Study On Agricultural Greenhouse Extraction Using CART Decision Tree Based On Sentinel

Agricultural greenhouses have significant social and economic benefits.They cover a wide variety of crops,including vegetables,fruits,food crops,flowers,saplings,etc.,for the safe supply of food and cash crops in China,farmers getting rich and well-off,and development of facilities agriculture Provided modern agricultural technology and equipment support.The timely and accurate acquisition of agricultural greenhouse coverage area and geographical distribution information is of great significance to increase farmers’ income,national agricultural overall planning,and agricultural output value estimation.Remote sensing,as a resource survey method with wide coverage,fast statistical speed and able to provide objective facts,is highly targeted in the extraction of agricultural greenhouse information.This paper takes Zhuanghe as a research area,uses Sentinel-2 remote sensing images with high spatial resolution and rich texture information as data sources,and comprehensively utilizes the characteristics of spectrum,index,texture,etc.to design CART based on multi-feature fusion data set Decision tree,and extracted the spatial distribution and quantity information of agricultural greenhouses in the study area,the conclusion is as follows:⑴The agricultural greenhouse area in Zhuanghe City is 2343.76 hectares,accounting for 2% of the total cultivated land area in Zhuanghe City.⑵The CART decision tree constructed by adding the remote sensing index and texture feature variables has a significant impact on the accuracy of agricultural greenhouse extraction.In the mining information of CART decision tree,the addition of remote sensing index can significantly improve the overall classification accuracy of remote sensing images by 5.5374%;texture features can also improve the overall classification accuracy of images by 5.6460%;both are added at the same time.The total classification accuracy is improved by 7.3833%.⑶The three remote sensing indexes RVI,EWI and MNDBI are the effective variables for CART to extract the greenhouse.In the process of selecting remote sensing feature index to participate in the construction of classified multi-feature data sets,through the analysis of nine indexes of RVI,NDVI,MSAVI,EWI,MNDWI,MNDBI,NDBI and RISI,finally selected three remote sensing of RVI,EWI and MNDBI Index,these three indexes participate in CART decision tree mining best.⑷The optimal window size for texture feature calculation is 11 × 11.The deciding factor that affects the extraction accuracy of texture features in agricultural greenhouses in the Sentinel-2 image is the size of the sliding window.The sliding windows of 5 × 5,7 × 7,9 × 9 and 11 × 11 are selected for CART decision tree classification In comparison,the sliding window size of the best texture feature is found to be 11 × 11.