Internal Quality Detection of Apple Based on Near Infrared Spectrum Characteristic Wavelength
Pengkun Quan,Juan Zhao,Mengsheng Zhang,Haihui Zhang
DOI: https://doi.org/10.13031/aim.201900500
2019-01-01
Abstract:Broadband near infrared spectroscopy (NIRS) is widely used in nondestructive testing of agricultural products, which has many problems, such as high cost, complex analysis process and redundancy of spectral information. In this paper, the quality information of Apple was predicted by using the selected characteristic wavelengths of Apple Internal quality. Through the near infrared spectroscopy detection platform, 595 apple samples with different maturity were selected to obtain the diffuse reflectance spectrum information of samples, and the soluble solids, titratable acid and hardness (sugar, acidity, hardness) of each sample were measured. Using a variety of feature band extraction algorithms, such as CARS, MWPLS, MCUVE and RF, the range of 10 characteristic bands most relevant to three parameters is finally optimized. The results showed that: When modeling with characteristic wavelength, the correlation coefficients of the prediction set were 0.86, 0.81 and 0.82, and the root mean square errors were 0.46, 0.018 and 0.33. The correlation coefficients of the modeling set were 0.82, 0.78 and 0.79, and the root mean square errors were 0.49,0.026 and 0.37. By optimizing the characteristic bands of three indexes, not only the number of wavelengths participating in modeling is significantly reduced, but also the performance of the regression model is close to that of the full spectrum modeling. The rapid detection of internal quality is realized, which provides a theoretical basis for guiding the development of portable equipment.