Study On Plastic Greenhouse Extraction From Multi-source Remote Sensing Imagery

In agricultural industry of China,the proportion of plastic greenhouses is increasing.The collection of statistical data is based on the way of reporting at the grassroots units.This method is insufficient to grasp the dynamic performance of reported quantities.Therefore,remote sensing methods are used to obtain timely and accurate information on the area and spatial distribution of plastic greenhouse.It is of great significance for the effective management and decision-making of relevant agricultural sectors.However,our country has a vast territory and plastic greenhouse has a wide distribution.If the medium-resolution image is directly used to obtain the area and spatial distribution information of the plastic greenhouse,the extraction accuracy is limited and the workload is high.If high-resolution satellite images are directly used to obtain information on the area and spatial distribution of plastic greenhouses,they will encounter problems such as large amounts of data,high costs,large workloads,and difficulties in ensuring data sources.To solve these problems,this thesis first starts from the perspective of geography and comprehensively considers a variety of limiting factors in the construction of plastic greenhouses and excludes the areas which is not suitable for the construction of plastic greenhouses.Taking Shandong Province as an example,the area with the potential of building plastic greenhouse is obtained by spatial analysis,and the scope of information extraction is reduced.Then,the distribution area of plastic greenhouse is extracted from the medium resolution Landsat8 OLI remote sensing image by object-oriented method.On this basis,the information of plastic greenhouse is precisely extracted based on the high-resolution satellite GF-1 remote sensing image of China.The research results show that:(1)From the point of view of geography,the unsuitable district of plastic greenhouse can be removed by using landform,climate,soil,land use types,which can effectively reduce the scope of using remote sensing to study the information extraction of plastic greenhouse.(2)Based on three indexes of potential segmentation error(PSE),segmentation strength(NSR)and Euclidean distance(ED),the optimal segmentation parameter combination on Landsat8 OLI and GF-1 remote sensing images for specific ground objects of plastic greenhouse can be determined.(3)Using Random Forest(RF)algorithm to analyze 47 spectral features,8 texture features,11 shape features and 6 thematic index features,it is concluded that the relationship between the number of features and the classification accuracy is gradually increasing and then decreasing.This method can effectively delete redundant and unrelated features and improve the performance of classifier.(4)Based on Landsat8 OLI remote sensing image,the distribution area of plastic greenhouse is extracted.The support vector machine,CART decision tree and random forest are compared.According to the principle of the minimum area of the leaking area,the support vector machine method presents a great advantage.Based on the GF-1 remote sensing image,the information of plastic greenhouse is extracted accurately.It is concluded that the random forest method has a strong recognition ability for plastic greenhouse.Through comparison and analysis with the results of artificial visual interpretation,it is concluded that the method can accurately extract the information of plastic greenhouse,and provide effective technical means for the information extraction of plastic greenhouse in the future.