Study On Rice Fine Recognition And Plant Density Inversion Based On Polarimetric SAR

Rice is one of the most important food crops in the world.In the face of the current global epidemic,the severe form of food security,accurate and real-time monitoring of rice production is of great significance for the formulation of China’s food policy,the allocation of agricultural resources and national food security.Polarimetric SAR is a kind of microwave remote sensing technology with the ability of all day and all-weather monitoring.Its polarization information greatly enriches the monitoring ability of rice.Based on multi temporal radarsat-2 polarimetric SAR data,this paper uses machine learning method to identify japonica rice and indica rice in some areas of Huai’an City,Jiangsu Province,and retrieve the plant density of their key phenological stages.The main research contents and results are as follows:(1)The effects of single temporal data set and multi temporal data set on the recognition accuracy of Japonica and indica rice in the study area were compared.The role of polarization decomposition parameters in the process of rice recognition was explored.The advantages and disadvantages of random forest and BP neural network models applied to rice recognition were also judged.The results show that the optimal model for identification of Japonica and indica rice can be constructed by using multi temporal polarization characteristic parameter data set with polarization decomposition parameters and random forest method.The accuracy of model verification is completely correct,and the mapping accuracy and user accuracy of Japonica and indica rice are more than 90%.(2)The japonica rice,indica rice and the overall recognition accuracy of different time phases and different types of polarization characteristic parameters in the study area were analyzed.The results show that the recognition accuracy of Japonica rice,indica rice and other features in the study area mainly depends on the representation ability of the basic data set for the unique structural characteristics of paddy field.The more the data set can reflect the structural characteristics of paddy field,the higher the recognition accuracy of the model for the two kinds of rice and other features in the study area.(3)Taking into account the differences between Japonica Rice and indica rice in types and planting patterns,two inversion models of rice plant density were constructed.For indica rice,with the increase of plant density,the weight of polarization decomposition parameters in the inversion model also increases,and the scattering component changes from surface scattering,secondary scattering to volume scattering.For japonica rice,the scattering components of polarization characteristic parameters significantly related to the plant density of Japonica Rice on September 7 were mainly realized as surface scattering and secondary scattering due to the planting mode,which was just opposite to that of Indica Rice in the same period.After analyzing the polarization characteristic parameters of three key phenological stages of rice,using the polarization characteristic parameters and rice plant density,the inversion model of plant density of Indica and japonica rice was constructed through elastic network,and the precision of inversion model of plant density of indica rice at seedling stage,tillering stage,heading and flowering stage was 0.959,0.961,0.976,respectively.The precision of plant density inversion model at heading and flowering stage of Japonica rice was 0.948.The above research results show that the full polarimetric SAR data can well realize the fine identification of rice and the inversion of plant density.This paper explores the potential application ability of radar parameters in rice monitoring and expands the inversion method of rice plant density.