Research On The Optical And SAR Scene Matching Method For Aircraft Navigation

Aircraft scene matching precision navigation technology is of great significance in improving the guidance accuracy,and a lot of research funds have been invested in it at home and abroad.Its two key technologies scene matching and suitable matching area selection are more challenging under the background of heterogeneous matching.Given the limitations of the existing scene matching and suitable matching area selection methods based on deep learning,further research is helpful to improve the performance of scene matching precision navigation.Herein,this paper analyzes and summarizes the deficiencies and limitations of the existing scene matching and suitable matching area selection methods.The effectiveness of structure information to resist the nonlinear intensity differences of heterogeneous images is confirmed,and the relationship between the deep features extracted from the deep network and the traditional suitable matching parameters is analyzed.We study how to make full use of image feature information to improve the performance of scene matching and suitable matching area selection from the perspective of refining structure features and multi-modal information fusion.The research contents and results are as follows:(1)A heterogeneous scene matching framework using multi-scale convolutional gradient features is constructed.In our method,a shallow pseudo-Siamese network is proposed to convolve multi-orientated gradient channel features in a multi-scale manner,resulting in a deep structure feature descriptor named multi-scale convolutional gradient features,which can effectively deal with nonlinear intensity differences between SAR and optical images.Experimental results show that the proposed method can overcome some limitations of current matching methods based on deep learning to a certain extent,which requires solving a huge number of model parameters by a large number of training samples,and can solve the problem that traditional hand-crafted structure features are limited by the richness of image structure information.The proposed method achieves satisfactory matching performance and computational efficiency.(2)A method for suitable matching area selection of reference maps based on multi-modal information fusion is proposed.The proposed method constructs a dual-branch multi-modal information fusion network.The network includes two parallel channels.The first channel regards the shallow suitable matching parameters as the language description of the image,and uses the natural language processing network to extract the deep suitable matching parameter vector.The second channel uses the deep network to extract the deep features based on the original image.Finally,we fuse the two multimodal information to make a strategic decision for suitable matching area selection.The results show that our method can effectively improve the performance of the suitable matching area selection.