A Deep Learning-Aided Remote Spectrally Tunable LED Light Source Integrated System

In recent years, the utilization of the spectrally tunable light sources (STLSs) is increasing dramatically across a wide range of applications. However, the existing multispectral systems still possess many crucial issues, such as relatively narrow spectral ranges, an unintegrated system, poor fittings, and relatively large sizes. In this study, a light-emitting diode (LED)-based STLSs integrated system which integrates the remote controlling, accurate power supply, and reliable heat dissipation has been carefully designed, constructed, calibrated, and demonstrated. A hybrid offline and real-time online feedback technique is proposed based on deep learning offline training and real-time online feedback fine-tuning to precisely achieve the target spectrum. The maximum deviation of common target spectral matching in the range of 380-780 nm is 2.75%. The system output maintains a high-level stability over a large dynamic range with only 2.03% maximum deviation. A dual-light output of the system facilitates the capability as standard luminance and illuminance calibration light source. The proposed integrated system has shown promising merits, such as compact size, high efficiency, and exactitude, making it an ideal universal tunable light source for multiple applications.