Estimation Of Potato Chlorophyll Content Based On Hyperspectral Analysis

Chlorophyll is the most important indicator pigment in photosynthesis,allowing plants to power the biosphere by converting light energy into chemical energy.From the perspective of precision agriculture,since chlorophyll content is directly related to plant stress and senescence,real-time monitoring of leaf chlorophyll content is of great significance for grasping the growth status of crops in real time.Traditional methods for determination of chlorophyll content are time-consuming and laborious.At present,the development of hyperspectral technology provides an effective method for estimating potato chlorophyll content.However,the determination of chlorophyll content was affected by canopy changes,growth period and soil background factors.Based on the potato field experiment with multiple nitrogen levels over many years,the canopy spectral parameters and SPAD values were collected in the key growth period of potato.Using the database established by PROSAIL model,a look-up table was established to estimate the chlorophyll content of potato during key growth period.At the same time,six kinds of spectral indices were used to optimize the band to find the optimal spectral index with the best correlation with potato chlorophyll content,and the estimation model of potato chlorophyll content in each growth period was established.Then,the machine learning regression algorithm was used to estimate potato chlorophyll content,and the sensitive band of potato chlorophyll content and the corresponding optimization modeling algorithm were found to provide a theoretical basis for the rapid and accurate monitoring of potato chlorophyll content.The main results of this study are as follows:1.PROSAIL model can well verify the spectral reflectance of potato at various growth stages.The highest coefficient of determination R~2between simulated reflectance and measured reflectance is 0.99,the root mean square error RMSE(mg/g)is 0.01,and the relative error RE is 6.55.There is no linear relationship between the chlorophyll content predicted by the look-up table and the measured chlorophyll content of potato,so the potato chlorophyll content cannot be estimated.2.The spectral index band optimization results based on measured spectral data and PROSAIL model simulation data show that the sensitive bands of optimized spectral index mainly focus on green light,yellow light and red light range.The field validation results showed that the growth period had a significant impact on the estimation accuracy of the optimized spectral index,and the correlation between the optimized spectral index and potato chlorophyll content was poor at the seedling stage,while the prediction ability was significantly improved at the tuber formation stage and starch accumulation stage.Through the comparison,based on the measured data of spectral index BNI estimate model accuracy is higher,have good linear fitting effect,model validation decision coefficient(R~2)is 0.65,the measured spectral data optimization BNI spectral indices and canopy cover of potato tuber expansion stage and starch accumulation stage,the optimal relevance between chlorophyll content,R~2was 0.76,showing a significant linear fitting relationship.3.The modeling results of machine learning showed that the estimation ability of the linear partial least squares method was poor compared with that of the nonlinear model.As for the nonlinear estimation model,the estimation accuracy of random forest is higher than that of support vector machine.The random forest model is less affected by year and growth period,and the prediction ability of optimized spectral index as input variable is the best.The coefficient of determination R~2,root mean square error RMSE(mg/g)and relative error RE(%)of the validation set were 0.73,1.56 and 1.40,respectively.In this study,the reflectance data of potato canopy were obtained by PROSAIL model simulation and spectral field measurement,and the look-up table method was not effective in estimating potato chlorophyll content.Through the measured spectral data and the PROSAIL model simulation database,the central bands of different calculated spectral indices can be optimized to estimate potato chlorophyll content.And spectral index and the spectral reflectance as variables to machine learning model,by comparison,the measured spectral data optimized BNI spectral indices and random forest modeling optimal correlation with potato,chlorophyll content and the reliability of the estimated model and the accuracy is better,to estimate the universality of potato chlorophyll content provides a theoretical support.