Analysis of soil Near-Infrared Spectra based on Multi-Fractal Theory. Issue 17 (2018)

• Record Type: Journal Article
• Authors: Wei, Yang; Peng, Zhou; Minzan, Li; Ziyuan, Hao; Zhen, Liu; Xiangqian, Yao
• Is Part Of: IFAC-PapersOnLine. Volume 51:Issue 17(2018)
• Page Start: 607
• Page End: 612
• Publication Date: 2018
• Subjects: multifractal spectrum -- soil nitrogen content -- path analysis -- feature extraction method; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2018.08.130
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: In order to effectively utilize the local absorption characteristics of spectrum and the amplitude and variation trend of spectral waveform, a multifractal spectrum based spectral signal singularity feature extraction method is proposed, and multifractal spectrum is introduced to characterize the singularity of spectral curve. Firstly, the generalized Hurst exponent, mass index and other related characteristic parameters of the spectrum in five different bands are extracted by using the popular multifractal detrend wave analysis. It is found that they all exhibit typical multifractal properties. 1400 nm~1800 nm and 1800~2200nm can better reflect the difference of multifractal characteristics than those of other bands. Then, the correlation analysis between the multifractal characteristic parameters in 1400~1800nm and 1800~2200nm bands and the soil total nitrogen content was carried out, and the relationship between seven multifractal parameters with high correlation was further analyzed by path analysis. Finally, five multifractal parameters were selected by path analysis method and were used as variable to build multivariate regression with soil total nitrogen content. The prediction R 2 and the verification R 2 of the regression model were 0.8690 and 0.8364, respectively. The experimental results show that the feature extraction method based on multifractal spectrum can be effectively applied to soil near infrared spectrum data processing.