Key Technology Research On Chlorophyll Content Monitoring Of Fruit Trees

Plant phenotypes are the results of interactions between plant genotypes and complex environments,including but not limited to plant morphology,nutritional elements,plant pigments and other parameters.Chlorophyll is the catalyst of plant photosynthesis.The research of chlorophyll is an important branch of plant phenotypic research.Chlorophyll content can reflect the photosynthetic capacity,the growth status and the deficiency of nutrients of plants.In the past,most of the studies on chlorophyll of fruit trees were focused on the level of single growth stage or single tree species,which is difficult to satisfy the evaluation of chlorophyll content for different tree species or different growth stages.Therefore,non-destructive detection of chlorophyll content in multi-growth stage and multitree species is of great significance for fruit tree growth monitoring,and helpful in fertilization and irrigation management.This dissertation investigated the chlorophyll content monitoring of fruit trees at different scales and for different tree species with multispectral imaging technology.The apple trees and pear trees were selected as the research object.Different image processing techniques and mathematical modeling methods were applied for data processing.The performance of estimation models constructed by different characteristic variables were deeply discussed,which provided theoretical basis and data support for fruit tree growth monitoring and management.The main research conclusions are as follows:(1)The apple trees and pear trees were selected as the research object,and spectral images of leaves were obtained by Parrot Sequoia multi-spectral imaging sensors,canopy images were obtained by UAV(Unmanned Aerial Vehicle)and multi-spectral sensors.To effectively eliminate the distortion of leaf images,geometric correction was performed based on camera parameters and principle of spatial transformation,which made a better result than the traditional checkerboard calibration method.In the processing of leaf images,the leaf region was extracted with the methods of the combination of image registration and threshold segmentation,solved the problems of the spatial difference of spectral images in different bands and the difficulty of the segmentation of the leaf region from the background region of images in RED band.In the processing of canopy images,the NDVI(Normalized Difference Vegetation Index)map and DSM(Digital Surface Model)was adopted to effectively separate the pure canopy pixels from the mixed pixels of bare soil,shadow and weeds.(2)At the leaf scale,the research of chlorophyll content monitoring with single tree species was studied.The SPAD value monitoring models were established with vegetation index(VI),spectral features and texture features as different categories of independent variables,and the stability and prediction accuracy of univariate regression,linear regression and non-linear regression models were compared.The results showed that the univariate model based on GNDVI showed stable prediction ability,and the R2 values of estimation models research on apple tree leaves and pear tree leaves were 0.600 and 0.587,respectively.Compared with different regression methods,the linear model showed strong stability and comparable predictive ability.Among the multivariate nonlinear models,GPR model had stronger stability than SVR and CNN model.Compared with the model results of different categories of independent variables,texture features improved the prediction ability of the model to a certain extent,but the optimization effect is not obvious.(3)At the canopy scale,the research of chlorophyll content monitoring with single tree species was studied.The texture indexes(NTI-AC and NTI-PC)were established according to the texture features and canopy SPAD value,and the bivariate regression model was established by combining the texture index with vegetation index.The highest R2 value was 0.550 for the model established with GNDVI and NTI-AC in the apple tree study,and the highest R2 value was 0.675 for the model established with GRVI and NTI-PC in the pear tree study,which increased by 7.4%and 37.8%respectively compared with the GNDVI univariate model.In the multivariate models,compared with other models trained with the multiple independent variables,the model based on the combination of spectral features and texture features had higher prediction ability.The GPR model yielded R2=0.788 and RRMSE=10.11%in the apple tree study,and the Ridge model yielded R2=0.739 and RRMSE=11.90%in the pear tree study.It can be concluded that the texture features can reflect the coverage and canopy structure of fruit trees in different growth stages,which plays an important role in chlorophyll content monitoring at the canopy scale.(4)The research of chlorophyll content monitoring with mixed tree species was studied,to explore the SPAD estimation model with adaptability of different tree species.The univariate model based on GNDVI performed stable results in the leaf SPAD value estimation of mixed tree species,but the performance of the validation set for single tree species at the canopy scale was not ideal,and the prediction accuracy is low.Comparing the model results yielded on spectral features with the model results yielded on spectral and texture features,the addition of texture features significantly improved the prediction accuracy of the models at both leaf and canopy scales.From the perspective of regression methods,GPR regression model not only had strong predictive ability,but also can be effectively applied to the monitoring of chlorophyll content of single tree species.(5)The fruit tree chlorophyll content monitoring system platform was developed to realize the visualization of image processing and monitoring results,which promoted the application and promotion of fruit tree chlorophyll content monitoring model and provided technical support for fruit tree management.