Abstract：The accurate detection of the content of chlorophyll in citrus leaves is of great significance to the nutritional status and the growth trend of citrus. A rapid and non-destructive method for diagnosing chlorophyll content in citrus leaves was studied in order to provide a reference for the detection of chlorophyll content in citrus leaves by Raman spectroscopy. A hundred and twenty slices of citrus leaves with different canopy heights and different geographical distributions were collected. The dust on the surface of the leaves was wiped off. The deionized water was used in the laboratory to clean it, dried in a sealed bag, and labeled with a label. The Raman spectra of citrus leaves were then collected. The parameters were set as follows: resolution 3 cm-1, integration time 15 s; laser power 50 mW. Three methods were used, such as baseline wavelet, iterative restricted least squares （IRLS）and asymmetric least squares (ALS), for background correction of Raman spectroscopy. After that, Partial least squares (PLS) method was used to establish the quantitative model. Subsequently, four methods of spectral pretreatment, like Savitzky-Golay convolution smoothing (SG smoothing), normalization, multiplicative scatter correction (MSC) and the Savitzky-Golay 1st derivative, were used to further optimize the spectra which had been treated by the background correction. The research process showed that the PLS model was established by the spectra of the original spectrum, Baseline Wavelet, IRLS, and ALS preprocessing. The correlation coefficients of the models were 0.858, 0.828, 0.885, and 0.862, respectively. The root mean square error cross validation, RMSECV were 5.392, 5.870, 4.934, and 5.336, respectively. The best principal component factors were 8, 3, 8 and 8 respectively. The RMSECV of the pre-processed PLS model deducted from the IRLS background was the smallest, the correlation coefficient was the highest, and the modeling effect was the best. SG smoothing, normalization, MSC and SG 1st Der preprocessing methods were used to preprocess IRLS background correction spectrum and establish PLS model. The results showed that: IRLS spectrum and its combination of SG smoothing, normalization, MSC and SG 1st Der The PLS of the four pretreatment methods of r were 0.885, 0.897, 0.852, 0.863, and 0.888, respectively. The RMSECV were 4.934, 4.715, 5.595, 5.182, and 4.962, respectively. The best principal component factors were 8, 8, 8, 8 and 5, respectively; the RMSECV of the PLS model after IRLS-SG smoothing was the smallest, and the model had the best effect. After verifying the PLLS model preprocessed by IRLS-SG, the predictive correlation coefficient r of the prediction set was 0.844, the root mean square error of prediction (RMSEP) was 5.29, and the prediction accuracy was high. Three kinds of background correction methods combined with four kinds of spectral pretreatment methods were used to quantitatively model the Raman spectra of citrus leaves. It can be concluded that the experimental results after IRLS background correction combined with the SG smoothing are optimal. The modeling set r is 0.897, the RMSECV is 4.715, the prediction set r is 0.844, and the RMSEP is 5.29, and the prediction accuracy is high. Studies have shown that Raman spectroscopy combined with background correction methods can provide a quick and easy analytical method for quantitative analysis of chlorophyll content in citrus leaves.
Key words：BaselineWavelet; Iterative restricted least squares; Asymmetric least squares; SG smoothing; Raman spectrum; Chlorophyll; Citrus leaves
[1] XIAO Yi, ZHU Xin-guang(肖 怡, 朱新广). Plant Physiology Journal(植物生理学报), 2016，(11): 1663.
[2] YUE Xue-jun, QUAN Dong-ping, HONG Tian-sheng(岳学军, 全东平, 洪添胜), 等. Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2015, 31(1): 294.
[3] Peng F, Liu S B, Xu H Y, et al. Advanced Materials Research, 2013, 726-731: 1411.
[4] LI Xiang, LÜ Yong(李 响, 吕 勇). Journal of Beijing Information Science & Technology University(北京信息科技大学学报自然科学版), 2013，(2): 27.
[5] FANG Cheng-xuan, LI Jian-hua, LIANG Yi-zeng(房承宣, 李建华, 梁逸曾). Journal of Instrumental Analysis(分析测试学报), 2012, 31(5): 541.
[6] Bian X, Chen D, Cai W, et al. Chinese Journal of Chemistry, 2011, 29(11): 2525.
[7] HAO Yong, CHEN Bin(郝 勇, 陈 斌). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2015，35(9): 2563.
[8] CHEN Shan(陈 珊). Central South University(中南大学), 2011.
[9] Lee K, Kwon H, You K. TDOA International Conference on Intelligent Informatics and Biomedical Sciences. IEEE, 2017: 92.
[10] HUANG Shi-hao, PAN Tie-zhu(黄时浩, 潘铁柱). Science and Technology and Innovation(科技与创新), 2016, (16): 81.
[11] JIANG An, PENG Jiang-tao, XIE Qi-wei, et al(姜 安, 彭江涛, 谢启伟, 等). Computers and Applied Chemistry(计算机与应用化学), 2012, 29(5): 537.
[12] LIU Yan-de, XIE Qing-hua, WANG Hai-yang, et al(刘燕德, 谢庆华, 王海阳, 等). Chinese Journal of Luminescence(发光学报), 2016, 37(9): 1135.
