Multi-channel fiber optical spectrometer for high-throughput characterization of photoluminescence properties

A Microlens Array Grating for Miniature Multi-Channel Spectrometers

Most existing multi-channel spectrometers are constructed by physically stacking single-channel spectrometers, resulting in their large size, high weight, and limited number of channels. Therefore, their miniaturization is urgently needed. In this paper, a microlens array grating is designed for miniature multi-channel spectrometers. A transmissive element integrating microlens arrays and gratings, the MLAG, enables simultaneous focusing and dispersion. Using soft lithography, the MLAG was fabricated with a deviation of less than 2.2%. The dimensions are 10 mm × 10 mm × 4 mm with over 2000 available units. The MLAG spectrometer operates in the 400-700 nm wavelength range with a resolution of 6 nm. Additionally, the designed MLAG multi-channel spectrometer is experimentally verified to have independently valid cells that can be used in multichannel spectrometers. The wavelength position repeatability deviation of each cell is about 0.5 nm, and the repeatability of displacement measurements by the chromatic confocal sensor with the designed MLAG multi-channel spectrometer is less than 0.5 μm.

Spectral Classification Based on Deep Learning Algorithms

Convolutional neural networks (CNN) can achieve accurate image classification, indicating the current best performance of deep learning algorithms. However, the complexity of spectral data limits the performance of many CNN models. Due to the potential redundancy and noise of the spectral data, the standard CNN model is usually unable to perform correct spectral classification. Furthermore, deeper CNN architectures also face some difficulties when other network layers are added, which hinders the network convergence and produces low classification accuracy. To alleviate these problems, we proposed a new CNN architecture specially designed for 2D spectral data. Firstly, we collected the reflectance spectra of five samples using a portable optical fiber spectrometer and converted them into 2D matrix data to adapt to the deep learning algorithms’ feature extraction. Secondly, the number of convolutional layers and pooling layers were adjusted according to the characteristics of the spectral data to enhance the feature extraction ability. Finally, the discard rate selection principle of the dropout layer was determined by visual analysis to improve the classification accuracy. Experimental results demonstrate our CNN system, which has advantages over the traditional AlexNet, Unet, and support vector machine (SVM)-based approaches in many aspects, such as easy implementation, short time, higher accuracy, and strong robustness.