Automatic design of infrared absorption filter for material discrimination device.

A method of using the material blend ratio of the light absorption filter as a neural net parameter

Abstract

We propose a method for discriminating materials from infrared reflected light using spectral filters optimized by a neural network. Normally, band pass filters focusing on a specific wavelength are used. However, the band pass filter is expensive. Therefore in this paper, we used organic materials which have complicated absorption spectrum as the filters.By optimizing the blending ratio of these organic material filters via the neural network, low-cost and lightweight material discriminating machines can be automatically designed. The accuracy (99.6\%) equivalent to that of a neural net with infrared spectroscopic spectra was obtained in the task of discriminating whether a plastic piece containing impurities is PP (polypropylene) or non-PP.