Abstract
In this study, a new prediction method for polyphenol oxidase (PPO) activity in tomato based on segmentation of Vis-NIR spectral graph characteristics has been proposed. Spectral data are miscellaneous and complex, which increases the prediction modeling complexity. However, spectral data depicted curve is usually approximately continuous and smooth, indicating spectral data is constrained by the smooth spectral curve. There is a strong correlation between adjacent spectral data, and spectral data information redundancy is large. To eliminate information redundancy, compress data, simplify modeling process and improve modeling effects, Vis-NIR spectral discrete points was imagined as a smooth curve, the smooth curve and coordinate axis enclosed area was regarded as a spectral graph. The whole spectral graph was divided into n small graphs, and the area/area-to-perimeter ratio characteristics of every small spectral graph were extracted. The extracted characteristics were used to establish prediction model of PPO activity. Experimental results showed that the proposed algorithm optimizes the modeling effect, reduces the modeling complexity and improves the modeling efficiency. When the spectral graph was averagely segmented into 21 small graphs, area-to-perimeter ratio characteristic was extracted, and multiple linear regression (MLR) was used, the modeling method performed the best. The root mean square error of calibration set (RMSEC) and prediction set (RMSEP) were 1.74 and 1.99 respectively, and the correlation coefficient of calibration set Rc and prediction set Rp were 0.98 and 0.97 respectively. This study provides a reference for nondestructive evaluation of tomato PPO activity using Vis-NIR spectral technology and the proposed spectral data processing method sheds light on spectral data processing.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call