Classification And Inversion Methods Research In Industrial And Mining Reclamation Area Based Remote Sensing Data

Land reclamation and ecological reconstruction were very important guarantee for coordination of mining resources and protection of land resources.It was also important measure to promote the construction of ecological civilization.For the better construction of ecological civilization and protection of land resource form three dimensions of quantity,quality and ecology.It was of great significance to use remote sensing technology to help land reclamation "quantity" survey(land use and cover classification)and "quality" monitor(soil components retrieval).However,in the industrial and mining land reclamation area,the strong topographic relief,the diversity,breakage,mixed distribution and scattered layout of the surface features and other factors constitute the difficulties for remote sensing image classification strategy selection and soil components inversion research.Therefore,this paper carried out a study on the classification and inversion methods of the mine reclamation area based on remote sensing data.Based on the GF-1 and landsat-8 OLI remote sensing data,the image classification research under different resolutions were carried out by using grid search(GS)for random forest(RF)classification,object-oriented analysis(OBIA),multiple endmember spectral mixture analysis(MESMA)and other methods.Based on the ASD FieldSpec 4 hyperspectral data,using the S-G convolution smoothing,first order differential transformation,logarithm spectrum transformation methods,e.g.,using the partial least squares regression(PLSR),and random forest regression(RFR)methods in the study of soil heavy metal content inversion.Through the above research could be well provided technical support and rational reference for land reclamation survey,planning,evaluation,monitoring and supervision.The conclusions were as follows:(1)Based on high resolution remote images(GF-1)and figured out 33 feature variables,the classification of land use and cover with GS-RF algorithm was carried out to get the highest accuracy was 88.16%.(2)Based on object-oriented analysis(OBIA)methods,using the parameters optimization of multi-scale segmentation method to the same experimental data and carrying out the classification,the results showed that the method could further improve the classification accuracy and precision was increased to 89.58%from 88.16%.(3)The classification of land use and cover with multiple endmember spectral mixture analysis(MESMA)based on medium-spatial resolution remote sensing data(Landsat-8),the results show that this method can reduce the time cost and improve the working efficiency on the basis of ensuring the accuracy of classification(90.5%).(4)Based on the ASD FieldSpec 4 hyperspectral remote sensing data,using PLSR method for spectral data modeling inversion with heavy metal content in soil.The result shows that the relevance of high-spectral data and heavy metal element content in soil was stronger,the accuracy of inversion modeling was high.The goodness of fiL of cadmium Cd R2 was 0.79.Innovative attempts had been made from the following aspects:(1)The knowledge framework of special information extraction of remote sensing data from the industrial and mining land reclamation area were presented systemically.(2)This research brought forward multi-feature detailed strategy for classification of land cover and use and proved the technical route of high precision classification in the complex environment.(3)Application of the information extraction mode of multi-feature fusion and construction of spectral index that was helpful for impervious surface information extracting were adopted to to strengthen the classification algorithm to identify low vegetation area,bare soil area and industrial and mining areas.(4)The grid search(GS)method based on out of bag(OOB)error estimation and the approach of multiple endmember spectral mixture analysis(MESMA)and genetic algorithm(GA)were used to improve the accuracy of classification and regression of target information based on machine learning technique.