphosate resistant horseweed, when collected within greenhouse conditions, was the only biotype reliably classified using significantly cooler temperature signatures than its susceptible counterpart. For field conditions, image based machine learning classifiers using thermal data were outperformed by classifiers made using additional multispectral data, suggesting thermal is not a reliable predictor of glyphosate resistance.

With the increasing demand of intelligent video surveillance systems, person re-identification (re-ID) plays an important role in intelligent video analysis, which aims at matching person images across non-overlapping camera views. It has gained increasing attention in computer vision community. With the advanced deep neural networks, existing methods have achieved promising performance on the widely-used re-ID benchmarks, even outperform the human-level rank-1 matching accuracy. However, most of the research efforts are conducted on the closed-world settings, with large-scale well annotated training data and all the person images are from the same visible modality. As a prerequisite in practical video surveillance application, there is still a large gap between the closed-world research-oriented setting and the practical open-world settings. In this thesis, we try to narrow the gap by studying three important issues in open-world person re-identification, including 1) unsupervised learning with large-scale unlabelled training data; 2) learning robust re-ID model with label corrupted training data and 3) cross-modality visible-thermal person re-identification with multi-modality data. For unsupervised learning with unlabelled training data, we mainly focus on video-based person re-identification, since the video data is usually easily obtained by tracking algorithms and the video sequence provides rich weakly labelled samples by assuming the image frames within the tracked sequence belonging to the same person identity. Following the cross-camera label estimation approach, we formulate the cross-camera label estimation as a one-to-one graph matching problem, and then propose a novel dynamic graph matching framework to estimate cross-camera labels. However, in a practical wild scenario, the unlabelled training data usually cannot satisfy the one-to-one matching constraint, which would result in a large proportion of false positives. To address this issue, we further propose a novel robust anchor embedding method for unsupervised video re-ID. In the proposed method, some anchor sequences are firstly selected to initialize the CNN feature representation. Then a robust anchor embedding method is proposed to measure the relationship between the unlabelled sequences and anchor sequences, which considers both the scalability and efficiency. After that, a top-{dollar}k{dollar} counts label prediction strategy is proposed to predict the labels of unlabelled sequences. With the newly estimated sequences, the CNN representation could be further updated. For robust re-ID model learning with label corrupted training data, we propose a two-stage learning method to handle the label noise. Rather than simply filtering the falsely annotated samples, we propose a joint learning method by simultaneously refining the falsely annotated labels and optimizing the neural networks. To address the limited training samples for each identity, we further propose a novel hard-aware instance re-weighting strategy to fine-tune the learned model, which assigns larger weights to hard samples with correct labels. For cross-modality visible-thermal person re-identification, it addresses an important issue in night-time surveillance applications by matching person images across different modalities. We propose a dual-path network to learn the cross-modality feature representations, which learns the multi-modality sharable feature representations by simultaneously considering the modality discrepancy and commonness. To guide the feature representation learning process, we propose a dual-constrained top-ranking loss, which contains both cross-modality and intra-modality top-ranking constraints to reduce the large cross-modality and intra-modality variations. Besides the open-world person re-identification, we have also studied the unsupervised embedding learning problem for general image classification and retrieval. Motivated by supervised embedding learning, we propose a data augmentation invariant and instance spread-out feature. To learn the feature embedding, we propose a instance feature-based softmax embedding, which optimizes the embedding directly on top of the real-time instance features. It achieves much faster learning speed and better accuracy than existing methods. In short, the major contributions of this thesis are summarized as follows. l A dynamic graph matching framework is proposed to estimate cross-camera labels for unsupervised video-based person re-identification. l A robust anchor embedding method with top-{dollar}k{dollar} counts label prediction is proposed to efficiently estimate the cross-camera labels for unsupervised video-based person re-identification under wild settings. l A two-stage PurifyNet is introduced to handle the label noise problem in person re-identification, which jointly refines the falsely annotated labels and mines hard samples with correct labels. l A dual-constrained top-ranking loss with a dual-path network is proposed for cross-modality visible-thermal person re-identification, which simultaneously addresses the cross-modality and intra-modality variations. l A data augmentation invariant and instance spread-out feature is proposed for unsupervised embedding learning, which directly optimizes the learned embedding on top of real-time instance features with softmax function

