Research On Remote Sensing Image Scene Classification Based On Deep Learning Lightweight Convolutional Neural Network

The rapid development of remote sensing observation technology has improved the information acquisition capability of the observation system,making remote sensing images presented with high-resolution characteristics.High-resolution remote sensing images contain more spatial,texture,and semantic feature information,which can express more detailed ground objects features and the related features between them.The remote sensing image scene classification task is dedicated to extracting and understanding the semantic feature information in remote sensing images,and plays an important role in the fields of urban resource management,agricultural monitoring,natural disaster detection,and geographic data acquisition.The convolutional neural network method based on deep learning is the mainstream method in the field of image processing.This thesis mainly studies the use of lightweight convolutional neural network to complete the task of remote sensing image scene classification.Aiming at the distribution characteristics of various objects in remote sensing images and the shortcomings of traditional convolution methods to extract features,this thesis proposes a remote sensing image scene classification method based on lightweight dilated convolutional neural networks.This method first applies dilated convolutional neural networks to extract global features and related features of ground objects in remote sensing images,and then applies low-weight and structured pruning algorithm to remove the redundancy in the network.The experiment compares the network performance and resource consumption of this method and the traditional convolutional neural network method.Experimental results show that the application of dilated convolution and network pruning algorithms improves the classification accuracy of convolutional neural networks and reduces the space and time complexity of the model.In order to improve the classification performance of lightweight convolutional neural networks in the remote sensing image scene classification task,this thesis takes the dilated convolutional neural network as the basic network model,uses the residual module to adjust the structure of the network model,and proposes a scene classification method for remote sensing images based on feature fusion and convolutional neural networks.This method mainly applies the residual module to fuse the features of different network layers in the convolutional neural network and the global features of different scales.The experimental results show that the convolutional neural network method based on feature fusion can extract more comprehensive features and effectively improve the classification performance of the convolutional neural network.In order to ensure the classification performance of the convolutional neural network and meet the requirement of a more lightweight convolutional neural network structure.This thesis takes the depthwise separable convolution method as the core,and proposes a remote sensing image scene classification method based on depthwise separable convolutional neural network.The use of depthwise separable convolution can further reduce the network model parameters and the amount of calculation to build depthwise separable convolutional neural network model.At the same time,the goal-based knowledge distillation method is used to make up for the loss of classification performance of the network model.The experimental results show that the remote sensing image scene classification method based on depthwise separable convolutional neural network guarantees the classification performance,and obtains a smaller model volume and lower model complexity than the typical lightweight convolutional neural network Squeeze Net and Mobile Net.