Fast Extraction And Application Of Land Leveling Information Based On Multi-scale Segmentation Of GF-1

In recent years,satellite remote sensing imaging technology has developed rapidly.Remote sensing imaging has become an important source of data in the field of soil and water conservation.Soil and water conservation work often uses visual interpretation as the main manual interpretation method to extract information from the image.This information extraction method has high classification accuracy and can meet a variety of image classification needs,however,it has high requirements for the classifiers themselves,and fails in performing high precision positioning.With the increase of high-resolution remote sensing image data,the rapid and accurate classification of images is required.The traditional manual interpretation methods can no longer meet the demand for high classification accuracy and efficiency of remote sensing images.This paper took the remote sensing image of No.1 in Lanzhou New District as an example,and used the object-oriented classification method via eCognition software to extract the image information of the unused leveling land in the image.The multi-scale segmentation method was used to segment the image multiple times,and different types of land use were defined according to different view characteristics.Image classification was performed by the threshold classification method to achieve efficient and accurate extraction of unused leveling land.According to statistics,the area of unused leveling land was 85.89.44 hm~2 and other land area was 71992.19 hm2.Under the eCognition software,the Producer of the unused leveling land in Lanzhou New Area was evaluated to be 0.871,the user accuracy was 0.937,the overall accuracy was 0.9505,and the Kappa coefficient was 0.829.The accuracy evaluation results showed that the image classification results were good and the image classification stability was good.Compared with the results of traditional manual interpretation,the boundary classification of automatic image classification was more accurate,which avoided the problem of classification boundary misalignment caused by the classifier's own problems in the process of manual interpretation.In summary,the eCognition software had high extraction accuracy for unused leveling land in Lanzhou New District and could meet the requirements for image classification accuracy in image classification and extraction.Multiple fusion and re-segmentation methods made the image classification more targeted and more accurate in extracting the information of unused leveling land.The similarity between image interpretation results and traditional manual interpretation was high,and the automatic interpretation operation was simpler and the interpretation efficiency was higher.It caould meet the demand for high classification accuracy and efficiency of image information extraction.