Research On Convolution Neural Network For Optical Remote Sensing Image Recognition

With the development of remote sensing technology,the remote sensing data acquired every day,especially after entering twenty-first Century,the remote sensing data is explosive growth.Along with the data,remote sensing image enhancement technology and remote sensing technology have also been developed rapidly.Deep learning is one of the most popular fields of machine learning,and the convolution neural network is one of the deep learning models,which has very broad prospects in the field of target recognition,target detection and other fields.Low illumination optical remote sensing image has a poor visual effect,and the difference of image features is low,which can not meet the requirements of recognition.Image enhancement technology of low illumination optical remote sensing is a technology to enhance the quality of the remote sensing image,and improve the contrast and visibility of remote sensing image.In this paper,a fast and effective enhancement algorithm is proposed,which is based on the characteristics of low illumination optical remote sensing image.This algorithm proposed a fast and effective enhancement algorithm.The proposed algorithm is based on the Retinex decomposition of the image.Remote sensing image target recognition is a complicated problem in the field of remote sensing.Remote sensing image is rich in content,and the shallow layer feature is extracted by hand,which can not accurately and effectively express the content of the image.Deep convolutional neural network is a hierarchical artificial neural network model,each layer is the abstract expression of the image,and the use of high level of neural network extraction is the image of high-level feature is advanced.In this paper,we introduce the convolutional neural network into the target recognition field of optical remote sensing image,and use the feature transfer method to train the convolutional network.In the UCMerced-Landuse data set,we can test the experiment.The accuracy rate of the recognition is improved to 97.6%,and the best results are achieved.The remote sensing image has many characteristics,and has different characteristics in different scales.This paper studies the recognition performance of the convolutional network,analyzes the robustness of the convolutional neural network to image scale,and proposes a super resolution Pyramid convolution network model,which can effectively improve the performance of remote sensing image target recognition.