Classification Of SAR Images Based On SIFT And Restricted Boltzmann Machines

SAR(Synthetic Aperture Radar)is an active microwave imaging radar,which can be installed in aircraft,satellites,spacecraft and other flight platforms.It's advantage is that it is with all-day,all-weather observation of the ground,and has a certain Surface penetrating ability.Thus,it plays a very important role in the field of remote sensing that other remote sensing means are difficult to achieve.It has drawn more and more attention from all over the world.SAR has been widely used in many fields such as national economy,military defense and scientific research.With the progress and development of SAR technology,SAR classification and interpretation based on SAR image data has become a hot topic for domestic and foreign scholars.In this paper,the feature extraction and SAR image classification of SAR images are studied theoretically and verified by experiments.The main research contents are as follows:At first,in this paper,the SIFT(Scale Invariant Feature Transform)feature descriptor and DSIFT algorithm are studied in depth.And experiments have validated the effect of DSIFT.However,there are still some problems,such as high computational complexity caused by the high dimension and so on.This paper overcomes the shortcomings of SIFT algorithm,and further studies the SAR-SIFT and DSAR-SIFT algorithms based on DSIFT algorithm.Experiments show that these two descriptors are more suitable for SAR image feature extraction.Then,the DSIFT and DSAR-SIFT descriptors are extended to the depth learning model.Through the Deep Boltzmann Machine,the characteristics of the image were studied in depth.And the Deep Boltzmann Machine based on DSIFT and DSAR-SIFT is studied and verified by experiments.Finally,based on the bottleneck feature extraction and the Relevance Vector Machine model,a Bottleneck Deep Boltzmann Vector method based on DSAR-SIFT is developed.By stacking two Deep Boltzmann Machine,the bottleneck vector is constructed.And the bottleneck vector is chosen as the input of the Relevance Vector Machine classifier.This method has a good classification performance on the classification of SAR images,and has achieved good results.At the same time,it meets the needs of SAR image classification.