Application of neural network for fiber-optic color sensor

The paper presents construction of a simple fiber-optic color sensor designed for recognition of objects from a set comprising an "a prori" defined, limited number of colors. Color sensing is based on diffusive reflectivity of examined surfaces measured in several separate spectral channels. Number and positions of channels depend on predetermined set of colors. The channels are fed by selected LED's sequentially, i.e. in Time-Division Multiple Access mode. Collected serially analogue signals are converted onto digital data enabling some processing variants. In order to extend the performance capability of such a sensor and to enable measurements of others colors, a simple neural network was applied.