An Experiment Of Automatic Classification Of High Score Remote Sensing Image Based On Edge Buffer And Convolutional Neural Network

The automatic classification system of remote sensing image is always a hotspot and a difficulty in the field of remote sensing image processing.The first scholars used pixel-based methods to classify images.After the resolution of the images was improved,there were a series of new methods such as object-oriented method.However,there was never an automatic classification system for industrial remote sensing images,that is to say,the existing methods have some problems in accuracy.With the development of science and technology,the advantages of convolutional neural network algorithm in the field of target detection are obvious,and many scholars have applied it to the automatic classification of remote sensing images,but influenced by traditional habits,many of these attempts are based on object-oriented segmentation,which limits the potential of neural network algorithms.In this study,a new automatic remote sensing image classification system is designed by using the edge buffer as the training unit of the neural network and with the idea of multi-scale.Through experiments,the new method achieves 97% accuracy in the test set and93% accuracy in the training set.The main contents of this study are as follows:(1)The specific theoretical reasons for the unsatisfactory performance of object-oriented segmentation in current image classification are analyzed.(2)A multi-scale edge buffer extraction algorithm is designed,which can extract the edge buffer in multi-direction and multi-scale,and restore the extracted data to the original image position.(3)Multi-layer convolutional neural network are used to learn the features of edge buffer,and several different convolution networks are designed to adapt to the learning of edge buffer features at different scales.(4)The integration learning method is used to determine the classification attribute of the remote sensing image,and the final classification result is obtained by fusing the result with the result of multi-scale segmentation.(5)The effect of the classification is illustrated by a large number of illustrations,then this paper give some suggestions to improve the algorithm.