Optimal scale extraction of farmland in coal mining areas with high groundwater levels based on visible light images from an unmanned aerial vehicle (UAV)

The accurate acquisition of information concerning farmland in coal mining areas with high groundwater levels can provide a basis for land dynamic monitoring and protection. In this study, visible light images from an unmanned aerial vehicle (UAV) were used as the data source, from which farmland located in the coal mining areas with high groundwater levels were extracted. Based on the optimal scale for image segmentation, which was determined to be 44, farmland was extracted using a sample-based, object-oriented extraction method and a feature combination-based hierarchical extraction method. The results showed that the Kappa coefficient of the latter was 0.87, the correct rate was 88%, the commission was 24%, and the omission was 12%; all of these were better than the corresponding results obtained using the sample-based, object-oriented extraction method. The accuracy of the hierarchical extraction method was verified using the images of the verification area. For these images, the Kappa coefficient of the feature combination-based hierarchical method was 0.96, the correct rate was 95%, the commission was 20%, and the omission was 5%; these were also better than the corresponding values obtained using sample-based, object-oriented extraction. Therefore, this study demonstrates that at a segmentation scale of 44, the hierarchical extraction method based on feature combination not only accurately extract farmland information from the mining area but also have better extraction accuracy than the traditional extraction method. This study thus provides a method and reference basis for the accurate extraction of information concerning ground objects in coal mining areas.