With the continuous improvement of people’s living standards, people are increasingly high requirements for food quality. Infrared analysis(IR)as a rapid detection method for accurate, sensitive, practical product quality has been accepted by the analysts. Infrared spectral technology is the combination of measurement technology, computer technology and chemometrics. We establish calibration model for collected infrared spectral with chemometrics method, then predict the unknown samples with the calibration model. While in the practical application of the process, due to the instability of collecting samples and the interference of external conditions, which will affect the calibration model. How to get robust calibration models used in actual production is an urgent problem. In this context, this paper research on some algorithms about infrared spectrum signal processing. We focus on the qualitative and quantitative analysis of complex mixtures by Infrared spectroscopy, and we propose several meaningful algorithms, it has important practical significance for quality control of complex mixtures. There are three main innovations of this paper: Based on the characteristics of the infrared spectral data, we focus on the unbalanced and high dimensional small sample classification problem, propose an improved algorithm based on window regression over-sampling(WRO)to generate virtual samples. The algorithm fully consider the distribution of samples as well as the additive and multiplicative information between adjacent samples in a local window. The experimental results show that: for the imbalanced data classification problem, compared with the traditional methods, the proposed algorithm can reduce the possibility of the minority class samples misclassified to some extent; For the high-dimensional small sample classification problems, compared with the traditional feature extraction methods, the proposed algorithm improves the classification accuracy. We research on the robustness of the partial least squares algorithm, pro-pose an improved ensemble algorithm creating diversity partial least squares (CDPLS). This algorithm is based on the “Bias - Variance - Covariance” decomposition, through the introduction of virtual samples to construct a series of diverse sub-models, and add a tradeoff parameter to ensure the correctness of the sub-models. Experimental results show that: Compared with the traditional methods, the proposed ...
修改评论