Publikationsserver der Ostbayerischen Technischen Hochschule Regensburg
Not a member yet
6172 research outputs found
Sort by
Recent Trends in Edge AI: Efficient Design, Training and Deployment of Machine Learning Models
No full textWith a rising demand for ubiquitous smart systems, processing and interpreting large quantities of data generated on the edge at a high velocity is becoming an increasingly important challenge. Machine learning (ML) models such as Deep Neural Networks (DNNs) are an essential tool of today’s artificial intelligence due to their ability to make accurate predictions given complex tasks and environments. However, Deep Learning is computationally complex and energy intensive. This seems to contradict the characteristics of many edge devices, which have only limited memory, computational resources, and energy budget available. To overcome this challenge, an efficient ML model design is crucial that incorporates available optimization techniques from hardware, software, and methodological perspective to enable energy-efficient deployment and operation on the edge. This work comprehensively summarizes recent techniques for training, optimizing, and deploying ML models targeting edge devices. We discuss different strategies for finding deployable ML models, scalable DNN architectures, neural architecture search, and multi-objective optimization approaches, to enable feasible trade-offs considering available resources and latency. Furthermore, we give insight into DNN compression methods such 182as quantization and pruning. We conclude by investigating different forms of cascaded processing, from simple multi-level approaches to highly branched compute graphs and early-exit DNNs
Abschlussbericht Projekt LeaP - Learning Poses : Posenerkennung mit Neuronalen Netzen
No full textDie Schätzung der 6D-Pose bekannter Objekte findet Anwendungen in der Robotik, der Luft- und Raumfahrtsteuerung und automatisierten Produktionsumgebungen. Es ist nach wie vor gängige Praxis, klassische Bildverarbeitungsmethoden zu verwenden, um eine hohe Präzision zu erreichen. Diese Methoden erfordern jedoch eine manuelle Parametrisierung der verwendeten Erkennungswerkzeuge für jedes einzelne Objekt.
Dies wirft die Frage auf, ob maschinelle Lerntechniken, insb. convolutional neural networks, Netzwerke, so trainiert werden können, dass sie die in der Industrie geforderte Präzision ohne individuelle Programmierung erreichen können.
Darüber sollen in Produktionsumgebungen Umwelteinflüsse wie Lichtverhältnisse minimiert werden.
Es werden verschiedene Netzstrukturen entwickelt und auf ihre Leistung bei diesem Szenario untersucht, mit 1 bis 6 Freiheitsgraden in der Aufgabenstellung und einer oder mehreren Kameras. Zusätzlich wird ein Ansatz mit reinforcement learning entwickelt und untersucht
Polymer embedding of membrane lungs for histological investigations of intra-device clot formation
No full textExtracorporeal membrane oxygenation (ECMO) is an invasive but potentially lifesaving treatment option for severe cardiac or respiratory failure. Despite its beneficial effect, coagulation-related complications, mainly due to clot formation, excessive bleeding and the accumulation of deposits in the membrane lung (ML) remain common, causing higher mortality. In this context, the formation of clots and other deposits in the ML is of particular interest. Previous histological examinations of the polymethylpentene fiber mats inside the ML could only be performed in a top view, prohibiting valid quantification and examination of the multi-layered deposits or fiber mat spanning structures. Our objective was the establishment of a polymer embedding to increase the mechanical stability of the deposits and thus enable cross-sectional microtome cutting through the ML hollow-fibers. Clinically used MLs (PLS, Getinge, Rastatt, Germany) were stabilized with a polymer resin (HistoCURE 8100). Specimens were cut out of the embedded MLs and microtome sections with a thickness of 10 µm were performed. In addition to standard histological staining with hematoxylin-eosin (HE) and Pappenheim (May-Grunwald-Giemsa), fluorescence DNA staining for nucleated cells with 4′,6-diamidino-2-phenylindole (DAPI) and SYTOX™ Green as well as immunohistochemical and immunofluorescence staining for the lysosomal enzyme myeloperoxidase (MPO) and von Willebrand factor (vWF) were established. The protocol provides a method for large volume embedding (400 mL). The cellular and extracellular deposits were securely fixed by the polymer scaffold allowing the examination of clots in MLs in native position which was not possible with conventional paraffin embedding. Multi-layered deposits and fiber mat spanning structures are no longer disrupted during specimen extraction and can now be quantified. Staining with HE, Pappenheim, DAPI, SYTOX™ Green, MPO, and vWF was successfully tested with this protocol. This method may be the foundation for new insights into the complex clotting phenomena observed in ML
Experimental Investigation of Shear-Induced Generation of Respiratory Aerosol: Simultaneous Measurements of Particle Quantities and Wave Topology
No full textDespite the high level of attention on infectious respiratory aerosol during the Covid19 pandemic, little is known about the processes how these particles form inside the respiratory system. Understanding the underlying fluid mechanical processes and their influencing factors would enable the development of drugs to suppress the generation of infectious aerosol. In the proposed work, we focus on the shear-induced mechanism of aerosol generation, which is supposed to occur mostly in the larger airways during coughing. In this process, high air velocities trigger Kelvin-Helmholtz waves in the mucus film, which lines the air vessels. Through a series of instabilities, particles detach from the crest of these waves.
