Research On Remote Sensing Image Classification Algorithm Based On Improved DenseNet

With the acceleration of the process of urbanization in my country and the continuous expansion of the scale of cities,the rational planning of cities is of great significance to the future development of cities.The current specific construction situation of the city can be obtained through the high-definition remote sensing image,which has important reference value for the next construction of the city.In this context,this article uses a neural network model to study scene classification of remote sensing images to obtain the city’s surface information.The main research contents are as follows:（1）Firstly,this paper analyzes the characteristics of urban remote sensing image dataset,preprocesses the dataset,and then compares and analyzes several typical models.The results show that the classification performance of DenseNet model is the best.The classification accuracy is 97.14% and 89.04% respectively on UCMerced＿Land Use dataset and RSSCN7 dataset,which is determined as the basic model of this paper.The experimental results of DenseNet model show that the classification accuracy of the model needs to be further improved,and the classification performance of the model also needs to be further improved.（2）The accuracy of model classification is not high enough,which may lead urban managers to inaccurate understanding of urban information,which is not conducive to managers to make correct decisions.Channel attention module is introduced to highlight the important channels in the process of feature reuse,and spatial attention module is introduced to highlight the position information of objects in the image.After the model is improved,the classification accuracy is improved to99.29% and 91.43% respectively on UCMerced＿Land Use dataset and RSSCN7 dataset.（3）The classification performance of the model for similar scenes is weak,which may lead to the misjudgment of urban managers on some urban information,and eventually lead to adverse consequences.The feature extraction ability of the model is enhanced by improving convolution,and then improve the model’s understanding of the high-level semantics of the image through feature fusion.Finally,experiments show the effectiveness of the improvement and further improve the classification performance.（4）Finally,the scene classification system is realized by using PyQt5 framework,which mainly includes two modules: image selection module and classification module.The system can segment the image selected by the user and display the classification results of each image to the user,so that the user can understand the image information.The system can also display some parameters of the model and the training process on the dataset,so that users can understand the model information.