Research On Information Extraction Method Of Crop Remote Sensing Image

Crops such as rice are important food crops in the world.Under the guidance of national policies,ensuring food production has become the focus of the work of local governments.The measurement of crop area is of great significance.The traditional measurement method mainly relies on manual completion,combining visual interpretation and classification and field sampling research to realize area estimation.This method requires a lot of labor,and the calculation results are inefficient,long-term,easily disturbed,and cannot achieve accurate agricultural conditions.The need for real-time control of information.With the development of science and technology,the traditional method of manually measuring the area of crop arable land in my country has been greatly improved.Therefore,in this thesis,remote sensing images of crop cultivation areas are screened,and the semantic segmentation model and object-oriented classification network are used to extract crop planting information quickly.The main research contents and results are as follows:(1)The remote sensing image acquisition method and the preprocessing method of crop remote sensing image are studied,and the crop phenological characteristics and the characteristics of shape,spectrum,and texture are analyzed by using the images of crops in different growth periods.Through the research on the phenological characteristics of crops and the image characteristics of the planting area,the existing survey vectors are used to label and generate accurate sample data sets,which lays the foundation for subsequent experiments.(2)Use two different methods of deep learning-based semantic segmentation model and object-oriented deep learning to extract information on crops.The semantic segmentation model is used for large-scale crop extraction,and the experimental accuracy rate reaches 88%.Build an object convolutional neural network,establish a complete data set,optimize the generated land cover classification results,and improve the accuracy of crop extraction.(3)According to the land cover classification results,a detection method of cultivated land change based on object convolutional neural network is proposed,and the overall classification accuracy reaches 90%.It can analyze the changes in the arable land planting area of crops and help the government to understand the agricultural situation information in time.