GKPA-Integrating global and key region for Optical-SAR medium resolution remote sensing image scene classification
Remote sensing image scene classification is the basis for practical applications such as natural disaster detection and urban function planning.Therefore,it is of great significance to study remote sensing image scene classification.Deep learning-based methods have become the most commonly used methods in the field of remote sensing image scene classification due to their powerful feature extraction capabilities.Such methods strongly rely on the quality of the data set.At present,most of the data sets in the field of scene classification are high-resolution optical images,which are easy to extract effective semantic features.When faced with medium-resolution scenes,the detailed information in the image is scarcer,the effective features are not significant enough,scene classification is more difficult,and the classification accuracy is not ideal compared with high-resolution scenes.In this case,using data from other modals such as synthetic aperture radar(SAR)to provide complementary features can effectively improve scene classification accuracy.Considering the above status quo,this paper constructed an optical-SAR remote sensing image scene classification data set(OS-RSISC),and based on this data set further proposed a perceptual attention dual-modal scene classification framework integrating global and key region(GKPA-RSSC).Experimental results show that GKPA-RSSC has the highest classification accuracy compared to comparing methods,with an overall accuracy of 81.97%.The comparative results of optical-SAR data and single-modal data verify the advantages of dual-modal data,and then verify the importance of the optical-SAR scene classification data set proposed in this article.
remote sensing image scene classificationdeep learningoptical-SAR datasetintegrating global and key areaperceptual attention mechanism
万方数据
维普
