Study On Hyperspectral Estimation Of Chlorophyll Content In Rape Leaves

In this study,rapeseed crops in Northwestern China were used as the object.Through three-year field experiment and observation of leaf spectrum and chlorophyll content,the characteristics of reflectance spectrum and chlorophyll content changes of rapeseed in different years and growth stages were analyzed and compared,and the leaf spectral reflectance and chlorophyll content were discussed.Select the commonly used spectral parameters with clear physical meanings,and build a remote sensing inversion model for predicting chlorophyll content.After comparing and verifying the three models of MLSR,PLSR and BP,as well as the accuracy test,as well as BP,MLSR-BP and MLSR-GA-BP’s comparison leads to the following conclusions:（1）The average chlorophyll content of rape showed a downward trend from the seedling stage to the mature stage.The SPAD values of the four growth periods except2018 showed significant differences from each other.The SPAD values of the seedling stage in 2018 showed significant differences from the other three growth periods,and mature There was a significant difference in the SPAD value between the three growth periods and the other three growth periods,and there was no significant difference in the SPAD value during the bud and blooming period.（2）The overall shape trend of the spectral reflectance of rapeseed in each growth period is the same.Between 400 nm and 500 nm,an absorption valley is formed;between600 nm and 700 nm,a second absorption valley is formed,and a strong reflection is formed at 550 nm.Peak,the reflectivity rises sharply at 670 nm-780 nm;a continuous high reflectivity platform is formed at 780 nm-1000 nm.In the Lvfeng area,the spectral reflectance of the seedling stage,bud and stalk stage,flowering stage,and mature stage showed an upward trend.The shape of the hyperspectral reflectance curve corresponding to the leaves of rape with different SPAD content is the same.In the visible light band,with the increase of SPAD content,the spectral reflectance becomes smaller.In the near-infrared band,as the SPAD content increases,the reflectivity increases simultaneously.（3）The correlation trend between SPAD value and original spectral reflectance of rape leaves at different growth stages is consistent.With the change of wavelength,there are 4inflection points,3 of which are located near 550 nm,670nm and 710nm;at 550 nm and710 nm The negative correlation coefficient reaches the peak;the wavelengths where the SPAD value of rape leaves and the original spectral reflectance are significantly correlated at the levels of 0.01 and 0.05 are basically located in the trilateral spectral band and the near-infrared band.（4）By analyzing the correlation between 31 kinds of spectral parameters and leaf chlorophyll content,the spectral parameters with large correlations in each growth period were selected to participate in the modeling.The number of spectral parameters participating in the modeling from 2017 to 2019 were 7 respectively.One,five,seven.In terms of the number of spectral parameters that reached a significant level,the SPAD values of rapeseed leaves at the seedling and mature stages of the multi-year experiment have a good correlation with the spectral parameters;in the bud and bloom stage,the correlation in 2018 is the worst,2017 and 2019 The correlations are similar and both are good.From the perspective of the maximum correlation of spectral parameters,the best is2019,with the maximum correlation coefficient of 0.908（VOG₂and MTCI）ranking first;the maximum correlation coefficient of 0.849（VOG₂and VOG₃）in 2017 is the second;the worst is 2018,The maximum correlation coefficient is 0.839（VOG₂）.（5）The chlorophyll content estimation models of different rapeseeds have been tested and verified for accuracy.The BP neural network model is the best,followed by the partial least square regression model,and the multiple linear stepwise regression model is the worst.The fit of the built models reached a very significant level.The modeling R²of the BP neural network model in 2017 was 0.917,0.925,0.947,and 0.827 in the four growth periods,respectively,and the verification R²reached 0.864,0.803,0.927,and 0.811respectively;the modeling R²of the BP neural network model in 2018 was in four The growth periods are 0.826,0.523,0.799,0.901,and the verification R²reaches 0.803,0.535,0.667,0.874,respectively;the modeling R²of the BP neural network model in 2019 is0.816,0.906,0.852,0.862 in the four growth periods,verification R²reached 0.785,0.889,0.735,0.859,respectively.Three years of rapeseed data were used to model by stratified sampling method and any combination of two years as the modeling set,and the other year as the verification set.The results also showed that the BP neural network model was the best,followed by partial least squares.Multiplying regression model,multiple linear stepwise regression model is the worst.（6）Based on the data of 2019,three models of BP,MLSR-BP and MLSR-GA-BP are established.The fit of the built models reached a very significant level.The effect of the first three growth period models is that BP is better than MLSR-GA-BP better than MLSR-BP.The effect of the model in the mature period in 2019 is that MLSR-GA-BP is better than BP than MLSR-BP.