分数阶微分预处理及PCA-SRDA的多模型融合对红富士苹果产地溯源

doi:10.3964/j.issn.1000-0593(2022)10-3249-07

Abstract
References
Related Citation (15)

Download: PDF (4554 KB)
Export: BibTeX | EndNote (RIS)

Abstract Apple’s origin traceability has important application value and practical significance. To explore new ways to trace apple’s origin, taking 671 Samples of Red Fuji apples from Aksu of Xinjiang Province, Yantai of Shandong Province and Luochuan of Shanxi Province as the research objects. The near-infrared transmission spectra of the samples at 590~1 250 nm are collected respectively, and then the techniques of Fractional Differential (FD) and Principal Component Analysis (PCA)-Spectral Regression Discriminant Analysis (SRDA) are used to fuse multiple models. An integrated learning model of the red Fuji apple’s origin traceability is constructed. Firstly, spectral data after spectral correction are divided into a training set and test set, and the fractional-order differential technique is used to preprocess the spectrum of the training set to obtain fractional-order differential spectra of different orders (order 0～2 and step size 0.1 in this paper). A new training set is constructed based on the prediction results of the base learner, built by combining different orders of fractional differential spectra and the PCA-SRDA algorithm, and the final classification prediction model is obtained by fusing the decision tree algorithm. Then, the corresponding order fractional differential is used to preprocess the spectrum of the test set, and the corresponding prediction results are obtained based on the established base learner. Finally, the results are formed into a new test set, and the final prediction results are output based on the established classification prediction model. The sample-set is randomly divided according to the ratio of 7∶3, and the experiment is repeated 200 times. The results show that the multi-model fusion and integration learning model combined with the fractional-order differential preprocessing, Linear Discriminant Analysis (LDA), SRDA, PCA-LDA and PCA-SRDA algorithms has a good Discriminant effect and strong robustness. Among them, The FD-PCA-SRDA multi-model fusion and integration learning model is the best, and the average accuracy and standard deviation of the training set are 97.33% and 0.49%, and the average accuracy and standard deviation of the test set are 94.84% and 1.48%, respectively. Therefore, the fractal-order differential technique and PCA-SRDA algorithm combined with the near-infrared transmission spectrum can successfully and effectively realize apple’s origin traceability.

Key words：Near-infrared transmission spectrum; Fractional differential; Principal component analysis-spectral regression discriminant analysis; Apple; Origin traceability

Received: 2021-08-08 Accepted: 2021-11-16

ZTFLH:

S661.1

Corresponding Authors: GUO Jun-xian E-mail: junxianguo@163.com

Cite this article:

HUANG Hua,NAN Meng-di,LI Zheng-hao, et al. Multi-Model Fusion Based on Fractional Differential Preprocessing and PCA-SRDA for the Origin Traceability of Red Fuji Apples[J]. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2022, 42(10): 3249-3255.

URL:

http://www.gpxygpfx.com/EN/10.3964/j.issn.1000-0593(2022)10-3249-07 OR http://www.gpxygpfx.com/EN/Y2022/V42/I10/3249