| China is the world’s largest consumer of copper and nickel mineral resources,with copper and nickel mineral resources consumption accounting for over 50% and 40% of global consumption respectively.The rapid development of new energy industries has further boosted the demand for copper and nickel mineral resources.The Beishan orogenic belt is located at the southern edge of the Central Asian orogenic belt.A large number of Paleozoic basaltic-ultrabasic rocks associated with copper-nickel ores are developed in the Beishan area,which is one of the main sources of copper-nickel deposits in China.The formation of large copper and nickel deposits by small basal-ultramafic rocks is a distinctive feature of copper and nickel ore formation in the Beishan orogenic belt.Therefore,the identification of the distribution characteristics of small basalticultrabasic rocks,is crucial to the exploration of copper-nickel mineral resources in the Beishan region.Remote sensing data provides multi-temporal and multi-level geological information for mineral resources exploration.The Beishan area with dry climate and highly exposed rocks is conducive to the automatic identification of basaltic-ultrabasic rocks by remote sensing data.Multi-source remote sensing data have their own advantages and disadvantages for the identification of basaltic-ultrabasic rocks.It is therefore necessary to apply the multi-source remote sensing information fusion technology to combine the advantages of multi-source remote sensing data for basalticultrabasic rock identification.However,traditional multi-source remote sensing data fusion methods had some problems,which hindered the improvement of lithological classification accuracy,mainly including:(1)When the traditional pixel-level image fusion algorithm was applied to the fusion of heterogeneous remote sensing data with different spectral response ranges,the spectral distortion of the fusion results was serious,which affected the lithological mapping accuracy;(2)In feature-level fusion,the data redundancy of the high-dimensional spectral and texture feature data set limited the efficiency and accuracy of lithological classification;(3)The traditional decision fusion classification strategy did not take into account the difference in classification performance of different classifiers for different lithological units,and could not effectively combine the advantages of different classifiers from different data sources for lithological classification.In order to solve the above problems,researches were carried at the pixel-level,feature-level and decision-level information fusion and lithological classification techniques for multi-source remote sensing data.A new GS-Sup Re ME fusion technique framework for multi-source remote sensing data was proposed to alleviate the spectral distortion problem of pixel-level fusion results.New bionic optimization algorithms such as GNDO(Generalized Normal Distribution Optimization),MPA(Marine Predators Algorithm),ASO(Atom Search Algorithm)were introduced for dimensionality reduction and feature optimization of high-dimensional spectral texture features.A novel decision-level fusion strategy(OAI-MAXV)that takes into account the difference in classification performance of different classifiers for different lithological units was proposed,which successfully improved the classification accuracy of basalticultrabasic rocks and other lithological units in the study area.The main results of the thesis are as follows.(1)Pixel-level fusion of multi-source remote sensing data and basaltic-ultrabasic rocks identificationThe advantages and disadvantages of multi-source remote sensing data for lithological identification were compared.A new framework for fusion of multi-sources remote sensing data,GS-Sup Re ME,is proposed by combining the GS(Gram-Schmidt)transform fusion algorithm and the Sup Re ME(SUPer-REsolution for multi-spectral Multi-resolution Estimation)algorithm.GS-Sup Re ME and the radiometric resolution normalisation technique were applied to pixel-level fusion of World View-2,Sentinel-2and ASTER data,and fusion data(24 bands,2m spatial resolution)with the spatial and spectral detection capability of the three data sources was obtained.Compared with the pre-fusion data,the fusion data has significantly improved the accuracy of the basalticultrabasic rock mapping in the study area.Compared to the classification results of World View-2,Sentinel-2 and ASTER data,the classification results of fusion data have25.39,10.56 and 1.45 percentages higher producer’s accuracy for basaltic-ultrabasic rock,respectively,and have 24.91,15.53 and 2.33 percentages higher user’s accuracy for basaltic-ultrabasic rock,respectively.(2)Multi-source remote sensing data feature-level fusion and basaltic-ultrabasic rock identificationA framework for feature-level fusion of multi-source remote sensing data was proposed,using a combination of de-correlation and feature optimization.PCA,MNF,ICA algorithms were applied to extract spectral features and to de-correlated of original information.GLMC algorithm was used to extract multi-scale texture features.The extracted spectral and texture features were combined to form a high-dimensional multisource remote sensing feature set.Bionic optimization algorithms such as GA,PSO,GDNO,MPA and ASO were applied for dimensionality reduction and feature optimization of the feature set,successfully improving the overall accuracy of lithological classification in the study area.Under the condition of 200 training pixels,the spectral texture feature set,obtained by the MPA algorithm,achieved the highest the overall classification accuracy(OA of 95.0143% and Kappa coefficient of 0.9355)and recognition accuracy of basal-ultramafic rocks(93.08%).The addition of topographic and radar remote sensing features further improved the overall accuracy and basalticultrabasic rock recognition accuracy in the study area.The highest overall classification accuracy(OA and Kappa coefficients of 95.9117% and 0.9469 respectively)and basalticultrabasic auto-identification accuracy(94.92%)was obtained for the feature dataset(ASO400+TOPO)with the combination of the spectral texture feature set(ASO400),obtained by the MPA algorithm under the condition of 400 training pixels,and topographic features(TOPO).(3)Decision-level fusion of multi-source remote sensing data and identification of basaltic-ultrabasic rocksA novel decision-level fusion classification strategy(OAI-MAXV)was proposed by combining the overall classification accuracy index(OAI)with the majority-vote decision strategy(MAXV).In particular,the OAI is used to reflect the difference in classification performance of different classifiers for different lithological units,and thus to optimise the traditional majority voting decision strategy.Compared with the traditional decision classification strategies of MAXV and adjusted majority-vote decision strategy(AMAXV),OAI-MAXV significantly improved the overall lithological classification accuracy and basaltic-ultrabasic rock identification accuracy in the study area.Six OAIs were constructed,and the OAI-V strategy was supplemented and improved by exploring the effect of the six OAIs on lithological classification accuracy.Decision-level lithology classification tests were carried out with different data and different classifiers.Results showed the OAI-V algorithm(OAI-V(K2))based on 2 × × was found to be the most effective in improving the overall classification accuracy of lithology,and the OAI-V algorithm(OAI-V(K6))based on 6 × was found to be the most effective in improving the classification accuracy of basal-ultramafic lithology.The main contribution of the dissertation is to propose technical frameworks for the information fusion of multi-source remote sensing data and lithological classification from the pixel level,feature level and decision level,which successfully combines the advantages of different remote sensing data for rock and mineral information identification and successfully improves the identification accuracy of basaltic-ultrabasic rocks in the study area.It is hoped that the results of the dissertation can provide technical ideas for copper-nickel mineral resource exploration in the Beishan orogenic belt and lithological mapping in similar areas. |
PDF Full Text Request