In the proposed work, we investigate the process of shear-induced aerosol generation in idealized experiments where we vary the air-flow characteristics and the properties of the mucus fluid. Our central aim here is to deduct an empirical model of the quantity and size distribution of generated particles depending on the mucus rheology and the local shear flow. Further, we observe the wave topology to better understand the coupling between the air flow and the waves.
In our experimental setup, we measure the quantity of created particles and the emerging waves simultaneously. To ensure controllable conditions, we simplify the complex flow conditions in the airways. We use a rectangular channel with the bottom wall covered in a mucus mimetic. The mucus mimetic fluid is a synthetic hydrogel developed to recreate the viscoelastic properties and low surface tensions of the mucus. Filtered pressurized air is guided through the channel to trigger shear-induced aerosol generation. After passing the mucus mimetic, the air enters into a collection chamber from where particles are sampled continuously by an aerosol spectrometer. To measure wave topology, we use planar laser induced fluorescence. For this, we stain the mucus mimetic with fluorescent dyes and illuminate a line on the surface of the fluid film with a 532 nm laser. A high-resolution camera captures the resulting fluorescent glow of the mucus mimetic. Figure 1 presents exemplary wave topology results from the experiments, employing varying air flow volume rates, mucus mimetic gel properties, and different configurations of the laser and camera. The resulting wave topologies exhibit significant variation. For the conference, we will conduct parameter studies
of the particle quantities and wave topology while varying the mucus mimetic properties and the flow
rate of the air. Additionally, we will present grid projection-based techniques to extend the single-line wave measurements and asses the entire surface of the mucus film
Design, simulation, fabrication, measurement, and analysis of 3D-printed radio frequency conductive structures
No full textAdditive manufacturing has emerged as a disruptive technology in the field of electronics, offering new possibilities for the fabrication of complex, lightweight, and customizable components. Conventional PCB manufacturing relies on multi-step processes and is limited to planar geometries, making it less suitable for rapid prototyping or integration into non-standard surfaces. In contrast, Fused Deposition Modelling (FDM) provides a cost-effective and
accessible approach to directly fabricate functional electronic structures by combining conductive and dielectric materials within a single printing process.
This thesis examines the design, simulation, fabrication, measurement and analysis of 3D-printed conductive transmission lines utilising FDM technology. A conductive filament and a dielectric filament are employed to realise fully integrated microstrip and coaxial structures. Electromagnetic simulations are performed to optimise the geometry for RF performance, and the printed prototypes are experimentally characterised using vector network analyser measurements. The analysis focuses on the transmission coefficients and the reflection coefficients, providing insights into impedance matching, insertion loss, and frequency-dependent behaviour.
