Research On Crop Classification Methods By Multisource Remote Sensing In Jiutai Area,Changchun

The spatial distribution of crop species is the basic premise of agricultural remote sensing monitoring,and the rapid development of various imaging remote sensing technologies provides the possibility to acquire a large-scale and multi-period crop planting distribution.At present,optical remote sensing data are mainly used to monitor crop planting distribution by remote sensing.However,the frequent cloud-rainy weather in the key cropping period of crops makes radar remote sensing technology with strong penetrating power provide strong basic data support for remote sensing identification of crops.For two different data sources,the Landsat8,Gaofen-1 multi-spectral data in optical remote sensing and Sentinel-1,RADARSAT-2 data in radar remote sensing were taken as examples,by exploring the different phases and related features of remote sensing images in Jiutai District of Changchun on the classification accuracy of crops,then the optimal time phase,feature band set and classification method suitable for crop classification were mined,and the research of agricultural classification monitoring by optical remote sensing and radar remote sensing was carried out.Based on remote sensing classification methods for crops,the methods for remote sensing classification of crops at home and abroad were collected and summarized.The remote sensing classification methods for crops were summarized from two kinds of data,optical remote sensing and radar remote sensing.Summarizing the natural situation and crop planting structure of the research area,collecting and processing Landsat8,Gaofen-1 optical remote sensing data and Sentinel-1 and RADARSAT-2 radar remote sensing data,could provide basic reflectance data and backscatter data for the research of crop remote sensing classification.Taking Landsat8 and Gaofen-1 PMS data as examples,the classification methods of optical remote sensing crops were studied.The decision tree was selected to identify crops based on the differences of various features in Landsat8 images in the research area.J-M distance was used to compare the classification separability of single-phase and single-phase PMS in Gaofen-1 and the optimal phase was selected.And the maximum likelihood method,decision tree and object-oriented classification method were used to compare the classification accuracy of remote sensing crops.The results showed that Landsat8 images at maturity were beneficial to crop classification,and the classification accuracy was up to 88.39%.The mid-July Gaofen-1 images of Jiutai district have good separability for soybeans.Compared with the decision tree classification,the object-oriented classification can better identify the boundaries of crop plots.The overall accuracy and Kappa coefficient are 94.68% and 0.93,respectively,which is more suitable for high-resolution remote sensing crop classification.Taking Sentinel-1 dual polarization data and RADARSAT-2 full polarization data as examples,the classification method of radar remote sensing crops was studied.Based on the temporal backscatter coefficient differences of various features on the Sentinel-1 dual-polarized data,a decision tree was established to identify corn and rice on multi-temporal Sentinel-1 data.Based on the differences of backscattering characteristics of all kinds of features on RADARSAT-2 full-polarization data,decision tree and object-oriented classification were adopted to classify crops respectively.The results showed that multi-temporal Sentinel-1 data were feasible for the classification of crops in Jiutai,and the overall accuracy is 80.57%.But Landsat8 images have a better classification effect;compared with decision tree classification,object-oriented classification can better reduce the influence of radar noise on the internal heterogeneity of crop plots,and its classification effect is better.The overall accuracy and Kappa coefficient were 81.63% and 0.72,respectively.