Crops ›› 2020, Vol. 36 ›› Issue (6): 180-188.doi: 10.16035/j.issn.1001-7283.2020.06.027

Previous Articles     Next Articles

Comparison of Estimation Methods for Growth Parameters of Winter Wheat Based on Full-Band Hyperspectral Data

Ji Jingchun1,2(), Liu Jianli1, Niu Yujie1,2, Xuan Kefan1,2, Jiang Yifei1,2, Deng Haodong1,3, Li Xiaopeng1()   

  1. 1Institute of Soil Science, Chinese Academy of Sciences, Nanjing 210008, Jiangsu, China
    2University of the Chinese Academy of Sciences, Beijing 100049, China
    3Hydrology and Water Resources College, Hohai University, Nanjing 210008, Jiangsu, China
  • Received:2020-02-26 Revised:2020-04-24 Online:2020-12-15 Published:2020-12-09
  • Contact: Li Xiaopeng E-mail:jcji@issas.ac.cn;lixp@issas.ac.cn

RichHTML

4

PDF (PC)

332

Abstract:

Hyperspectral data monitoring in crop growth status has the characteristics of nondestructive and efficient, which will be the developing direction of modern agriculture. In order to simplify the hyperspectral data processing procedure and apply to the real-time monitoring of crop growth, the original hyperspectal reflectance was directly used to complete the whole process from modeling to estimating the crop growth patameters and three methods including partial least square regression (PLSR), support vector regression (SVR) and feed forward neural network (FNN) were used to estimate the growth parameters (aboveground biomass, leaf area index, total nitrogen content and chlorophyll concentration) of winter wheat at multiple key growth stages (jointing, booting, anthesis and milk ripe stage) based on full-band hyperspectral data, respectively. The modeling and estimation performances of the three methods were also compared. For the SVR method, the mean R2 of the above growth parameters during the four growth periods was between 0.89 and 0.98, while MAPE was between 1.70% and 7.53% in calibration set and the mean R2 of the above growth parameters during the four growth periods was between 0.90 and 0.94, while MAPE was in between 4.04% and 7.46% in validation set. Both the interpretation of parameters and the estimation accuracy were higher than PLSR and FNN methods, which illustrated that the SVR method performed best in estimating winter wheat growth parameters using all-band spectral reflectance in the tested methods. With the maturity of UAV-based hyperspectral technology, the SVR method can be used to process a wide range of field hyperspectral information obtained fromaerial photography, facilitate modeling and parameter inversion, and reflect crop growth status in time.

Key words: Hyperspectral remote sensing, Partial least squares regression, Support vector regression, Feedforward neural network, Monitoring of crop growth parameters

Fig.1

Experimental design of nitrogen rate test N0, N1, N2, N3 and N4 indicate 0, 150, 190, 230 and 270kg/hm2"

Table 1

Number of nodes in the hidden layer of FNN"

生长参数
Growth parameter		 拔节期
Jointing
stage		 孕穗期
Booting
stage		 扬花期
Anthesis
stage		 乳熟期
Milk ripe
stage
地上部生物量Aboveground biomass 15 15 15 15
LAI 15 25 25 15
全氮含量
Total nitrogen content		 25 15 15 15
叶绿素浓度
Chlorophyll concentration		 25 25 15 15

Fig.2

Growth parameters of winter wheat under different nitrogen fertilizer gradients Different lowercase letters indicate significant difference at 0.05 level"

Fig.3

Canopy hyperspectral reflectance of winter wheat under different nitrogen fertilizer gradients"

Fig.4

Fitted results for calibration samples of PLSR, SVR and FNN model"

Table 2

The evaluation indicators of PLSR, SVR and FNN model in calibration set"

