The goal of this thesis is to develop an on-line probe that can be implemented to measure the in situ cell density and viability in bioreactors. This task includes not only hardware development, but also the development of suitable software that can fast and accurately process the signal generated by the hardware.

In spite of the diversity of the methods for on-line measuring cell density or viability separately, it appeared to be a nontrivial task to build a probe suitable for both properties. Even a simple combination of any two of the techniques would lead to great technical complexity. In order to avoid the system complexity, a special probe based on dark field microscopy has been proposed in this work. A reflective condenser has been designed and built to achieve dark field illumination and high contrast images. The imaging system of the probe is also designed and configured in accordance with the system's requirements.

In order to obtain accurate cell density and viability, programs have been written to evaluate the images captured by the probe. The core of the programs is implementing classifiers based on supervised machine learning. Different pattern recognition methods have been utilized in order to find the best way of image processing for the system.