Speaker
Description
This paper attempts to perform spectral discrimination of fluorescence images of liquid scintillators acquired by complementary metal oxide semiconductor (CMOS) image sensors using the discriminative ability of deep convolutional neural networks without any special effort. As semiconductor fab-processing technology has advanced, so has the processing technology of optical elements in image sensors. There is a trade-off between the pixel size of an image sensor and the signal-noise ratio and high color reproduction. Manufacturers of commercial complementary metal oxide semiconductor (CMOS) image sensors do not provide users with spectral response data for their CMOS sensors. We generated training images with a light-emitting diode module programmable on a single-board computer. We demonstrated the feasibility of inferring the spectral response backward from the discriminant values of a deep convolutional neural network. As a follow-up to the previous study, considering the operational characteristics of neutrino experiments, the feasibility of implanting a deep convolutional neural network for monitoring the attenuation distance and spectral response of light in a liquid scintillator was confirmed in terms of supervised learning. In the future, we will optimize efficient transformer implantation with limited computational resources for the characteristics of the Internet of Things.
