share it - The Open Access and Research Data Repository of the University Libraries in Saxony-Anhalt
Not a member yet
91636 research outputs found
Sort by
MLUH-E-10_3, Falco peregrinus Tunstall, 1771, eggshell
No full textInventory No.: MLUH-E-10_3, Object: eggshell, Species: Falco peregrinus Tunstall, 1771, Preservation: complete preservation, Locality_loc.: Havelberg (Mark Brandenburg), Locality today: Saxony-Anhalt, Country: GermanyCollector_leg.: G. Schulz, Date: 03/04/1921, Collection_coll.: M. Schönwetter, published in Handbuch der Oologie, Schönwetter, Max: Vol. I, p. 189, Identification by: M. Schönwetter, Aquisition: 1939, Aquired from: G. Schulz, ex Collection: ex. Coll. G. Schulz, Additional Information: Gustav Schulz was based in Neustadt an der Dosse; Found in 16m height on a pin
MLUH-E-256_1, Buteo rufinus rufinus (Cretzschmar, 1829), eggshell
No full textInventory No.: MLUH-E-256_1, Object: eggshell, Species: Buteo rufinus rufinus (Cretzschmar, 1829), Preservation: good preservation, Locality_loc.: Smyrna, Locality today: Izmir, Country: TurkeyCollector_leg.: Dr. T. J. Krüper, Date: 20/04/1904, Collection_coll.: M. Schönwetter, published in Handbuch der Oologie, Schönwetter, Max: Vol. I, p. 164, Identification by: M. Schönwetter, Aquisition: 1927, Aquired from: Dr. Henrici, ex Collection: ex. Coll. Dr. Henrici, ex. Coll. Dr. T. J. Krüper, Additional Information: According to the Catalogue 255 and 256 are one clutch. Label says they ae differen
Association of emergency room admissions and weekdays in musculoskeletal medicine : results from a major trauma centre in Germany
No full textBackground
Seven days is a week. This ancient concept of structuring our everyday lives has survived several millennia. The repetitious cycle of work and rest is still shaping our routine, influencing the occurrence of diseases and necessities in emergency departments.
Methods
We analysed the admissions to a trauma emergency department of a level 3 trauma centre from 2018 to 2024, looking for changes in admission frequency based on the seven days a week.
Results
Data from 53,597 patients were collected, of whom 45.4% (24,336 of 53,597) were women. The mean age was 35.9 ± 25.4 years. A strong association emerged between the day of the week and the number of admitted patients. In particular, Saturday and Sunday had the most admittances, whereas Thursday was the least busy (P < 0.01).
Conclusions
While we cannot present data on the reasons for this increase, it is probable to account for this rise in a different activity profile of the patients in comparison to the work week. Other factors that might influence this are the patients’ obligations during the work week and the availability of medical care limited to ERs on weekends. Independent of all these reasons, these data may help healthcare providers allocate their resources based on patient volume and emergency conditions
MLUH-E-117_1, Falco naumanni Fleischer, JG, 1818, eggshell
No full textInventory No.: MLUH-E-117_1, Object: eggshell, Species: Falco naumanni Fleischer, JG, 1818, Preservation: complete preservation, Locality_loc.: Smyrna, Locality today: Izmir, Country: TurkeyCollector_leg.: Dr. T. J. Krüper, Date: 24/05/1864, Collection_coll.: M. Schönwetter, published in Handbuch der Oologie, Schönwetter, Max: Vol. I, p. 192, Identification by: M. Schönwetter, Aquisition: 1927, Aquired from: Dr. Henrici, ex Collection: ex. Coll. Dr. Henrici; ex. Coll. Dr. T. J. Krüper, Additional Information: signed with a
Effect of Antioxidant Activity of Naringin and CoQ10 Against Acetaminophen-Induced Nephrotoxicity in Male Rats
No full textAcetaminophen overdose is a known cause of nephrotoxicity, primarily through oxidative stress mechanisms. This study evaluated the potential renoprotective effects of the antioxidants naringin and coenzyme Q10 (CoQ10) against acetaminophen-induced kidney damage. Forty-two male rats were divided into six groups (n=7): control, acetaminophen (1g/kg), naringin (100mg/kg), CoQ10 (100mg/kg), acetaminophen + naringin, and acetaminophen + CoQ10. Treatments were administered orally for 60 days. Serum levels of urea, creatinine, calcium, phosphate, and the activities of antioxidant enzymes (SOD, GPx) and malondialdehyde (MDA) were assessed. Acetaminophen administration caused a significant (p<0.05) decrease in SOD, GPx, calcium, and phosphate levels, alongside an increase in MDA, urea, and creatinine levels compared to the control. Co-treatment with both naringin and CoQ10 significantly ameliorated these changes. Notably, naringin exhibited a more potent protective effect, normalizing kidney function parameters and oxidative stress markers to levels comparable with the control group. The findings demonstrate that naringin and CoQ10 confer protection against acetaminophen-induced nephrotoxicity by attenuating oxidative stress. Naringin proved to be a more effective therapeutic agent in this model, suggesting its potential for clinical application in preventing drug-induced kidney injury
