| Moisture content is an important indicator to judge the maturity and harvest time of wheat.Accurate acquisition of wheat moisture content in large-area is helpful to effectively plan the harvesting sequence,guide the reasonable scheduling of the operation path of the combine harvester,and ensure the accurate harvesting of wheat at maturity.Satellite remote sensing has wide detection area,multiple ways to obtain information and multi-scale.It can obtain wheat moisture content information in large-area through surface spectral reflection.However,due to the influence of atmospheric environment,it is easy to appear that the quality of satellite images is poor to extract effective spectral information in critical growing period.Satellite remote has high accuracy,convenience,flexibility,high applicability and more available datas.The fusion of multi-source remote sensing information from satellite and earth,the ground remote sensing can supplement the missing spectral information caused by the shortcomings of satellite remote sensing,and the satellite remote sensing can also make up for the shortcomings that the research results and methods of ground remote sensing which are limited to small-area applications,thus forming the information synergy of satellite-ground linkage and insufficient ground compensation in the sky.In this paper,based on the high time series remote sensing datas obtained from the fusion of satellite and local multi-source remote sensing information,a prediction model with NDVI,NDWI and EVI vegetation indices as feature sets and wheat moisture content as target variables was constructed.The on-line detection technology of wheat moisture content of combine harvester was used to verify the timely harvest and its application in guiding the rational planning of harvest sequence and path scheduling of combine harvester.The main research contents of this paper include the following aspects:1.The method of multi-source remote sensing information fusion was proposed,and the satellites Landsat-8 and Sentinel-2 were fused in time and space.In view of the limited information obtained from single satellite remote sensing data,this paper puts forward STARFM(Spatial and Temporal Adaptive Reflection Fusion Model)algorithm to fuse the remote sensing information of multi-source satellites Landsat-8 and Sentinel-2,which could effectively improve the accuracy and time series length of extracting the vegetation index information in wheat filling maturity,and obtain high-time series Landsat-8 remote sensing datas.A prediction model with NDVI,NDWI and EVI vegetation indices as feature sets and wheat moisture content as target variables was established by partial least squares method,and the test set RMSE=2.399;As for the correlation between the actual value and the predicted value of the sample moisture content,the determination coefficient of the training set R~2=0.878.2.A prediction model of wheat moisture content based on ground remote sensing hyperspectral information was constructed.The feasibility analysis of predicting wheat moisture content by hyperspectral information of wheat husk was carried out,and the best regression model of wheat and husk moisture content was determined by correlation analysis.Competitive adaptive reweighting method(CARS)was used to optimize the spectrum sensitive band of wheat husk moisture content,and partial least squares regression method was used to construct wheat husk moisture content prediction model.The root mean square error of prediction set RMSEP=2.676,the determination coefficient of prediction set R~2p=0.945,and the relative analysis error RPD=3.362.3.Based on multi-source satellites Landsat-8 and Sentinel-2,the reflectivity of Ranbo(Blue),Red Wave(Red),Near Infrared Wave(NIR)and Short Wave Infrared Wave(SWIR)related to the vegetation indices NDVI,NDWI and EVI of the ground features were analyzed,the equivalent reflectivity and average equivalent reflectivity formula were adopted to complete the conversion between the spectral reflectivity of ground objects and the remote sensing spectral reflectivity of multi-source satellites,the high-time series vegetation index NDVI,NDWI and EVI curves in the filling maturity stage based on satellite-ground remote sensing information fusion were constructed,a prediction model of wheat moisture content based on random forest algorithm was constructed,with the determination coefficient R~2=0.963 and the maximum relative error of 0.159.In order to realize the application of multi-source information fusion prediction model of wheat moisture content in field machinery production,based on satellite-ground multi-source remote sensing fusion information and wheat moisture content prediction model,a prediction model of wheat harvest time in the field was established,with the difference of wheat moisture content as feature sets and the difference of days as dependent variable.Through data verification,the absolute error range of moisture content was 0.11%～0.34%,which met the requirement of prediction accuracy.4.The on-line detection system of grain water content in combine harvester was designed.COMSOL finite element simulation software was used to optimize the structural parameters of the ipsilateral arc capacitive moisture content sensor,and the parameters of the number of plates,plate size and plate spacing were determined.Based on the working principle and structure of moisture content sensor,the experimental study on the influence of four factors,temperature-frequency-capacitance-bulk density on moisture content detection of six different wheat varieties was carried out.The mathematical model of the relationship between moisture content and temperature,frequency,capacitance and bulk density was established by using the least square regression method,support vector regression method and BP neural network method,and the comparative analysis and significance test was carried out.The determination coefficients R~2of training set and test set of wheat mositure content prediction model based on BP neural network was 0.896 and 0.893 respectively,and the root mean square error RMSE was 1.317 and1.342 respectively,compared with the other two models,its stability and prediction ability was best.5.The test bed dynamic performance test and field comprehensive test were carried out.An on-line test bed for water content detection of wheat combine was established,and a model for wheat water content detection was constructed.In the field wheat water content online detection test,the stability and repeatability test of the wheat water content index of the online detection system were carried out,and the detection accuracy and repeatability of the system were analyzed.The maximum relative error of the detection result of the system water content was 1.67%,the detection stability was good,and the detection error was less than 5%,which met the application requirements.In terms of the prediction verification and application of multi-source remote sensing information for wheat water content,the average relative error of the predicted and actual wheat water content was 0.38%.According to the prediction model of wheat harvest time based on satellite-based remote sensing fusion information and wheat predicted water content information,the wheat water content of field combine was detected online at the predicted harvest time by the field combine.Through comparison and analysis of the detected value of the system water content and predicted value of predicted water content at the predicted harvest time,the results tend to be consistent with the absolute error of 0.36%.The application effect of multi-source information fusion of wheat water content prediction in mechanized wheat harvest production was verified. |
PDF Full Text Request