Research On Convolution Neural Network Classification Of Remote Sensing Image Based On Massive Interpretation Mark

How to quickly obtain remote sensing information from the massive remote sensing data and to realize intelligent interpretation has always been an important subject in the field of remote sensing technology.And the artificial neural network technology has achieved remarkable achievements in this field.Through the analysis of the current situation at home and abroad,it is found that many studies have improved the classification accuracy by improving the algorithm,and the classification accuracy has reached more than 90%.However,these studies are aimed at specific research area and experimental data.When the research area and experimental data are changed,it is very likely that the classification accuracy cannot be achieved.So,this paper improves the automation degree and classification accuracy of remote sensing image by changing the source of sample data,controlling the quality of sample data and ensuring the number of sample data.In this paper,the theory of deep learning and artificial neural network are combined,and the most successful convolutional neural network model is applied to the classification of high resolution remote sensing images.The advantage of deep learning is that feature extraction and learning are automatically acquired from large data.The convolutional neural network can reduce the number of parameters greatly and reduce the difficulty of training by two magic weapons: local region apperceive and weight sharing.The experimental data of this paper is not specific to a certain area,but from interpretation marks that extract the multiple remote sensing image data.The main results of this study are as follows:(1)To summary and analysis the application status of artificial neural network and convolution neural network in remote sensing image automatic classification,and then to introduce the theoretical basis of remote sensing image classification based on artificial neural network,the advantages of deep learning and the characteristics and model structure of convolutional neural networks;(2)Based on the Visual Studio 2015 software platform,the implementation of convolution neural network framework is completed by C# programming language and WPF interface framework;(3)Set the interpretation marks as the sample data of the convolutional neural network.The interpretation marks have two characteristics: the large amount of data and the insurance of true value.To ensure the quality of the interpretation marks by controlling the source of the interpretation and standardized storage.To guarantee the number of interpretation marks by generating by batch;(4)The sample data generated by the interpretation mark is used as the experimental data,and the classification experiment of the high resolution remote sensing image is completed by using the convolution neural network frame.The result is about that The higher the image resolution,the better the quality of the object samples,the more the number of training samples,the accuracy is better,however,as the number of training samples continues to increase,the rate of accuracy increase is slowed down;Finally,the classification accuracy of this paper is 86.17%.Within the available range and eliminates the limitation of aiming at specific research area and experimental data.It has very important practical significance of research on realize the intelligent interpretation.