Urban Functional Area Integrating Remote Sensing Image And Text Information Classification Method Research

Most of the existing classification technologies of urban functional areas rely on big data analysis and data mining,and it is difficult to obtain better classification results due to the influence of data itself.High-resolution remote sensing images are rich in spatial information,and the classification of urban functional areas can be realized by extracting features of remote sensing images to identify the types of ground objects and their spatial positions.In the face of complex scene classification,by improving ResNet152 remote sensing image classification model,text information and high-resolution remote sensing images are merged to improve the classification acc of city functional area.The main research contents include:1.Select the depth convolution neural network suitable for this remote sensing image data set.The data set used in this paper is preprocessed.In the first step,all black and all white images are removed.In the second step,contrast enhancement,brightness enhancement,random rotation,MSRCR and other methods are used to enhance the data.The preprocessed data set contains 25,000 remote sensing images with 5,000 images in each scene,which are divided into training dataset and test dataset according to the ratio of 4:1.Comparing the remote sensing image classification models,four network models,Vgg-16,ResNet-50,ResNeXt-101 and ResNet-152,are selected for experimental test.Experimental show the overall accuracy and Kappa coefficient of ResNet-152 model are higher than those of the other three models.Finally,the ResNet-152 model is selected as the remote sensing image classification model,and the ResNet-152 model is improved to realize the urban functional area classification model.2.Improvement of remote sensing image classification model based on dual-pool feature weighting structure.On the basis of traditional pooling methods,two pooling methods,max pooling and mean pooling,are selected to form a double-pooling structure.The network models with max pooling,mean pooling,median pooling and double pooling are compared and verified on remote sensing image data set NWPU-RESISC45 and color image data set Cifar-10.The network model with double pooling structure is superior to the traditional pooling method in classification accuracy.On the basis of ResNet-152 model with double pool structure,the feature recalibration strategy of SE module is introduced,and the importance of each feature can be obtained through training.The learned features are input into XGboost classifier to complete remote sensing image classification.Compared with ResNet-152,the classification accuracy of SE-ResNet152+XGboost model is improved by 2.24%.3.The urban functional area classification model is established by integrating text information and remote sensing images,and the classification of five urban functional areas,namely residential areas,schools,industrial parks,commercial areas and government districts,is realized.The urban functional area division model is realized from three aspects:(1)Text information classification model,which uses the user’s visit information to realize the urban functional area division.The feature module is constructed by feature extraction of visiting information,and XGboost model is used to train and classify.(2)SE-ResNet152+XGboost remote sensing image classification model.(3)Used the idea of ensemble learning,the urban functional area classification network is constructed,which integrates text information and remote sensing images.Text information classification network and SE-ResNet152+XGboost model are regarded as two weak classifiers,and model fusion is realized through XGboost algorithm ensemble learning.Experimental results show that the classification accuracy of the urban functional area classification model based on ensemble learning is greatly improved compared with other classification networks,and the recognition accuracy reaches 70.71%,14.09%higher than ResNet-152 model,and 11.85%higher than SE-ResNet152+XGboost model,which can better accomplish the task of urban functional area classification.