DETERMINATION OF PIGMENTS CONCENTRATION OF OILSEED RAPE (BRASSICA NAPUS L.) LEAVES USING HYPERSPECTRAL IMAGING

作者：He, Y.; Zhang, C.; Liu, F.^*; Kong, W. W.; Cui, P.; Zhou, W. J.; Huang, L. X.

来源：Applied Engineering in Agriculture, 2015, 31(1): 23-30.

DOI：10.13031/aea.31.10690

• Extreme learning machine
• Hyperspectral imaging
• Leaf pigments
• Oilseed rape
• Successive projections algorithm

A hyperspectral imaging system covering the spectral range 380-1030 nm was used to estimate leaf pigment concentration. Spectral information of rape leaves were extracted from the hyperspectral images. Partial least squares (PLS), least squares-support vector machine (LS-SVM), and extreme learning machine (ELM) were applied to build calibration models using spectra of 500-900 nm to determine the concentrations of chlorophyll a (Chl a), chlorophyll b (Chl b), total chlorophyll (tChl), and carotenoids (Car). For full spectra, ELM models performed best with correlation coefficient of prediction (r(p)) of 0.929, root mean square error of prediction (RMSEP) of 0.096 mg g(-1) and residual prediction deviation (RPD) of 2.37 for Chl a; r(p) of 0.883, RMSEP of 0.046 mg g(-1), and RPD of 2.01 for Chl b; r(p) of 0.912, RMSEP of 0.142 mg g(-1), and RPD of 2.29 for tChl; and r(p) of 0.914, RMSEP of 0.030 mg g(-1) and RPD of 2.29 for Car. The results showed that PLS, LS-SVM, and ELM models for chlorophyll a, chlorophyll b, total chlorophyll, and carotenoids obtained strong performances. Seventeen, 15, 19, and 8 sensitive wavelengths were selected for chlorophyll a, chlorophyll b, total chlorophyll, and carotenoids estimation by successive projections algorithm (SPA), respectively. For sensitive wavelengths, ELM models performed best with r(p) of 0.906, RMSEP of 0.108 mg g(-1) and RPD of 2.27 for Chl a; r(p) of 0.897, RMSEP of 0.043 mg g(-1), and RPD of 2.18 for Chl b; r(p) of 0.901, RMSEP of 0.145 mg g(-1), and RPD of 2.25 for tChl: and r(p) of 0.893, RMSEP of 0.034 mg g(-1) and RPD of 2.14 for Car. The results showed that PLS, LS-SVM, and ELM models based on the sensitive wavelengths also obtained good performance closed to the calibration models using full spectra. Compared with all developed models, the ELM models using full spectra and sensitive wavelengths obtained the best performances with highest re and lowest RMSEP, and highest RPD values for Chl a, Chl b, tChl, and Car concentration estimation, respectively. The overall results indicated that hyperspectral imaging with ELM method was an efficient technique for leaf pigment content determination, and the selected sensitive wavelengths would be helpful to develop portable instrument or on-field monitoring sensors in the precise agricultural management.

• 出版日期2015
• 单位浙江大学