Investigation of reward-related processing improvements in dual-task situations
Get PDFEvidence from single-task studies suggests that reward improves cognitive performance, yet its role in dual-tasking (DT) remains unclear. This dissertation addresses four open questions: (1) which DT processes are affected by reward, (2) whether reward effects transfer between tasks, (3) whether reward strategies can be flexibly adapted, and (4) whether reward-related improvements reflect mere preparation. Using the psychological refractory period (PRP) paradigm with reward prospect for task 1 (Study 1) or task 2 (Study 2), results showed that reward improved pre-motoric processes in task 1 and transmitted benefits to task 2. Study 3 applied trial-wise reward cues and varied the cue–target interval (CTI), showing that while longer CTIs enhanced performance, reward effects extended beyond preparation. Together, the findings suggest that reward enhances DT performance, involves inter-task transmission, and cannot be explained by preparation alone
Evidence of heteroepitaxy and solid solutions in lattice matched ternary covalent organic framework systems
Get PDFCovalent organic frameworks (COFs) are crystalline, porous materials with the possibility for broad applications, but their structural diversity remains constrained by simple net topologies, limiting functional versatility. To address this challenge, we developed a strategy for incorporating linkers with normally mismatched geometries, exemplified by a [4-c + 2-c + 3-c] system incorporating pentagonal motifs for 2D tiling. Central to this approach is the derivation of a length ratio parameter, α, which provides a quantitative guide for evaluating the compatibility of linkers in ternary systems. Investigating a model system with close to ideal α, we demonstrate that precise size matching enables the formation of localized solid solutions and heteroepitaxial interfaces, as seen by transmission electron microscopy. These findings showcase a pathway for expanding the structural and functional complexity of COFs, opening new avenues for tailored material design
34. Schweißtechnische Fachtagung : Tagungsband zur gleichnamigen Fachtagung am 15. Mai 2025 in Barleben
No full textLiteraturangabe
Capturing dissent: forensic photography of graffiti in the late German Democratic Republic
Get PDFThrough a document analysis of archival materials, this paper explores visual landscapes of graffiti, produced by the German Democratic Republic’s Ministry for State Security (MfS) photographers in Leipzig from 1980 to 1989. Capturing visual dissent through forensic photography and its subsequent displacement from public view are two entwined territorial practices that appear to concern aesthetics yet are inherently political. Four selected photographs illustrate the main findings: First, a logic of invisibilisation as a means of deterritorialisation, and second, contradictions in MfS photo practices that highlight the contingent character of repairs to the brittle architecture of state sovereignty in the late GDR
Efficient and robust face recognition in the wild
Get PDFFace recognition stands as the superior biometric technique for identity authentication,
finding extensive applications in our daily lives, like access control, finance, entertain-
ment, and public security. Despite the widespread integration of face biometrics, most
current face recognition systems are tailored for environments where accurate control
governs the process of capturing facial images.
In recent years, rapid advancements in face recognition techniques have unfolded
across both academic and industrial sectors. This transformation has been driven
by key factors, including the availability of substantial annotated training datasets,
the rise of convolutional neural network based deep architectures, the affordability and
power of computational resources, and the emergence of refined loss functions. Despite
the considerable strides and achievements, persistent challenges await resolution.
