Significance A variety of sensing applications in the Internet of Things (IoT) raise the diverse needs for sensors. The sensing ability of material composition is a weak part of the sensing layer of the IoT. Due to the technical limitations, traditional analysis methods, such as mass spectrometry and chromatography, cannot be directly applied in the field of the IoT. With the advantages of speediness, lossless and high-efficiency, the technology of near-infrared (NIR) spectroscopy can be applied to the applications of composition analysis. In recent years, the trend of commercialization of spectral analysis has promoted the development of miniaturization, networking and imaging of analytical equipment. The devices based on filters and array detectors are compact without moving parts, which makes them ideal for IoT applications. The miniaturized instruments have been successfully applied in many industrial and agricultural production fields such as petrochemical industry, grain screening, and pharmaceutical industry. With the successful application of machine learning and deep learning, NIR spectral analysis has an intelligent trend of component sensing.

InGaAs focal plane arrays (FPAs) have the advantages of working near room temperature, high detection rate, good uniformity and stability, which are beneficial to realize the miniaturization design of the NIR photoelectric system. As the core photoelectric sensor, InGaAs FPAs are widely used in NIR spectral analysis equipments. This paper introduces the key technologies of the NIR spectral sensing IoT with several kinds of compact InGaAs spectral sensors.

Progress NIR spectral sensing IoT (Fig. 1) integrates spectral analysis techniques by using customized micro spectral sensing nodes. In recent years, Shanghai Institute of Technical Physics, Chinese Academy of Sciences and Shandong University have made good progresses in the research and application of the NIR spectral sensing IoT based on InGaAs FPAs. Spectral sensing nodes can be regarded as customized NIR spectrometers improved by miniaturization, networking and portable design. The wireless communication modules are added on the basis of spectral components, PFA, signal acquisition circuits and optical test accessories, which can meet the application requirements of the IoT. In order to realize the miniaturization design of spectral sensing nodes, two kinds of micro spectral sensors are introduced, which are integrated multi-channel filter and integrated linear variable filter (LVF). The performance of spectral sensors is compared and analyzed with the preparation process. The research and application of NIR sensing nodes, cloud server and mobile phone client are further introduced. Finally, the future development of NIR sensing IoT is proposed.

Based on a 256×1 FPA, the 64-channel filter integrated spectral sensor (Fig. 2) was developed. Considering the impact of filter’s alignment offset with photosensitive elements on the performance of the spectral sensor, four adjacent photosensitive elements were combined as a spectral channel. LVF is a kind of wedge-shaped dielectric thin-film Fabry-Perot narrow-band pass filter. The spacer layer of narrow-band filter was made into wedge-shaped by special coating process. The spacer layer at different positions corresponds to different equivalent optical thickness, thus corresponding to the central wavelength of linearly varying transmittance. The LVF with 900-1700nm was used to develop a LVF type spectral sensor (Fig. 3, Fig. 4), which was coupled with 256×1 and 512×2 FPA respectively. The 256×1 spectral sensor used a single large photosensitive element as a spectral channel. The 512×2 spectral sensor used a combination of multiple small photosensitive elements as a spectral channel, which could reduce the adverse effects of blind elements and nonuniformity of the FPA on the spectral signal. The test results show that the multi-channel filter has higher resolution compared with the LVF (Fig. 5). However, benefitting from the continuous gradient structure, the LVF is less affected by the edge mutation effect and alignment problem between filter and channels. Therefore, the uniformity and consistency between the spectral channels of LVF spectral sensor are better than these of the multi-channel filter (Fig. 6).

For the spectral sensors with different structures, the corresponding spectral sensing nodes were further developed. The optical structure and wireless communication interface had been integrated in the spectral sensing node (Fig. 7). In the network structure, the cloud server platform, the analysis software and the green tea origin identification spectral analysis model were studied (Fig. 8). In the research and development of the next generation spectral sensor, the digitization in the sensor was realized by integrating successive-approximation register structure analog-to-digital converter in the readout circuit.

Conclusion and Prospect This paper introduces the system architecture and key technologies of the NIR spectral sensing IoT. In order to realize the miniaturization design of spectral sensing nodes, two kinds of micro spectral sensors are introduced, which are integrated multi-channel filter and integrated LVF. The performance of spectral sensors is compared and analyzed with the preparation process. The research and application of NIR sensing nodes, cloud server and mobile phone client are further introduced. Finally, the future development of NIR sensing IoT is proposed. In the long term, the future development direction of spectral sensing IoT is intellectualization. The integration of system on chip (SoC), wireless communication module and intelligent analysis algorithm will be further realized in the intelligent spectral sensors.