abstract: Traumatic injuries are the leading cause of death in children under 18, with head trauma being the leading cause of death in children below 5. A large but unknown number of traumatic injuries are non-accidental, i.e. inflicted. The lack of sensitivity and specificity required to diagnose Abusive Head Trauma (AHT) from radiological studies results in putting the children at risk of re-injury and death. Modern Deep Learning techniques can be utilized to detect Abusive Head Trauma using Computer Tomography (CT) scans. Training models using these techniques are only a part of building AI-driven Computer-Aided Diagnostic systems. There are challenges in deploying the models to make them highly available and scalable. The thesis models the domain of Abusive Head Trauma using Deep Learning techniques and builds an AI-driven System at scale using best Software Engineering Practices. It has been done in collaboration with Phoenix Children Hospital (PCH). The thesis breaks down AHT into sub-domains of Medical Knowledge, Data Collection, Data Pre-processing, Image Generation, Image Classification, Building APIs, Containers and Kubernetes. Data Collection and Pre-processing were done at PCH with the help of trauma researchers and radiologists. Experiments are run using Deep Learning models such as DCGAN (for Image Generation), Pretrained 2D and custom 3D CNN classifiers for the classification tasks. The trained models are exposed as APIs using the Flask web framework, contained using Docker and deployed on a Kubernetes cluster. The results are analyzed based on the accuracy of the models, the feasibility of their implementation as APIs and load testing the Kubernetes cluster. They suggest the need for Data Annotation at the Slice level for CT scans and an increase in the Data Collection process. Load Testing reveals the auto-scalability feature of the cluster to serve a high number of requests. / Dissertation/Thesis / Masters Thesis Software Engineering 2020

The phase separation process of a rubber modified epoxy system during cure was analyzed by a model developed on the basis of a thermodynamic description of binary mixture and constitutive equations for nucleation and growth rates. As epoxy resins are cured, rubber molecules are precipitated from the epoxy matrix to a non-equilibrium composition due to the decrease in the configurational entropy and the increase in the viscosity with conversion. If phase separation takes place in a metastable region, this model can monitor the changes of rubber compositions in both phases as well as the changes in the number and size of rubber particles upon conversion of polymerization. The particle size distribution at the completion of phase separation was also calculated. The effect of cure temperature on the final morphologies of a rubber modified epoxy system was discussed. The computed particle size distributions for piperidine and diaminodiphenyl sulfone cured systems showed good agreements with experimentally measured values. Depending on the activation energy for viscous flow of the epoxy matrix relative to that for the polymerization, the particle size distribution may show bimodal or unimodal distribution. The size of rubber rich phase increases to a maximum and then decreases with an increase in cure temperature. However, due to limitations of temperature range to probe in an actual experiment, one may observe only either decreasing or increasing particle size as cure temperature increases. The number of rubber particles per unit volume increases for the DGEBA/DDS/ETBN system as cure temperature increases in the temperature range of 30 °C to 220 °C. Fracture toughness of cured DGEBA/DDS/ETBN system was analyzed in terms of morphologies generated by the temperature variation. Since the volume fraction of rubber particles did not change with cure temperature, the critical stress intensity factor did not vary significantly with cure temperature as expected. However, increases in cure temperature produced smaller but more numerous particles. The critical stress intensity factor normalized by the number density of particles exhibited dependence on the radius of particles to the third power. On the other hand, the critical stress intensity factor normalized by the radius of particles showed a linear dependence with respect to the number density of particles. / Ph. D.

The microtuble-associated protein tau is known to regulate neuronal micro- tubule dynamics and is involved in several neurodegenerative diseases collec- tively called tauopathies. Besides the formation of tau-containing aggregates this group of diseases is characterized by changes on different anatomical lev- els in the nervous system. Morphological changes in the dendritic arbor of neu- rons or subcellular compartments can be investigated with microscopy-based and image informatical methods. Furthermore, the functional processes that constitute these changes can be predicted with bioinformatical methods and based on these predictions investigated with biological experiments. Two different bioinformatical disciplines contribute to the study of neurobio- logical processes. Due to advances in microscopy and imaging coupled to the tremendous advances in computer technology, image informatics techniques and workflows are necessary to analyze the acquired data with greater pre- cision. The classical bioinformatics on the other hand covers the analysis of molecular evolution, phylogeny and the prediction of protein function. This work aims to assist neurobiologists with computational methods in ongo- ing reasearch questions. The development of computer-assisted or fully auto- mated workflows for image analysis has been achieved on different levels. A machine learning algorithm has been trained to determine the density of neu- rons in tissues. Workflows for analysis of morphological changes of dendritic arbors, like process thickness or branching pattern, have been implemented. Existing workflows for dendritic spine analysis have been optimized and the volume and movement behavior of subcellular compartments like ribonucle- oparticles have been analyzed. Image analysis workflows have been adapted for the analysis of molecular distributions after photoactivation. Additionally, techniques from data mining workflows have been adapted to extract and filter trajectories from single molecule tracking approaches to assist the inferrence of biophysical parameters. Sequence data from public available databases have been collected to recon- struct tau and other related sequences in a broad range of species to infer phy- logenetic trees and to perform hidden-Markov-model analysis. Using this ap- proach it has been possible to illuminate the relations in the MAPT/2/4 family and predict putative functional sequence motifs for further bioinformatical or biological investigations.