This thesis makes significant contributions to in-the-wild face recognition, partic-
ularly concerning human-robot interaction, from three perspectives: model enhance-
ment, loss function innovation, and network design. By enhancing current face recog-
nition framework capabilities, designing novel loss functions, and carefully developing
network architectures, this thesis aims to navigate the challenges of recognizing faces
within dynamic and uncontrolled environments, where humans and robots interact.
Firstly, we address the complexities of human-robot interaction (HRI), highlighting
the challenges of real-time face recognition. Emphasizing the need for fast process-
ing and high accuracy, we adopt lightweight convolutional neural networks for our
proposed face recognition framework. The integration of the state-of-the-art ArcFace
loss function and the RetinaFace method for face detection, combined with an online
real-time face tracker, empowers our system to adeptly handle challenges such as vary-
ing illumination, different head poses, and occlusions. By merging tracking data with
recognized identities, we enhance the system’s performance in unconstrained settings,
resulting in improved recognition accuracy and processing speed. Evaluations within
our HRI system, "RoSA," showcase significant advancements while also highlighting
areas for further refinement.
Secondly, we explore the transformative role of margin-based softmax loss func-
tions in face recognition. Traditional methods, which rely on a static, single margin,
may not effectively address diverse real-world data. In response, we introduce the
JAMsFace loss function, which offers flexible margin settings based on the class distri-
bution. Harnessing joint adaptive margins in both angle and cosine spaces, JAMsFace
refines feature discernibility and effectively addresses the challenge of class imbalance.
Comprehensive evaluations across various datasets validate the efficacy of JAMsFace, signaling a shift towards more adaptive face recognition methodologies.
Finally, we present RobFaceNet, a network specifically designed for face recogni-
tion. Balancing computational efficiency with accuracy, RobFaceNet employs a multi-
feature approach and integrates the modified h-swish activation function. We further
enhance RobFaceNet with an attention-based bottleneck, incorporating either a CA
or SE attention module, to boost its facial feature discernment capabilities. Rigor-
ous evaluations against state-of-the-art face recognition models confirm RobFaceNet’s
superior performance, underscoring the potential of lightweight models in real-world
scenarios.
In conclusion, this thesis encapsulates a comprehensive journey through the complex
landscape of face recognition in dynamic and uncontrolled environments, specifically
within the context of human-robot interactions. Addressing fundamental challenges,
innovating within the scope of loss functions, and devising efficient network designs
underscores a clear roadmap toward achieving more seamless and natural interactions
between humans and robots.Gesichtserkennung gilt als überlegene biometrische Technik zur Identitätsauthentifi-
zierung und findet umfangreiche Anwendungen in unserem täglichen Leben, wie Zu-
gangskontrolle, Finanzen, Unterhaltung und öffentliche Sicherheit. Trotz der weit ver-
breiteten Integration von Gesichtsbiometrie sind die meisten aktuellen Gesichtserken-
nungssysteme für Umgebungen maßgeschneidert, in denen eine genaue Steuerung den
Prozess der Erfassung von Gesichtsbildern bestimmt.
In den letzten Jahren haben rasante Fortschritte in den Techniken zur Gesichts-
erkennung sowohl im akademischen als auch im industriellen Bereich stattgefunden.
Diese Transformation wurde durch Schlüsselfaktoren vorangetrieben, darunter die Ver-
fügbarkeit umfangreicher annotierter Trainingsdatensätze, der Aufstieg von tiefen Ar-
chitekturen auf der Grundlage von Convolutional Neural Networks, die Erschwing-
lichkeit und Leistungsfähigkeit von Rechenressourcen und das Auftreten raffinierter
Verlustfunktionen. Trotz der erheblichen Fortschritte und Erfolge warten weiterhin
anhaltende Herausforderungen auf Lösungen.
