Scene Classification Of Remote Sensing Image Based On Convolutional Neural Network

Remote sensing images have the characteristics of wide coverage,strong time continuity and large amount of data.How to effectively and quickly classify and identify remote sensing image scenes has become a research hotspot in the field of remote sensing image applications.Scene classification of remote sensing images is a process of mapping image content to semantic content tags.The diversity,multi-resolution and complex features of remote sensing image data make it difficult to extract accurate feature information from remote sensing images,increasing the difficulty of accurately adding scene classification labels.The traditional image scene classification method cannot deal with the problem of remote sensing image scene classification quickly and accurately due to noise sensitivity,insufficient recognition accuracy and high processing complexity.Compared with the traditional classification method,the image scene classification method based on convolutional neural network extracts the feature descriptors by using the image as the direct input and using the trained model for automatic extraction,which can make the feature extraction precision higher.The classification effect is better.Based on the convolutional neural network method,the neural network training is used to establish the mapping relationship between the underlying features of remote sensing images and high-level semantics.Two different convolutional neural network models are used to study the classification of remote sensing image scenes.The main work is as follows:(1)A method for classification of remote sensing image scenes based on convolutional neural network is proposed.This method solves the problem of remote sensing image scene classification by establishing a fourteen-layer convolutional neural network model,and introduces dropout technology to improve the generalization ability of the model.The Softmax classifier is used to classify and output the experimental results.High-level semantic features of remote sensing images.In the specific experiment,the RSSCN7 data set is used for specific verification.Firstly,the input data is processed by data enhancement,and then through a series of parameter comparison discussions,the parameters used in the experiment are determined,including the convolution kernel size and dropout.The selection of the loss rate and the setting of the learning rate,etc.,can be seen through specific experiments.The experimental model has achieved96% accuracy on the data set,and can effectively identify the scene category of the remote sensing image.Its accuracy is 3.35 percentage points higher than the AlexNet convolution network,which is 9.82 percentage points higher than the traditional CNN,showing the advanced performance of the convolutional neural network model.(2)A method for classification of remote sensing image scenes based on multi-scale residual neural network is proposed.The method is based on an improved multi-scale residual neural unit,and a residual neural network structure with 56 layers depth is designed to process the multi-scale residual neural network model for remote sensing image scene classification.The experiment used NWPU Data Set data set to carry out specific remote sensing image scene classification verification.The experimental model achieved the highest classification accuracy of 99.37% on the data set,which has a good recognition effect.The accuracy is higher than the 44-layer depth multi-scale residual neural network by 0.24 percentage points,which is 0.68 percentage points higher than the 32-layer depth model;it is 3.64 percentage points higher than the proposed convolutional neural network structure,showing the multi-scale The residual neural network model identifies powerful performance.