An innovative multispectral sensor for rapid wine adulteration detection using wavelength selection algorithms.

Effective detection of wine adulteration can protect the rights and interests of consumers and producers. The combination of spectroscopic technology and chemometric methods can achieve adulteration discrimination of wine, providing feasibility for developing consumer-grade wine adulteration detection equipment. This study attempts to use a wavelength selection algorithm to guide the development of a four-channel, low-cost spectroscopic device for rapid identification of wine adulteration. Firstly, this study utilized a miniature spectrometer to obtain the Vis-NIR spectra of wine (only the Vis portion was used). Secondly, by integrating the successive projections algorithm (SPA), spectral degradation, and traversal methods, four wavelengths and their parameters related to wine adulteration discrimination are determined for the selection of filters. Then, a CCD camera and four filters were packaged into a 4-channel spectral sensor. Finally, a random forest (RF) discrimination model was established, and an improved spectral index (SI) combined method was used to enhance the model's performance. The results indicate that the snapshot-based multispectral sensor can achieve high accuracy in identifying wine adulteration (97.47%) and has the characteristics of small size and low cost. At the same time, the research findings also demonstrate that by integrating the SPA, SI, spectral degradation, and traversal methods, effective central wavelengths and resolution specifications can be provided for the construction of dedicated spectral sensors.