Identification And Classification Of Massive High-resolution Remote Sensing Images Based On Deep Learning

In the context of global environmental change,food security is the core issue of the sustainable development of human society.The distribution pattern of global farmland and the accurate evaluation of its evolution characteristics are the basis for guaranteeing grain production.At present,the distribution of global farmland is well grasped by multi-source remote sensing satellite inversion and statistical investigation,but the spatial distribution of farmland with high precision is still lacking,and the existing data are quite different in regional scale,which limits the global and regional crop Yield assessment and the improvement of grain yield forecasting,which makes food security problems with greater uncertainty.Therefore,it is necessary to construct the application system with high precision farmland identification and classification ability through the innovative method,and carry out the related research and application.Based on the high resolution remote sensing image data,the system constructs a high resolution image recognition and classification platform framework,and has been successfully applied in the Loess Plateau and Ningxia irrigation area.Based on the TensorFlow depth learning framework,this paper extracts the convolution and pool structure outside the classifier from the Inception-v3 network structure through the migration learning,and combines the Logistic regression classifier to realize the feature extraction and classification of the remote sensing image data Training and identification of pictures and other key technologies,while the development of a large-scale high-resolution images can be self-learning image itself,the core of the characteristics of the method.The results of training and recognition in this study show that the classification accuracy is 0.74,and the system framework of this study has a good application prospect.This study further suggests that with the development of distributed parallel computing framework and deep learning,the deep learning can improve the classification accuracy and the lack of resolution in terms of the recognition and classification of massive high-resolution remote sensing images.In the follow-up study,this paper will further improve the system framework,enhancing the farmland pattern and identifying farmland boundaries,to achieve high-precision farmland mapping and its application,to provide technical support to ensure food security.