Performance-Steigerung multispektraler Systeme durch eine analytische Auswahl der Spektralbänder am Beispiel von a-Si:H Photodioden

The objective of this present dissertation is to increase the measurement precision and performance of the non-destructive characterization of materials by means of their reflective properties. The paper describes the development of an analytical procedure for separating those specific spectral bands exhibiting a high information density. The basis for such characterization is the light that is reflected or absorbed by a material and which thereby acquires an unequivocal spectral signature. Analytic separation enables similar measurement results to be obtained with a minimal amount of spectral bands as compared to the more complex hyperspectral procedures. As is invariably the case in such hyperspectral procedures, several hundred bands are recorded for each measurement, since characteristic extraction does not take place until after the measurement is taken based on information acquired through empirical and probabilistic means. The paper documents the various stages of development in which parameters are defined and an algorithm implemented for automatic separation. It all starts with the voltage-tunable multispectral photodiodes that were developed at the institute, which have a dynamically changeable information density per pixel. The results that are obtainable with this detector type with respect to the number of spectral bands that can be achieved cannot be analyzed with current trichromatic methods used in colorimetry. The parameter ΔCL was defined based on the general requirement of the external separation. In the case of a sensor with an arbitrary number of spectral bands for a given database of materials and under a certain type of lighting, the function returns a value for ΔCL that is an independent dimension for external separation. Since, however, even an extremely high degree of separation does not provide any certainty with regard to how reliably the measurement was done, the second parameter μ(P) was defined according to the membership function from the area of fuzzy sets. μ(P) can be used to determine with what probability the measured sample P involves a material from the database of materials or whether the sample is unknown. In tandem with the development of the parameters and the analytic procedures resulting from them were experimental measurements for the purpose of being able to distinguish between whitish powder samples. The process of setting up a database in this regard received special attention, which - along with the powder samples´ spectral reflection curves - also involves the spectral irradiance of diverse real and standard light sources as well as the spectral sensitivity of diverse sensors. This allows for the possibility of resorting to simulated values in addition to actually measured values to provide a basis for widely applied, multivariate analysis in order to reengineer the diodes. Special significance is given to the development of such multispectral diodes that have been produced entirely with planar technology and optimized by shrewdly selecting the bands in conjunction with hyperspectral imaging for remote sensing, environmental monitoring, and touch-free and non-destructive material analysis. In established hyperspectral systems, the optics, which are upstream of the sensor and whose purpose is to spectrally separate the incident light, lead to unwieldy and expensive equipment. This situation has so far prevented such measurement systems from being widely distributed. By implementing an analytic procedure for separating spectral bands that exhibit high information density, a procedure was able to be devised that - even without expert knowledge - produces the best combination of spectral bands as output. Not least, this property proves that reducing the spectral bands by shrewdly selecting them paves the way for a new generation of hyperspectral systems with a broad range of fields of application.