| Maize is one of the main grain reserve crops.Timely and effective monitoring of maize growth plays an important role in national economic development and social stability.Canopy chlorophyll is an important indicator of maize growth and nutritional status,and its content directly affects maize yield and grain quality.Therefore,it is particularly important to monitor the canopy chlorophyll density(CCD)of maize canopy quickly and accurately.The traditional agricultural monitoring method need destructive sampling,which is time-consuming and laborious,and contradicts the demand of modern agriculture to save costs and increase efficiency.In recent years,unmanned aerial vehicle(UAV)remote sensing and image processing technology has developed rapidly,and a series of crop CCD estimation methods have been proposed.However,the CCD estimation model based on the single scale feature of the UAV image is more susceptible to the influence of the canopy structure,thus ignoring the contribution of the leaf chlorophyll density(LCD)information,resulting in the model estimation results inconsistent with the objective facts.In addition,the blindness and randomness of UAV flight time make the observation geometry of light source and sensor unstable,which may also affect the estimation of CCD.In view of the above problems,this research focuses on the two key tasks:accurate estimation of CCD based on multi-scale image feature fusion and analysis of the impact of observation time on CCD estimation model.The performance of random forest(RF),support vector machine(SVM)and multiple linear regression(MLR)algorithms are compared and the best model was selected based on the accuracy and efficiency of LCD,leaf area index(LAI)and CCD estimation,;Based on the best model,the best feature dataset estimated by CCD was selected from four feature datasets(spectral feature dataset,texture feature dataset,wavelet feature dataset and fusion feature dataset),and then the best feature type and feature quantity were determined.The precision of CCD indirect estimation and direct estimation models were compared,and the CCD mapping of UAV scale at the whole growth stage of maize was realized;Based on the multi-frequency observation experiment data of UAV on the same day and the simulation data of PROSAIL model,the change rule of typical vegetation indices at different times of the day and its impact on the CCD estimation results were analyzed.The main conclusions are as follows:(1)RF model performs best in LCD,LAI and CCD estimation(R_a~2 values is 0.89,0.97,and 0.96 respectively);The estimation results of LCD,LAI,and CCD based on UAV image fusion feature dataset are the best(R~2 is 0.91,0.97,and 0.97 respectively,RMSE is 6.59μg/cm~2,0.35,and 24.85μg/cm~2 respectively),the best feature number are the top 24 of the fusion feature dataset,and the feature types include vegetation indices,texture and wavelet,accounting for 54%,29%and 17%respectively;Compared with CCD estimated by indirect method(R~2=0.96,RMSE=26.85μg/cm~2),the result of CCD directly estimated from fusion feature dataset is slightly better.The results show that multi-scale image feature fusion can capture the canopy chlorophyll information of maize more accurately,which provides a new technical support for the accurate acquisition of CCD in precision agriculture.(2)In the same day,the vegetation indices values of UAV with good correlation with the measured CCD change with time,and the lower the value is the closer to noon,while the vegetation indices values simulated by PROSAIL at different times of the day have little difference;The correlation between the same vegetation indices of the UAV obtained at different observation times on the same day and the measured CCD is quite different,and the difference between different growth stages and different indices is inconsistent;The correlation between the same simulated vegetation indices and CCD is not significant at different times on the same day;CCD estimation models based on UAV data at different observation times had significant differences in estimation results,R~2 is 0.53 at the lowest and R~2 is 0.80 at the highest.The results show that in the traditional spectral data acquisition time range(between 10:00 am.and 14:00 pm.),the UAV image acquisition time still has an impact on the maize CCD estimation,and the closer to the noon time,the higher the estimation accuracy.The research results can provide technical reference for the optimization of field UAV experiment and the accurate estimation of crop CCD.Figure[23]Table[11]Reference[105]... |
PDF Full Text Request