| Due to the characteristics of microwave imaging,Synthetic Aperture Radar(SAR)effectively solves the problem of imaging quality under poor lighting and bad weather conditions,achieving large-scene observation of the ground.SAR have been widely used in disaster monitoring,crop estimation,military information collection,enemy detection,etc.,playing an important role in both military and civilian fields.Airports are responsible for the take-off,landing and parking of aircraft,thus they are highly valuable infrastructures in both civil and military fields.Airport region detection plays an important role in the fields of airport navigation and air reconnaissance.Large-scene SAR images contain a large number of airport-like regions such as roads,rivers and shadows,which lead to an increased false alarm rate of the algorithm.In addition,the fully polarimetric SAR images contain various polarimetric features,how to make the best use of the polarimetric features for airport region detection is also a major problem.In response to the above problems,this paper develops the study of airport region detection under large-scene SAR images based on the deep features and physical features in the airport region of SAR images.The main work are as follows:(1)Aiming at the problem that the airport region detection task under large scene single polarimetric SAR images is severely interfered by airport similar regions,fusion deep feature and line segment feature for single polarimetric airport region detection method is proposed.Based on the common deep detection model,the line segment detection branch is added to obtain line segment feature,which achieves the addition of parallelism constraint in the deep detection results,reducing the false detection generated by airport similar regions effectively.Finally,there is no publicly available airport region dataset for single polarimetric SAR images,a large scene single polarimetric SAR image airport dataset is constructed in this paper.The validation of the proposed method is performed on this dataset.Experimental results show that the airport region detection method based on deep feature and line segment feature has an accuracy of 91.4% and a recall of 86.5%.(2)Aiming at the problem that the polarimetric features in the fully polarimetric SAR images are not fully utilized and the physical features are not fully extracted,a fully polarimetric airport region detection method based on the scattering components and covariance matrix is proposed.Specifically,this method uses fuzzy clustering to classify the polarimetric SAR image based on the four scattering components,and obtain the airport candidate regions.Then,a line segment detection algorithm for polarimetric SAR image is designed by considering that the existing line segment detection algorithm does not use the multi-polarimetric information of full polarimetric SAR image.It consists of constant false alarm rate edge detection algorithm,statistical region merging algorithm and contrario approach,which can achieve fast line segment detection on full polarimetric SAR image.Finally,airport regions are obtained by validating the airport candidate regions with line segments.In addition,line segment detection algorithm and airport region detection algorithm are validated by UAVSAR image provided by NASA Propulsion Technology Laboratory. |