模型Model 生育期
Growth stage		 地上部生物量
Aboveground biomass		 LAI 全氮含量
Total nitrogen content		 叶绿素浓度
Chlorophyll concentration
R2 RMSE
(g/m2)		 MAPE
(%)		 R2 RMSE MAPE
(%)		 R2 RMSE
(%)		 MAPE
(%)		 R2 RMSE
(μmol/m2)		 MAPE
(%)
PLSR 拔节期Jointing 0.90 18.61 10.58 0.93 0.13 9.43 0.90 0.12 4.96 0.91 34.51 7.03
孕穗期Booting 0.87 63.87 8.73 0.95 0.34 9.22 0.92 0.09 3.36 0.91 37.43 6.65
扬花期Anthesis 0.92 79.83 6.57 0.93 0.39 8.46 0.91 0.10 4.44 0.86 50.92 8.40
乳熟期Milk ripe 0.91 124.83 8.09 0.94 0.36 7.84 0.94 0.13 6.40 0.94 35.01 8.19
SVR 拔节期Jointing 0.92 16.31 7.53 0.95 0.11 5.46 0.98 0.05 1.70 0.94 28.02 4.93
孕穗期Booting 0.91 53.30 6.36 0.97 0.25 4.55 0.97 0.05 1.70 0.93 32.54 5.32
扬花期Anthesis 0.94 72.77 5.38 0.94 0.35 5.96 0.94 0.08 3.14 0.89 45.00 6.49
乳熟期Milk ripe 0.95 94.46 5.25 0.95 0.32 5.65 0.97 0.10 3.54 0.96 26.25 4.70
FNN 拔节期Jointing 0.93 15.37 7.58 0.96 0.10 6.03 0.97 0.06 2.13 0.96 24.50 4.45
孕穗期Booting 0.92 49.69 6.80 0.96 0.29 5.96 0.96 0.07 2.25 0.93 33.44 5.26
扬花期Anthesis 0.94 71.72 5.49 0.95 0.35 7.13 0.93 0.09 3.46 0.90 45.00 7.35
乳熟期Milk ripe 0.95 93.57 5.59 0.96 0.30 6.65 0.97 0.10 4.24 0.96 27.70 5.93

Fig.5

Predicting result of growth parameters of PLSR, SVR and FNN model"

Table 3

The evaluation indicators of PLSR, SVR and FNN model in validation set"