These findings confirm that FDM-based additive manufacturing can produce functional RF structures with reliable performance, despite material and fabrication limitations. The work demonstrates the potential of 3D printing as a flexible and low-cost method for prototyping radio frequency components, highlighting future directions that include improved material characterisation, connector integration, and the design of more advanced components, such as
filters and antennas
Antimuslimische Einstellungen junger Menschen in Deutschland: Erkenntnisse aus einer bundesweiten Befragung
No full textIn den letzten Jahrzehnten hat die zunehmende Pluralisierung der Gesellschaft zu intensiven Debatten über kulturelle, religiöse und nationale Identitäten geführt. Studien zeigen eine Zunahme demokratiefeindlicher Einstellungen in der deutschen Gesellschaft. Insbesondere Muslim:innen und Menschen, die als solche wahrgenommen werden, sind von Vorurteilen betroffen, die durch rechtspopulistische Narrative verstärkt werden. Dabei wird auch die Bedeutung der symbolischen Macht nach Bourdieu deutlich. Bestehende Hierarchien und Stereotypen führen zu sozialer Ausgrenzung und Diskriminierung von Muslim:innen. Die symbolische Gewalt manifestiert sich in rassistischen Strukturen und islamfeindlichen Demonstrationen. Dieser Beitrag präsentiert Ergebnisse einer Befragung zu antimuslimischen Einstellungen junger Menschen zwischen 16 und 27 Jahren in Deutschland. Die Ergebnisse zeigen, dass antimuslimische Einstellungen durch intensiven Kontakt mit Muslim:innen verringert werden können, insbesondere durch interethnische Freundschaften. Auf Basis der Kontakthypothese wird betont, dass freiwillige, gleichberechtigte Begegnungen in einem kooperativen Kontext Vorurteile abbauen können. Somit zeigt die Studie die Notwendigkeit auf, interkulturelle Begegnungen und Bildungsmaßnahmen zu fördern, um gegenseitiges Verständnis zu stärken. Zudem wird die Bedeutung von sicheren Räumen und Empowerment für junge Muslim:innen hervorgehoben, um ihre Partizipation und Selbstwirksamkeit zu fördern
Extracting Product Improvement Insights from Social Media Comments Using Machine Learning: a Case Study in the Automotive Industry
No full textThis paper presents a scalable machine learning pipeline for extracting actionable, product-related insights from user-generated social media comments. Leveraging sentence embeddings from SBERT and unsupervised clustering (k-Means and agglomerative), the approach structures informal and noisy comments from Instagram and YouTube into topic groups intended to support thematic analysis. A case study on feedback regarding BMW vehicles, comprising more than 26,000 comments, illustrates how the pipeline can reveal recurring user concerns, such as design critiques, usability issues, and technology-related expectations, even in short and unstructured social media comments. The proposed pipeline operates without labeled data or manual annotation, enabling scalable application and transferability across product categories and industries. By transforming large-scale, unstructured consumer feedback into interpretable themes, the pipeline provides product teams with an efficient and structured basis for data-driven product development and improvement
Comparative validation of surgical phase recognition, instrument keypoint estimation, and instrument instance segmentation in endoscopy: Results of the PhaKIR 2024 challenge
No full textReliable recognition and localization of surgical instruments in endoscopic video recordings are foundational for a wide range of applications in computer- and robot-assisted minimally invasive surgery (RAMIS), including surgical training, skill assessment, and autonomous assistance. However, robust performance under real-world conditions remains a significant challenge. Incorporating surgical context – such as the current procedural phase – has emerged as a promising strategy to improve robustness and interpretability.
To address these challenges, we organized the Surgical Procedure Phase, Keypoint, and Instrument Recognition (PhaKIR) sub-challenge as part of the Endoscopic Vision (EndoVis) challenge at MICCAI 2024. We introduced a novel, multi-center dataset comprising thirteen full-length laparoscopic cholecystectomy videos collected from three distinct medical institutions, with unified annotations for three interrelated tasks: surgical phase recognition, instrument keypoint estimation, and instrument instance segmentation. Unlike existing datasets, ours enables joint investigation of instrument localization and procedural context within the same data while supporting the integration of temporal information across entire procedures.
We report results and findings in accordance with the BIAS guidelines for biomedical image analysis challenges. The PhaKIR sub-challenge advances the field by providing a unique benchmark for developing temporally aware, context-driven methods in RAMIS and offers a high-quality resource to support future research in surgical scene understanding
Developing a smart and scalable tool for histopathological education—PATe 2.0
No full textDigital microscopy plays a crucial role in pathology education, providing scalable and standardized access to learning resources. In response, we present PATe 2.0, a scalable redeveloped web-application of the former PATe system from 2015. PATe 2.0 was developed using an agile, iterative process and built on a microservices architecture to ensure modularity, scalability, and reliability. It integrates a modern web-based user interface optimized for desktop and tablet use and automates key workflows such as whole-slide image uploads and processing. Performance tests demonstrated that PATe 2.0 significantly reduces tile request times compared to PATe, despite handling larger tiles. The platform supports open formats like DICOM and OpenSlide, enhancing its interoperability and adaptability across institutions. PATe 2.0 represents a robust digital microscopy solution in pathology education enhancing usability, performance, and flexibility. Its design enables future integration of research algorithms and highlights it as a pivotal tool for advancing pathology education and research