Recognition Method Of Rare Earth Mining Based On Deep Learning In High-resolution Remote Sensing Images

As an extremely important strategic resource in China,rare earth is widely used in microelectronics,chemical industry,chip,machinery,national defense,new energy and other high-tech fields.At a time when countries around the world step up their competition for strategic resources and the trade war between China and the United States continues to escalate,how to overexploit rare earth mining areas in real time and efficiently,monitor and control illegal mining,and standardize the management of rare earth mining,The effective protection of rare earth resources has become a problem that can not be ignored in rare earth industry.In this paper,with the help of the rapid development of artificial intelligence depth learning technology in recent years,the feature recognition method of rare earth mining in southern Jiangxi is constructed by using high resolution remote sensing image as the data source,which is used as rare earth line.The supervision of the industry provides technical support.The rare earth mining area in Lingbei,southern Gannan is selected as the study area,and the French 0.5m resolution Pleiades image is used as the data source to study the deep learning method of remote sensing image ground object recognition in rare earth mining area.The main contents are as follows:(1)the obtained Pleiades image data are preprocessed,such as fusion,cutting recognition,data enhancement and image size normalization,and the data set is tagged and position information according to the actual type of rare earth mining area ground objects.The data are divided into training samples and test samples to form the deep learning data set of remote sensing images in rare earth mining areas.(2)according to the deep learning theories such as neural network model,activation function,convolution neural network,classifier and non-maximum suppression algorithm,a rare earth mining feature recognition algorithm based on Mask RCNN is constructed.(3)the recognition of rare earth mining characteristics is realized by adjusting the superparameters of the model,and the preliminary experimental results are obtained.(4)according to the experimental results,the NDWI of Pleiades image is added to the dataset to optimize the performance of the algorithm,and different backbone networks are used to test the recognition effect of rare earth mining features.Finally,the accuracy of rare earth mining feature recognition is improved to 95.53%.In summary,this paper combines the popular deep learning technology with remote sensing technology,and the recognition accuracy of rare earth mining features is high,and the recognition accuracy can be further improved to 95.53% after adding the normalized water body index of remote sensing image.It can provide technical support to rare earth regulators.