Sampling method for establishing tobacco near-infrared model

Abstract

The invention discloses a sampling method for establishing a tobacco near-infrared model, comprising the following steps of: obtaining near-infrared scanning spectrograms of tobacco samples in two mutually independent sample sets (A) and (B); carrying out spectrum preprocessing on the near-infrared scanning spectrograms; obtaining the spectrum Mahalanobis distance di from a sample i to the sample set (A); obtaining the component value distance di' from the sample i to the sample set (A); obtaining the integral Euclidean distance Di from the sample i in the sample set (B) to the sample set (A); selecting samples in the sample set (B) according to specific principles and then adding to the sample set (A) to form a new sample set (A'), and finally obtaining a new near-infrared model A' by the establishment of the sample set (A'). The invention overcomes the problem that the prior art can not give attention to predictive ability and model authenticity and provides a sampling method being capable of getting rid of invalid exceptional data; the established model has good predictive ability and applicability and can be widely applied to tobacco industry.

Description

Technical field [0001] The invention relates to a method for establishing a tobacco near-infrared model, in particular to a sample selection method for establishing a tobacco near-infrared model. technical background [0002] Since the 1990s, the application of near-infrared technology in tobacco has become more and more extensive, and various tobacco companies have established their own near-infrared models according to their actual conditions. After entering this century, the domestic tobacco industry has set off a wave of mergers and acquisitions. The originally scattered cigarette factories have combined to form a cross-regional, large-scale industrial company. In the research of near-infrared technology, the ensuing problem appeared: the original near-infrared model established by various tobacco factories can no longer meet the needs of research under the large group model. The reality is that different tobacco factories have established different near-infrared models, but...

AI Technical Summary

Problems solved by technology

The disadvantage of using the empirical selection method to select samples is that it requires a huge amount of sample, and this method ignores the physical and chemical properties of the sample itself, which is easy to cause errors; the disadvantage of the concentration identification method is that it ignores the spectral differences of the samples, which may The prediction ability of the model is low; the disadvantage of the spectral identification method is that it ignores the difference in the intrinsic quality of the sample, which easily leads to model distortion

Images