Research On Multi-source Remote Sensing Data Collaborative Deep Learning Classification Method

In the context of remote sensing big data,the collaborative classification of multi-source remote sensing data is a hot issue in the intelligent processing of remote sensing information.Deep learning as one of the best methods of big data processing and analysis,has been proven to be a breakthrough method in many research fields.Therefore,this article is based on the hyperspectral,multispectral,and lidar data,the research is mainly conducted from the following three aspects.First of all,in view of the traditional multi-source remote sensing image classification method that not only requires manual extraction of features,but also cannot guarantee the overall optimal solution,a classification algorithm of multi-source remote sensing image based on dual-branch convolutional neural networks is proposed.3D-1D parallel convolutional neural network branches and 3D convolutional neural network branches are designed to extract spatial and spectral features of hyperspectral and multispectral images,the features extracted by branches are fused,and use it for classification tasks.Secondly,in view of the over-fitting problem of convolutional neural networks when the number of training samples is insufficient,a classification algorithm of multi-source remote sensing data based on a pseudo-siamese weight sharing network is proposed.By integrating the advantages of hyperspectral and lidar data,finding the internal information of different data,we can achieve a more comprehensive description of ground objects.The3D-2D cascaded convolutional neural network and the 2D convolutional neural network are designed to extract the feature of the hyperspectral and lidar data respectively.Among them,the two networks use weight-sharing convolutional neural networks in the last three2 D convolutional layers to reduce the number of parameters,guide the networks to learn from each other,promote the process of feature fusion and prevent network over-fitting.Finally,aiming at the problem of a large number of labeled samples that are difficult to obtain from remote sensing data under the classification tasks,a classification algorithm for multi-source remote sensing data based on unsupervised feature learning is proposed,the “encoding-decoding” step is designed to train the network,the hyperspectral,multispectral,and lidar data are respectively provided to the multi-source convolutional autoencoder for feature learning,and then use the learned features to reconstruct the image.The network can automatically learn effective data features in an unsupervised manner and use it for classification tasks.