Automatic interpretation method for winter wheat based on deformable fully convolutional neural network

Abstract

The present invention discloses an automatic interpretation method for winter wheat based on a deformable fully convolutional neural network (FCN). The method includes the following steps: step: using a region corresponding to a winter wheat planting area as a research region, and obtaining a high-resolution remote sensing image of the research region; step: preprocessing the obtained remote sensing image: extracting geometric changes of sizes, fields of view, postures, and partial deformation of winter wheat objects in different spatial positions in the high-resolution image as sample data; step: establishing the deformable FCN; step: inputting the sample data to the deformable FCN to implement training of the deformable FCN; and step: inputting a to-be-recognized remote sensing image of winter wheat to the trained deformable FCN, and outputting a prediction graph of a winter wheat planting area in a to-be-recognized region.