Subfield Boundary Extraction Based On UAV High-Resolution Imagery

In recent years,our country’s agriculture has developed towards scale,precision,and modernization.A large number of large-scale agricultural management organizations such as family farms,agricultural cooperatives,and agricultural socialized service organizations have emerged all over the country.For large-scale modern agriculture,the subfield boundary is the basic unit of its agricultural production management,so the subfield boundary information is the basic information that must be mastered for the development of large-scale modern agriculture.Rapid and accurate extraction of subfield boundary information has the great significance to large-scale modern agricultural field management,agricultural planning and other operations.The traditional field boundaries are obtained by artificial visual interpretation,which is a highly subjective and time-consuming method.With the rapid development of aerial remote sensing technology,high spatial resolution remote sensing images can more accurately reflect the spatial distribution characteristics of ground objects,and the extraction of the subfield boundaries provides the effective data support.The traditional remote sensing satellite image has low spatial resolution and mixed pixels,which is not suitable for the extraction of subfield boundaries.For this reason,this paper uses U AV high-resolution remote sensing images as the main data source,and the object-oriented method based on Canny edge detection operator and the improved Gaussian mixture model method are studied.By comparing the advantages and disadvantages of the two methods in sub field boundary extraction,the subfield boundary extraction in wheat planting area is realized.The research results of this paper can provide reliable methods and models for boundary extraction from UAV remote sensingimages of the North China Plain.The research contents and results are as follows:(1)Combined with the characteristics of subfield boundary,a rule set suitable for subfield boundary extraction was constructed by object-oriented method.Based on the Canny edge detection operator,the UAV remote sensing image is segmented at multiple scales.The segmented image is combined with the characteristics of the UAV remote sensing image and the boundary characteristics of the subfield,and the optimal segmentation scale is selected and an appropriate rule set is constructed,to realize the extraction of subfield boundaries in highresolution remote sensing images.The final results show that the object-oriented method based on the Canny edge detection operator can extract the boundaries of subfields with an accuracy of more than 93%.This method can accurately and effectively extract the subfield boundary,and can provide information support for the large-scale agricultural development in wheat planting area.(2)For high-resolution UAV images,multiple parameter selections were performed through eCognition software and ESP plug-in to obtain the optimal spatial resolution and optimal segmentation scale for accurately extracting the boundaries of wheat fields,to achieve accurate boundaries of subfields.The experimental results show that the optimal spatial resolution is 15cm,the optimal segmentation scale is 45,and the shape factor and compactness factor are 0.5 and 0.6,respectively,which provide necessary basic information for the extraction of the boundaries of wheat planting areas in the North China Plain.(3)The Gaussian mixture model based on NDVI realizes the automatic extraction of subfield boundaries.Firstly,the Gaussian mixture model is constructed by bayesian decisionmaking and maximum likelihood method,the normalized vegetation index is used as the training data of the Gaussian mixture model,the initial segmentation threshold parameters are selected according to experience and statistical methods,and finally the EM algorithm is repeatedly iterated.When the algorithm converges,use the findcontours function to perform contour detection on the segmented binary image to extract the subfield boundary.The results show that the extraction accuracy of the two areas in the study area is more than 94%.This method has great potential in the extraction of the boundary of the subfield,and has a high degree of automation.