L’histologie est la méthode de choix pour l’étude ex vivo de la distribution spatiale des molécules qui composent les organes. En particulier, l’histologie permet de mettre en évidence les marqueurs neuropathologiques de la maladie d’Alzheimer ce qui en fait un outil incontournable pour étudier la physiopathologie de la maladie et pour évaluer l’efficacité de candidats médicaments. Classiquement, l’analyse de données histologiques implique de lourdes interventions manuelles, et de ce fait, est souvent limitée à l’analyse d’un nombre restreint de coupe histologiques et à quelques régions d’intérêts. Dans ce travail de thèse, nous proposons une méthode automatique pour l’analyse quantitative de marqueurs histopathologiques en trois dimensions dans le cerveau entier de rongeurs. Les images histologiques deux-dimensionnelles sont d’abord reconstruites en trois dimensions en utilisant l’imagerie photographique de bloc comme référence géométrique et les marqueurs d’intérêts sont segmentés par apprentissage automatique. Deux approches sont proposées pour détecter des différences entre groupes d’animaux: la première est basée sur l’utilisation d’une ontologie anatomique de cerveau qui permet détecter des différences à l’échelle de structures entières et la deuxième approche est basée sur la comparaison voxel-à-voxel afin de détecter des différences locales sans a priori spatial. Cette méthode a été appliquée dans plusieurs études chez des souris modèles de déposition amyloïde afin d’en démontrer l’utilisabilité. / Histology is the gold standard to study the spatial distribution of the molecular building blocks of organs. In humans and in animal models of disease, histology is widely used to highlight neuropathological markers on brain tissue sections. This makes it particularly useful to investigate the pathophysiology of neurodegenerative diseases such as Alzheimer’s disease and to evaluate drug candidates. However, due to tedious manual interventions, quantification of histopathological markers is classically performed on a few tissue sections, thus restricting measurements to limited portions of the brain. Quantitative methods are lacking for whole-brain analysis of cellular and pathological markers. In this work, we propose an automated and scalable method to thoroughly quantify and analyze histopathological markers in 3D in rodent whole brains. Histology images are reconstructed in 3D using block-face photography as a spatial reference and the markers of interest are segmented via supervised machine learning. Two complimentary approaches are proposed to detect differences in histopathological marker load between groups of animals: an ontology-based approach is used to infer difference at the level of brain regions and a voxel-wise approach is used to detect local differences without spatial a priori. Several applications in mouse models of A-beta deposition are described to illustrate 3D histopathology usability to characterize animal models of brain diseases, to evaluate the effect of experimental interventions, to anatomically correlate cellular and pathological markers throughout the entire brain and to validate in vivo imaging techniques.

An image processing method for automatic quality assurance of Ericsson products is developed. The method consists of taking an image of the product, extract the product labels from the image, OCR the product numbers and make a database lookup to match the mounted product with the customer specification. The engineering innovation of the method developed in this report is that the OCR is performed using machine learning techniques. It is shown that machine learning can produce results that are on par or better than baseline OCR methods. The advantage with a machine learning based approach is that the associated neural network can be trained for the specific input images from the Ericsson factory. Imperfections in the image quality and varying type fonts etc. can be handled by properly training the net, a task that would have been very difficult with legacy OCR algorithms where poor OCR results typically need to be mitigated by improving the input image quality rather than changing the algorithm.

abstract: In nature, it is commonly observed that animals and birds perform movement-based thermoregulation activities to regulate their body temperatures. For example, flapping of elephant ears or plumage fluffing in birds. Taking inspiration from nature and to explore the possibilities of such heat transfer enhancements, augmentation of heat transfer rates induced by the vibration of solid and well as novel flexible pinned heatsinks were studied in this research project. Enhancement of natural convection has always been very important in improving the performance of the cooling mechanisms. In this research, flexible heatsinks were developed and they were characterized based on natural convection cooling with moderately vibrating conditions. The vibration of heated surfaces such as motor surfaces, condenser surfaces, robotic arms and exoskeletons led to the motivation of the development of heat sinks having flexible fins with an improved heat transfer capacity. The performance of an inflexible, solid copper pin fin heat sink was considered as the baseline, current industry standard for the thermal performance. It is expected to obtain maximum convective heat transfer at the resonance frequency of the flexible pin fins. Current experimental results with fixed input frequency and varying amplitudes indicate that the vibration provides a moderate improvement in convective heat transfer, however, the flexibility of fins had negligible effects. / Dissertation/Thesis / Masters Thesis Mechanical Engineering 2019

京都大学 / 0048 / 新制・課程博士 / 博士(地球環境学) / 甲第18487号 / 地環博第121号 / 新制||地環||25(附属図書館) / 31365 / 京都大学大学院地球環境学舎環境マネジメント専攻 / (主査)教授 勝見 武, 准教授 田中 周平, 准教授 乾 徹 / 学位規則第4条第1項該当 / Doctor of Global Environmental Studies / Kyoto University / DFAM