Synergistic Integration Of Multiple Classifiers To Better Extract Parcel Of Rice Field From Multi-scale Remotely Sensed Imageries

Rice is an important food crop in China,so timely and accurate acquisition of rice planting area is of great significance to China’s food security.At present,the acquisition of rice planting area in many areas of China is mainly based on manual reporting,which is faced with problems such as high monitoring cost and relatively lagging time of survey results.In order to better solve the above problems,this thesis takes Chongming District,Shanghai as the research area,and proposes a method of integrating multi-temporal and multi-scale remote sensing images and multi-classifier to achieve the extraction and change analysis of rice planting area at the plot level.The main research contents and achievements of the thesis are as follows:1.Firstly,the data of ZY3 is segmented at multiple scales by FNEA,and the high-resolution plot boundaries are extracted.The results show that moderate shape factor weight and relatively small compactness factor weight can better preserve the characteristics of plot boundary;Different optimal segmentation scales can be set for different sizes of parcels in the study area to get better results of parcel boundary extraction.2.The pixel based rice extraction in the study area is realized by comprehensively using radar data,multi-spectral data and multi classifier integration method.According to the rice phenological characteristics,the minimum value of Sentinel-1 images located in the rice transplanting period is calculated,and the characteristic data set is obtained by screening NDVI,RVI and NDWI of Sentinel-2 image located in the rice growing period combined with J-M distance.Using Dempster Shafer evidence theory,KNN,RF and SVM are integrated into multiple classifiers to classify rice and other ground objects in the study area based on pixels,and the overall classification accuracy is 88%.3.The high-precision plot boundary data and pixel based rice extraction results are fused to extract rice planting fields,and the user accuracy and producer accuracy of the extraction results are improved to 95%.4.Using the extraction method of rice planting field proposed in this paper,the extraction of rice planting field in Chongming district from 2018 to 2020 is realized.The extraction results were compared with the statistical data to verify the accuracy of the extraction results,and the temporal and spatial variation characteristics of rice planting in this area were further analyzed.