Study On Plant Growth Monitoring Based On Multi-Source Data Form UAV And Satellite Remote Sensing

With the increase of the world’s population,the reduction of cultivated land,and the decline of the quality of cultivated land,global food security is facing enormous challenges.Plant growth monitoring based on remote sensing images is an effective means to promote the development of agricultural science and technology,achieve precision agriculture and crop improvement,and ensure food security.In this study,we take wheat and maize as the main research targets and focus on the unmanned aerial vehicle(UAV)based monitoring methods regarding plant geometric characteristics parameters(plant height,canopy coverage,leaf area index),stress resistance parameters(lodging),biochemical parameters(leaf chlorophyll content),phenological stage(heading date)and production performance parameters(aboveground biomass).And the fine classification of crops at small and medium scales was achieved by the fusion of UAV and satellite data.On this basis,the plant height monitoring model was extended to multi-crop growth monitoring in large area field trials.It provided the theoretical basis and technical support for high-throughput plant growth monitoring for precision agriculture and crop improvement.The main contents and conclusions of this study are as follows:(1)A method for continuous plant growth simulation was proposed.The plant height was estimated by using percentile calculation and digital surface model of the early and growth stages of the plant.Plant canopy coverage was estimated using 9 color features extracted from UAV images combined with support vector machine.The Plant’s continuous growth monitoring results were also obtained by fusing multiple time points observation data with plant growth curves.(2)A heading date estimation method based on the fusion of UAV data and plant growth curves was proposed.The wheat heading date was estimated by calculating wheat plant height growth acceleration based on the growth curve.Compared with the previous methods of phenology modeling and field IOT image detection,it neither required environmental data of several years and plant parameters for model parameter adjustment,nor required high-density data calculation,and had the advantages of high throughput and low cost.The proposed model achieved an estimation accuracy of mean absolute error of2.81 days and root mean square error of 3.49 days on 72 wheat experimental fields with different sowing dates,densities,and varieties.(3)A method for precise monitoring of plant lodging by combining plant height and spectral features was proposed.The combination of spectral data and elevation data significantly improved the recognition effect and accuracy of plant lodging identification.The accuracy of the proposed method for lodging identification was 98.41%,and the kappa coefficient was 0.97.Based on the lodging identification,the accuracy of the proposed method for estimating wheat heading date could be further improved.(4)The estimation models of maize leaf area index,leaf chlorophyll content and aboveground biomass based on multi-feature fusion were constructed.The estimation methods were studied at multigrowth stages with 7 groups of feature sets of texture features,vegetation indices and plant height indexes and 6 machine learning methods.The results showed that texture features were effective parameters for estimating leaf area index,ear leaf chlorophyll content,and aboveground biomass.Plant height metrics were not suitable for the estimation of leaf area index,leaf chlorophyll content,and aboveground biomass after the silking stage.Combining multiple features(texture features,vegetation indices,plant height metrics)could obtain relatively stable performances at multiple growth stages.(5)A method for fine classification of small-and medium-scale crops by fusing UAV data and Sentinel data was proposed,and the effect of spatial resolution of UAV images on classification by fusion data was explored.The fusion of UAV data and satellite data could obtain high spatial resolution multispectral data.The crop classification accuracy was significantly improved(by 10%–16%)using the fused data compared with the original UAV data and satellite data.When fusing UAV data with satellite data,it was not the case that higher spatial resolution derived better classification results.By comparing the fusion of UAV data of different resolutions and Sentinel-2A data,it showed that when the UAV data is 0.1 meters,the fusion with Sentinel data achieved the best classification accuracy.The above conclusions were proved by repeated experiments in the verification area.(6)The combination of fine crop distribution maps and crop growth monitoring algorithms enabled the simultaneous monitoring of the multi-crop field scale in the large area experimental field.Taking plant height monitoring as an example,in the case of a large regional experimental field with multiple crops planted irregularly,the plot-scale crop distribution map was used to substitute the plant height monitoring model into the large regional experimental field,and automated simultaneous monitoring of maize,rice,soybean,and buckwheat plant heights at the plot scale was achieved.