Accuracy Improvement of Avalanche Photodiode based Filter-less Spectroscopy by Deep Learning

In this paper, we report a highly sensitive filter-less spectroscopy method using an avalanche photodiode (APD) by deep learning. Many different band pass filters or grating methods are commonly employed. Electron multiplying charge-coupled devices (CCD) are commonly used for high sensitivity sensing. However, conventional devices consisting of filters or gratings provide a relatively huge system size. In addition, the light intensity diminishes at the filters/gratings. Previously, experimental measurements using the proposed APD-based spectroscopy method exhibits the sensing capability to detect and divide dual wavelength light without band pass filters or grating for applying fluorescence and excitation lights detection. However, the proposed spectroscopy method has a problem in calculation with noises. The noises lead an inaccurate answer. So we propose a new method with deep learning instead of the calculation for improving spectroscopy accuracy.