Remote Sensing Imagry Reconstruction And Classification Via Deep Learning

Remote sensing imagery as one of an important methods of earth observation for human has become a highly acadmical research field.Remote sensing imagery has been widely applied in many fields,including agriculture,forestry,military,urbanization and natural disaster analysis,etc.Deep learning especially Convolution Neural Network(CNN)has been a hot topic in the research of computer vision.Deep learning has achieived success in image classification,image semantic segmentation,instance segmentation,object detection and image caption.Meanwhile,deep learning based methods have achieved quite good results in the field of remote sensing image analysis and processing.Based on deep learning,the paper focus on researching in super-resolution reconstruction and multi-source classification for remote sensing imagery.The spatial resolution of remote sensing images is limited due to optical imaging mechanism,system hardware and atmosphere condition.Especially,it's difficult for hyperspectral image to obtain high spatial resolution with a high spectral resolution.Higher spatial resolution can better express the details of remote sensing images,which can help improve image interpretation.High spatial resolution remote sensing images can better remain the details and obtain more accurate analysis results.Hyperspectral imagery(HSI),as an important part of remote sensing imagery,is often used in the task of classification and object annlysis owning to its rich spectral information.The misclassification of HSI is often caused by the 'different body with same spectrum'and 'same bady with different spectrum'.Consider combining HSI and other type of remote sensing image,like LiDAR etc.,can imporove the performance of classification.The main construction of this paper are as follows:1.Analysis the reason of low spatial resolution of remote sensing images,and build a degraded model for this situation.With the basis,deep convolution neural network are designed to convert the low-resolution image to a high resolution version to achieve the task of super-reoslution,including SRDCN,DSRDCN,ESRDCN.2.With the situation of a single hyperspectral data has a weak ablity to distinguish different objects with same spectral.Consider importing data from other source help hyperspectral data for classification.And a two-branch convolution neural network is designed for this task.And the feature are extracted and fused at the same time through stacking operation.