Study Of Remote Sensing Image Scene Classification Based On Deep Learning

Remote sensing technology has developed rapidly,and the spatial resolution of remote sensing images has also been continuously improved.The information contained in high-resolution remote sensing images is more abundant,and it can express more detailed features of ground objects and the distribution patterns of ground objects.The different spatial relationships between ground objects make up different semantic scenes,and mining higher-level scene semantic information is an important direction for current remote sensing image understanding.In recent years,with the rapid development of deep learning technology and its successful application in computer vision,the research on scene classification of remote sensing images ushered in new op portunities.Based on deep learning technology and aiming at high accuracy of classification accuracy,this paper studies the classification methods of remote sensing image scenes.This paper explores the principle and mechanism of deep learning image feature extraction for remote sensing images,and verifies the advantages of convolutional neural network model for image feature extraction.Based on the convolutional neural network model and the XGBoost algorithm,a remote sensing image scene classification framework is designed,and the pre-training convolutional neural network is fine tuned using the transfer learning method.This ensures the accuracy of the classification,improves the efficiency of the framework,and reduces the training cost.The classification accuracy reached 95.57% and 83.35% on the UC-Merced dataset and the NWPU-RESISC45 dataset.This study was based on the remote sensing images of Haidian District of Beijing.Use POI data and shared bicycle data to supplement information on socio-economic activities that are lacking in remote sensing images.Haidian District,Beijing was selected as the research object,and scene classification was performed based on the actual urban functional area category.The final overall classification accuracy reached 80.94%.Finally,based on WebGIS technology and TensorFlow framework,this paper designs and implements a prototype platform for remote sensing image scene classification.Integrate the algorithms and models mentioned in this article,and support the management of data and visualization.