Research Of Remote Sensing Image Scene Classification Based On DCNN

With the continuous development of DCNN technology,the magnitude of satellite data has increased geometrically,and the information that can be expressed by satellite images has become more abundant.In the field of scene classification,which mainly converts image classification technology based on deep learning theory,and improves DCNN according to the classification results.The main content here is as follows:(1)After comparing five commonly used remote sensing image scene classification data sets,it is understood that high-resolution remote sensing image data sets have problems such as lack of categories,lack of quantity,and lack of regional characteristics.Therefore,in order to obtain better scene classification results,a large-scale,multi-category data set needs to be produced.This article introduces the process of collecting the data set in detail,and determines the necessary conditions for making the data set according to the classification method of this article,such as the resolution,illumination angle,and selected area of various scenes.24 types of scenes are collected,each of which is about 400.Finally,the highresolution remote sensing image scene classification data set needed in this article is produced.(2)Propose a remote sensing image scene classification method that can enhance the feature extraction ability of the network.Add branches to AlexNet to improve the feature expression ability of the image and filter out the background interference information of the image.Apply this method to the data set proposed in this paper,change the network structure according to the actual classification results,adjust the learning rate,optimization function and other hyperparameters,and finally obtain better classification results.(3)Through multiple sets of cross experiments on different data sets and different classification models,the results between the performance of the data set and the classification model are obtained,and the model is continuously optimized and improved through the classification results to make it in training It can converge quickly and achieve better classification results in remote sensing image scene classification.(4)Design and develop a visual and interactive scene classification system based on web applications.Users upload data at the front end of the platform,and display the classification results on the main interface of the system after scene classification in the background.And provide user results preview,download and other functions.