Refine
H-BRS Bibliography
- yes (3)
Departments, institutes and facilities
- Institut für Sicherheitsforschung (ISF) (3) (remove)
Document Type
- Doctoral Thesis (3) (remove)
Language
- English (3)
Has Fulltext
- no (3)
Keywords
- Bacteria (1)
- Bakterien (1)
- Chemometrics (1)
- Chemometrie (1)
- Infrared (1)
- Infrarot (1)
- Raman (1)
- Spectroscropy (1)
- Spektroskopie (1)
Microorganisms not only contribute to the spoilage of food but can also cause illnesses through consumption. Consumer concerns and doubts about the shelf life of the products and the resulting enormous amounts of food waste have led to a demand for a rapid, robust, and non-destructive method for the detection of microorganisms, especially in the food sector. Therefore, a rapid and simple sampling method for the Raman- and infrared (IR)-microspectroscopic study of microorganisms associated with spoilage processes was developed. For subsequent evaluation pre-processing routines, as well as chemometric models for classification of spoilage microorganisms were developed. The microbiological samples are taken using a disinfectable sampling stamp and measured by microspectroscopy without the usual pre-treatments such as purification separation, washing, and centrifugation. The resulting complex multivariate data sets were pre-processed, reduced by principal component analysis, and classified by discriminant analysis. Classification of independent unlabeled test data showed that microorganisms could be classified at genus, species, and strain levels with an accuracy of 96.5 % (Raman) and 94.5 % (IR), respectively, despite large biological differences and novel sampling strategies. As bacteria are exposed to constantly changing conditions and their adaptation mechanisms may make them inaccessible to conventional measurement methods, the methods and models developed were investigated for their suitability for microorganisms exposed to stress. Compared to normal growth conditions, spectral changes in lipids, polysaccharides, nucleic acids, and proteins were observed in microorganisms exposed to stress. Models were developed to discriminate microorganisms, independent of the involvement of various stress factors and storage times. Classification of the investigated bacteria yielded accuracies of 97.6 % (Raman) and 96.6 % (IR), respectively, and a robust and meaningful model was developed to discriminate different microorganisms at the genus, species, and strain levels. The obtained results are very promising and show that the methods and models developed for the discrimination of microorganisms as well as the investigation of stress factors on microorganisms by means of Raman- and IR-microspectroscopy have the potential to be used, for example, in the food sector for the rapid determination of surface contamination.
The detection of human skin in images is a very desirable feature for applications such as biometric face recognition, which is becoming more frequently used for, e.g., automated border or access control. However, distinguishing real skin from other materials based on imagery captured in the visual spectrum alone and in spite of varying skin types and lighting conditions can be dicult and unreliable. Therefore, spoofing attacks with facial disguises or masks are still a serious problem for state of the art face recognition algorithms. This dissertation presents a novel approach for reliable skin detection based on spectral remission properties in the short-wave infrared (SWIR) spectrum and proposes a cross-modal method that enhances existing solutions for face verification to ensure the authenticity of a face even in the presence of partial disguises or masks. Furthermore, it presents a reference design and the necessary building blocks for an active multispectral camera system that implements this approach, as well as an in-depth evaluation. The system acquires four-band multispectral images within T = 50ms. Using a machine-learning-based classifier, it achieves unprecedented skin detection accuracy, even in the presence of skin-like materials used for spoofing attacks. Paired with a commercial face recognition software, the system successfully rejected all evaluated attempts to counterfeit a foreign face.
The use of manually fed machines (e.g. table saws) bares risks of injury that are clearly above the average level of other high risk workplaces.
The wide use of such machines causes severe problems for occupational safety and implies high costs for medical treatments and accident annuities.
This thesis presents a new concept of a multispectral sensor to monitor an area in front of a danger zone to detect the user’s limbs and trigger safeguarding measures to prevent an accident in time.
The sensor concept realizes a contact-free material classification, which comprises the development of a system design and specific safety requirements with respect to international safety standards.
Furthermore, a prototypical implementation using four wavebands, which were determined for skin detection through an analysis of reflectance spectra acquired specifically for this purpose, was built.