Research On Monitoring The Growth Of Oilseed Rape At Seedling Stage Based On Multi-Source And Multi-Scale Remote Sensing Analysis

The study focused on the seedling stage of oilseed rape,which has a decisive impact on growth,development,and yield formation.The key technologies of low-altitude remote sensing by unmanned aerial vehicles and high-altitude remote sensing by satellites were employed to obtain multi-source remote sensing data of oilseed rape at four time points during the seedling stage,as well as multi-source non-remote sensing data from meteorological stations and physiological indicators of oilseed rape growth.Combined with image processing algorithms,multi-source data fusion algorithms,and machine learning algorithms,the study established four remote sensing quantitative monitoring models for oilseed rape leaf area index(LAI),aboveground biomass(AGB),leaf nitrogen content(LNC),and chlorophyll content(SPAD).Through model calibration strategies,the established oilseed rape models in the field were combined with satellite remote sensing data to achieve the establishment of a regional oilseed growth monitoring and prediction models.The main research contents are as follows:(1)An improved image fusion algorithm RMGF was proposed by introducing the reptile optimization algorithm(RSA)into the multiscale decomposition image fusion algorithm based on the guided filtering(GF)algorithm.The improved image fusion algorithm RMGF was compared with the 6 image fusion algorithm in four benchmarks based on satellite remote sensing images,and the results showed that the proposed algorithm can improve the resolution of MS and effectively retain the spectral information of MS images and the spatial texture information of PAN images in the fusion of low-resolution multispectral(MS)images and high-resolution panchromatic images(PAN)images.The RMGF was used for UAV multi-source remote sensing data to obtained high-resolution multispectral images,and achieve the fusion of UAV multi-source remote sensing data,providing more accurate data for the subsequent establishment of a model for monitoring the growth of oilseed rape in large fields.(2)The study proposed four model frameworks based on multimodal data,and four machine learning models(SVR,PLS,BPNN and NMR)were used to compare the monitoring models of physiological indicators of oilseed rape growth established by the four model frameworks.The results showed that model framework based on the fusion of multi-source remote sensing/non-remote sensing data had the highest average accuracy(R2=0.7454),which was 28%,14.6% and 3.7% higher than the average accuracy of the other three model frameworks,respectively,while the accuracy of the oilseed rape growth model established by SVR is generally higher under each multimodal model framework.In addition,model framework added meteorological data to improve the robustness of the monitoring model for multi-scenario applications and provided theoretical support for extrapolating the model to large scales.The oilseed rape monitoring model established by SVR was used for the whole field,breaking through the limitation of experimental plots to achieve remote sensing inversion of LAI,AGB,LNC and SPAD for oilseed rape in large fields,and the inversion map of oilseed rape growth was drawn.(3)To realize multi-scale oilseed rape growth monitoring,the study obtained multi-source remote sensing data and oilseed rape growth area vector data of Jingzhou region through GEE cloud remote sensing computing platform,among which the accuracy of oilseed rape growth area vector data reached 96.88%,which was used to accurately obtain multi-source data of oilseed rape at large scale.Finally,the oilseed rape growth monitoring model based on multi-source data fusion established in large fields was extended to large scales to obtain the inverse map of oilseed rape growth monitoring in Jingzhou at large scales,which provided technical support for the staff of Jingzhou government and Agricultural Bureau to make decisions related to agricultural production.