模型
Model		 生育期
Growth stage		 地上部生物量
Aboveground biomass		 LAI 全氮含量
Total nitrogen content		 叶绿素浓度
Chlorophyll concentration
R2 RMSE
(g/m2)		 MAPE
(%)		 R2 RMSE MAPE
(%)		 R2 RMSE
(%)		 MAPE
(%)		 R2 RMSE
(μmol/m2)		 MAPE
(%)
PLSR 拔节期Jointing 0.89 20.03 11.15 0.92 0.14 10.75 0.90 0.12 5.31 0.91 34.67 7.13
孕穗期Booting 0.87 64.19 8.73 0.95 0.35 9.44 0.92 0.09 3.36 0.91 37.31 6.78
扬花期Anthesis 0.92 80.70 6.60 0.93 0.40 8.62 0.91 0.11 4.57 0.86 50.38 8.31
乳熟期Milk ripe 0.91 127.58 8.34 0.94 0.37 8.00 0.94 0.13 6.46 0.94 36.01 8.44
SVR 拔节期Jointing 0.90 19.09 9.16 0.93 0.13 6.98 0.95 0.08 3.31 0.92 33.13 6.36
孕穗期Booting 0.89 59.99 8.23 0.96 0.31 6.69 0.93 0.08 3.05 0.92 35.20 6.16
扬花期Anthesis 0.93 77.44 5.99 0.93 0.39 7.38 0.90 0.11 4.77 0.86 50.97 8.11
乳熟期Milk ripe 0.94 103.96 6.44 0.94 0.35 7.15 0.95 0.11 5.04 0.95 31.26 6.74
FNN 拔节期Jointing 0.88 20.82 11.09 0.92 0.14 8.42 0.94 0.10 3.87 0.89 39.24 7.93
孕穗期Booting 0.87 64.40 9.22 0.92 0.42 9.26 0.91 0.09 3.63 0.88 44.20 7.83
扬花期Anthesis 0.91 88.01 7.19 0.90 0.49 10.34 0.90 0.11 4.76 0.86 52.52 8.71
乳熟期Milk ripe 0.92 118.21 7.51 0.92 0.42 9.60 0.94 0.13 6.16 0.94 36.61 8.44
[1] Li X C, Zhang Y J, Bao Y S, et al. Exploring the best hyperspectral features for LAI estimation using partial least squares regression. Remote Sensing, 2014,6(7):6221-6241.
doi: 10.3390/rs6076221
[2] Jia M, Li W, Wang K K, et al. A newly developed method to extract the optimal hyperspectral feature for monitoring leaf biomass in wheat. Computers and Electronics in Agriculture, 2019,165(1):1-7.
[3] Duan D D, Zhao C J, Li Z H, et al. Estimating total leaf nitrogen concentration in winter wheat by canopy hyperspectral data and nitrogen vertical distribution. Journal of Integrative Agriculture, 2019,18(7):1562-1570.
doi: 10.1016/S2095-3119(19)62686-9
[4] Huang L S, Zhang H S, Ding W J, et al. Monitoring of wheat scab using the specific spectral index from ASD hyperspectral dataset. Journal of Spectroscopy, 2019: 9153195.
[5] Bao Y D, Mi C X, Wu N, et al. Rapid classification of wheat grain varieties using hyperspectral imaging and chemometrics. Applied Sciences, 2019,9(19):4119-4135.
doi: 10.3390/app9194119
[6] 田明璐, 班松涛, 常庆瑞, 等. 基于无人机成像光谱仪数据的棉花叶绿素含量反演. 农业机械学报, 2016,47(11):285-293.
[7] 梁栋, 杨勤英, 黄文江, 等. 基于小波变换与支持向量机回归的冬小麦叶面积指数估算. 红外与激光工程, 2015,44(1):335-340.
[8] Curran P J, Dungan J L, Macler B A, et al. Reflectance spectroscopy of fresh whole leaves for the estimation of chemical concentration. Remote Sensing of Environment, 1992,39(2):153-166.
doi: 10.1016/0034-4257(92)90133-5
[9] Osborne S, Schepers J S, Schlemmer M. Using multi-spectral imagery to evaluate corn grown under nitrogen and drought stressed conditions. Journal of Plant Nutrition, 2004,27(11):1917-1929.
doi: 10.1081/PLN-200030042
[10] Thenkabail P S, Smith R B, De Pauw E. Hyperspectral vegetation indices and their relationships with agricultural crop characteristics. Remote Sensing of Environment, 2000,71(2):158-182.
doi: 10.1016/S0034-4257(99)00067-X
[11] 陆国政, 李长春, 杨贵军, 等. 基于成像高光谱仪的大豆叶面积指数反演研究. 大豆科学, 2016,35(4):599-608.
[12] Svante W, Sjöström M, Eriksson L. PLS-regression:a basic tool of chemometrics. Chemometrics and Intelligent Laboratory Systems, 2001,58(2):109-130.
doi: 10.1016/S0169-7439(01)00155-1
[13] Cherkassky V. The nature of statistical learning theory. IEEE Transactions on Neural Networks, 1997,8(6):1564.
doi: 10.1109/TNN.1997.641482 pmid: 18255760
[14] Hsu C W, Lin C J. A comparison of methods for multiclass support vector machines. IEEE Transactions on Neural Networks, 2002,13(2):415-425.
doi: 10.1109/72.991427 pmid: 18244442
[15] Fulkerson B. Machine learning,neural and statistical classification. Technometrics, 1997,37(4):459-459.
doi: 10.1080/00401706.1995.10484383
[16] Cheng C M, Wei Y C, Xu J J, et al. Remote sensing estimation of chlorophyll a and suspended sediment concentration in turbid water based on spectral separation. Optik, 2013,124(24):6815-6819.
doi: 10.1016/j.ijleo.2013.05.078
[17] 梁栋, 杨勤英, 黄文江, 等. 基于小波变换与支持向量机回归的冬小麦叶面积指数估算. 红外与激光工程, 2015,44(1):335-340.
[18] 贺佳, 刘冰峰, 郭燕, 等. 冬小麦生物量高光谱遥感监测模型研究. 植物营养与肥料学报, 2017,23(2):313-323.
[19] 贾学勤, 冯美臣, 杨武德, 等. 基于多植被指数组合的冬小麦地上干生物量高光谱估测. 生态学杂志, 2018,37(2):424-429.
No related articles found!
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!