Research On Geological Body Recognition Method Based On Deep Learning

Geological mapping refers to the process of filling in various geological bodies and geological phenomena on the base map to form a geological map on the basis of field observation and research on a certain scale.Accurate identification of geological bodies can effectively assist geological mapping.In view of the fact that most of the geological body identification methods based on machine learning only use image data and the obtained geological information is single,there are many limitations in the application ability of this method.In this paper,deep learning technology is applied to geological body recognition,and the information of different modal data is deeply excavated.From the feature extraction and fusion of different modal data and the improvement of loss function in the training process of multi-modal feature fusion model,designing and optimizing the geological body identification model to realize the accurate identification of various geological bodies in the area to be mapped,so as to better assist geological mapping.The main work of this paper are as follows:(1)Geological body recognition based on multi-modal feature fusion.In order to obtain more geological information to improve the ability to identify geological bodies,this paper completely considers the data of two different modalities of geophysics,geochemistry data and remote sensing image data in the area to be mapped,proposing a geological body recognition model based on multi-modal feature fusion.Through the cross-validation results,it is found that the model has obvious advantages over the geological body recognition model that only uses remote sensing image data or geophysics,geochemistry data,and the overall classification accuracy is improved by 2.79% and14.08% respectively.It proves that the recognition model based on multi-modal feature fusion can realize more accurate geological body recognition,and then effectively assist geological mapping.(2)Model optimization method based on long tail datasets.The number of sampling points of various geological bodies in the predict area showing long-tailed data distribution,which makes the model easy to be biased towards categories with a large number of sampling points,resulting in a poor performance of the model in identifying categories with a small number of sampling points.The cross entropy loss function is a commonly used loss function in multi-classification tasks.Based on the Cross entropy loss function,this paper proposes a balanced-class loss function.According to the cross-validation results,the loss function can effectively guide the training process of the geological body recognition model,improving the stability of model identification ability,so as to solve the problem of unbalanced model recognition ability caused by the long-tailed distribution of sampling points of various geological bodies in the area to be predicted.