Research On High-resolution Remote Sensing Image Scene Classification Based On Deep Learnin

With the rapid development of land observation technology,the number of optical remote sensing images has increased exponentially.It takes a huge amount of time and economic investment to manually process massive data,so an effective data extraction technology is urgently needed.In recent years,deep learning and related technologies have developed rapidly in the fields of computer vision,natural language processing,and language recognition.It also have been successfully applied in the field of remote sensing information processing.This paper focuses on the combination of deep learning and high-resolution remote sensing image classification tasks.A deep neural network high-resolution remote sensing image classification model based on dual-stream neural network and lightweight convolution fusion global feature extraction operator was established to realize automatic classification of high-resolution remote sensing images.The main work of this paper is as follows:Aiming at the problem that the small sample size of the dataset makes it difficult to fit the deep neural network during training,a deep neural network model based on deep transfer learning is proposed.The deep neural network is pre-trained on a large-scale dataset of the same task,and the prior knowledge obtained after training is transferred to a small-scale dataset.The comparative experiments show that this method can effectively alleviate the overfitting phenomenon caused by the small number of data samples and shorten the training time.The existing convolutional neural network cannot overcome the problems of large intraclass diversity and high inter-class similarity of remote sensing images.A two-stream remote sensing image scene classification model based on multi-scale attention combined with graph neural network is proposed.Firstly,multi-scale feature extraction is introduced on the basis of channel attention,and secondly,long-distance spatial information is extracted through graph convolutional neural network,and a weighted fusion strategy is used to fuse multiple features to achieve classification.In the experiment,a variety of data enhancement methods are used to train the model,which effectively improves the generalization ability of the model.Aiming at the requirements of specific scene tasks,a high-resolution remote sensing image scene classification model based on lightweight network joint knowledge distillation is proposed.The scale of the model is compressed by lightweight network design,which greatly reduces the amount of parameters.The knowledge distillation method is used to improve the model’s ability to distinguish difficult categories,thereby improving the overall classification accuracy.The effectiveness of each module in this network is verified through ablation experiments.The experimental results show that this network can effectively reduce model complexity and improve classification accuracy.