Multi-scale Rehabilitated Vegetation Classification From Remote Sensed Data In Coal Mining Site

The development of high spatial resolution satellite broadens the application of remote sensing technology.Traditional pixel-based methods are not suitable for fine vegetation classification,while object-based classification method can avoid complex spectral character with the texture,spectrum and geometrical characteristics all being considered.A multi-level structure of forest,shrubby and grass generated at west dump in coal mining site after years of reclamation.High spatial resolution images combined with object-based classification have the potential to reflect ecological factors so as to complete comprehensive and dynamic analysis for mining environment on the basis of information extraction.The results of object-based classification can be used for remote sensing investigation for mine environment.The west dump of Pingshuo surface coal mining site was selected as study area.The Worldview-2 imagery with high spatial resolution was used for Object-based classification to support rehabilitation of vegetation.First of all,local variance measure,Euclidean distance 2 index and segmentation error methods were chosen for selecting the optimal segmentation scale.The similarity index was applied to evaluate the accuracy of classification results of each method.The results show that the Euclidean distance 2 index significantly improves the classification precision of object-based image analysis with both arithmetic and geometric discrepancies being considered.Secondly,segments based on the optimal scale and characteristics such as brightness values,shape index was adopted to complete the coarse vegetation classification.Multiple image layers and the results of reclamation vegetation characteristics were generated to establish membership function classification rule set for the fine vegetation classification.Finally,confusion matrix was used for accuracy assessment and the overall accuracy arrived at 82.89%.Spectral characteristics mixed with vegetation index can distinguish woodland from other land features effectively.Moderate coverage grassland has similar characteristics with grass-shrub,so the phenomenon of misclassification led to the lowest accuracy.With abundant spectral and texture characteristics,high spatial resolution Worldview-2 imagery can reflect the spatial information of different vegetation community.Object-based classification method makes full use of object information using high spatial resolution images,which can provide higher accuracy results.Thus,multi-level object-based classification rules and technical processes can be further used for mining environment monitoring and ecological function restoring.