Research On Remote Sensing Image Classification Method Based On Deep Fusion

Remote sensing image has abundant information,the fusion method uses the abundant information of remote sensing image to improve the classification accuracy,but the fusion feature have problem of low information utilization and poor complementarity.To solve the problems,this paper studies the remote sensing image fusion method from two aspects:spectral information fusion of hyperspectral images and multi-source remote sensing image fusion.The main work of this paper is as follows:(1)To solve the problem of low effective information utilization,a multi-layer fusion spectral-spatial dense features deep fusion method is proposed.The spectral information and spatial information are processed through two channels to avoid spectral-spatial loss.Densenet achieves feature reuse and improves the utilization of information.More discriminative deep spectral-spatial feature is obtained by multi-layer fusion method.The effectiveness of the algorithm is proved on three different data sets.And the accuracy keep high when training sample reduced.(2)To improve the complementarity of fusion spectral-spatial features,we propose a multi-view spectral-spatial features deep fusion method.This method combines the traditional feature extraction method with the deep learning method,and describes the hyperspectral image in multiple views.The model still has high accuracy in the simple structure and converges faster.The multi-view method enhances the complementarity of features,improves the quality of fusion features,and reduces the dependence of the model on a large number of training data.The effectiveness of the algorithm is proved on three data sets and the performance is stable when the training samples are reduced.(3)To improve the fusion efficiency of multi-source features and fully fuse the hyperspectral spectral-spatial features and LiDAR elevation features,we propose a multisource remote sensing features deep fusion method.This method fuses the LiDAR elevation feature and hyperspectral spatial feature according to the feature meaning through multi-view,and fuses the spectral feature and spatial-elevation feature through the multi-layer fusion structure.This method makes full use of the complementarity of elevation features and spectralspatial features and improve the efficiency of multi-source feature fusion.