Research And Demonstration Of Multi-temporal Images Extraction For Crop Planting Area

As a mainstay of agricultural development,crop production involves social stability and food security.Due to the changes in climate,environment and other issues,various natural disasters have occurred frequently,which not only severely affected the crop production,but also caused economic losses of farmers.In order to ensure the interests of farmers and agricultural production,China has gradually expanded the scope of policy-based agricultural insurance pilots and supported the development of policy insurance for major crop.Therefore,research on efficiently obtaining crop planting area and spatial distribution by using high-resolution satellite remote sensing images,which could provide reliable data for government to supervise underwriting financial funds of agricultural insurance.First of all,the thesis uses multi-temporal Sentinel-2 and Google Earth images as the data source,and adopts decision tree classification and logical operation.Then,the test by analyzing the changes of the spectral index of winter wheat and paddy in their respective phenology-based periods,to establish the decision tree model of winter wheat and phenology-based algorithm of paddy.Eventually,the methods precisely extract planting area and spatial distribution of the two crops.The results show that the multi-temporal images could comprehensively reflect the characteristics of the crop throughout the growth cycle,while avoiding the extraction error caused by the different sowing stage,so that the crops could be recognized accurately.This thesis takes winter wheat and paddy planted in Xiangfu District,Kaifeng,Henan Province as the research objects,and adopts the winter wheat decision tree model and paddy phenology-based algorithm devised in this test to extract the two crops planting area.The results are obtained through experiments and assessment of accuracy:The extraction result of winter wheat planting area was 62.85 thousand hectares,the extraction accuracy was 97.51%,and the overall accuracy was 97.39%;the extraction result of paddy planting area was 4.47 thousand hectares,the extraction accuracy was96.75%,and the overall accuracy was 97.81%.Compared with the data published by the government statistics,the error of the two crops extraction results is less than 5%,which meets the accuracy requirements of the Henan Provincial Finance Department.Sentinel-2 images are superior data source for crop classification and monitoring because of its high-resolution,economical and accessible performance.The two methods are intelligible and tractable to implement on computer,which is feasible to the business application of agricultural insurance underwriting supervision.