Diese Dissertation trägt zur Gesichtserkennung unter realen Bedingungen bei, ins-
besondere im Zusammenhang mit der Interaktion zwischen Mensch und Roboter, aus
drei Perspektiven: der Verbesserung von Modellen, der Innovation von Verlustfunk-
tionen und dem Design von Netzwerken. Durch die Verbesserung der Fähigkeiten des
aktuellen Gesichtserkennungsrahmens, die Entwicklung innovativer Verlustfunktionen
und die sorgfältige Gestaltung von Netzwerkarchitekturen zielt diese Arbeit darauf ab,
die Herausforderungen bei der Erkennung von Gesichtern in dynamischen und unkon-
trollierten Umgebungen zu bewältigen, in denen Menschen und Roboter interagieren.
Erstens behandeln wir die Komplexitäten der Mensch-Roboter-Interaktion (HRI)
und betonen die Herausforderungen der Echtzeit-Gesichtserkennung. Mit Schwerpunkt
auf schneller Verarbeitung und hoher Genauigkeit verwenden wir leichte Convolutional
Neural Networks für unseren vorgeschlagenen Gesichtserkennungsrahmen. Die Inte-
gration der hochmodernen ArcFace-Verlustfunktion und der RetinaFace-Methode zur
Gesichtserkennung, kombiniert mit einem online Echtzeit-Gesichts-Tracker, ermöglicht
es unserem System, Herausforderungen wie unterschiedliche Beleuchtung, verschiedene
Kopfpositionen und Verdeckungen geschickt zu bewältigen. Durch die Zusammenfüh-
rung von Tracking-Daten mit erkannten Identitäten verbessern wir die Leistung des
Systems in nicht eingeschränkten Umgebungen und erzielen eine verbesserte Erken-
nungsgenauigkeit und Verarbeitungsgeschwindigkeit. Bewertungen innerhalb unseres
HRI-Systems, "RoSA", zeigen signifikante Fortschritte und weisen gleichzeitig Bereiche
für weitere Verbesserungen auf. Zweitens untersuchen wir die transformative Rolle von margenbasierten Softmax-
Verlustfunktionen in der Gesichtserkennung. Traditionelle Methoden, die auf einem
statischen, einzelnen Margin basieren, können vielfältige realweltliche Daten mögli-
cherweise nicht effektiv bewältigen. Als Reaktion darauf führen wir die JAMsFace
Verlustfunktion ein, die flexible Margin-Einstellungen basierend auf der Klassenvertei-
lung bietet. Durch die Nutzung gemeinsamer anpassbarer Margen sowohl im Winkel-
als auch im Cosinus-Raum verfeinert JAMsFace die Merkmalsunterscheidbarkeit und
bewältigt effektiv die Herausforderung der Klassenungleichgewicht. Umfassende Be-
wertungen in verschiedenen Datensätzen bestätigen die Wirksamkeit von JAMsFace,
was auf eine Verschiebung hin zu adaptiveren Methoden in der Gesichtserkennung
hinweist.
Schließlich präsentieren wir RobFaceNet, ein speziell für die Gesichtserkennung ent-
wickeltes Netzwerk. RobFaceNet balanciert Recheneffizienz und Genauigkeit aus und
verwendet einen multi-feature Ansatz und integriert die modifizierte h-swish Akti-
vierungsfunktion. Wir verbessern RobFaceNet weiter mit einem aufmerksamkeitsba-
sierten Engpass, der entweder ein CA- oder SE-Aufmerksamkeitsmodul enthält, um
seine Fähigkeiten zur Merkmalsunterscheidung im Gesicht zu steigern. Rigorose Be-
wertungen im Vergleich zu modernsten Gesichtserkennungsmodellen bestätigen die
überragende Leistung von RobFaceNet, was das Potenzial von leichten Modellen in
realen Szenarien unterstreicht.
Zusammenfassend fasst diese Dissertation eine umfassende Reise durch das komplexe
Gebiet der Gesichtserkennung in dynamischen und unkontrollierten Umgebungen zu-
sammen, insbesondere im Kontext der Interaktion zwischen Mensch und Roboter. Die
Bewältigung grundlegender Herausforderungen, die Innovation im Rahmen von Ver-
lustfunktionen und die Entwicklung effizienter Netzwerke unterstreichen einen klaren
Weg zur Erreichung nahtloserer und natürlicherer Interaktionen zwischen Menschen
und Robotern.Literaturverzeichnis: Seite 123-14