Building The Convolutional Neural Network Based On Multi-Source Data And Its Application For Lithological Mapping

The early lithological identification method can only reflect the lithology near the surface,and it is difficult to identify the deep lithology,and it is even impossible to detect the change of underlying lithology in the covered area,which limits the effect of geological survey.At present,the research on lithological identification of covered area is in the stage of exploration,the identification precision is low,and the mature and effective technical route and identification method have not been formed.The multisource data fusion technology provides an effective way to combine the data features of the same object from different angles and synthesize the advantages of various data sources,thus providing abundant information for lithological identification.Therefore,how to make full use of the advantages of the existing multi-source geological data and improve the accuracy of lithological identification in the covered area is worthy of indepth study.Based on this question,this study takes the Baolige area in Jilin,Inner Mongolia as the research area,and introduces the geochemistry data and remote sensing image data into the lithological identification,firstly,remote sensing image data and geochemistry data are fused by multi-source data fusion technology,making full use of the advantages of different types of data.Then,the convolutional neural network and random forest algorithm model are used for lithological mapping.The main work includes:(1)Fusion of Sentinel-2A and ASTER remote sensing images.Gram-Schmidt fusion technology is used to fuse Sentinel-2A and ASTER remote sensing images,which can improve the spatial resolution and enrich the spectral information of remote sensing images.(2)Fusion of remote sensing images and geochemical data.The mineral characteristic spectrum is influenced by the mineral chemical composition,and there is correlation between remote sensing image and geochemistry data.In this thesis,we use multi-source data fusion technology to fuse the two,and the fused data has original geochemistry features and rich spatial texture details.(3)Using convolutional neural network model for lithological mapping.Convolutional neural network model is a neural network model with multiple hidden layers,which can learn and extract deeper abstract features of data,and then achieve the purpose of improving the accuracy of lithologic unit classification.The results show that the convolutional neural network is an effective method for identifying lithologic units and can make full use of the spatial features of the data to solve problems such as “Salt and pepper phenomenon” caused by random forest,the obtained results of lithological mapping coincide highly with the original geological map,and the overall classification accuracy is 83%,indicating that the lithological mapping model has important research value and application potential in the covered area.