Aircraft Target Detection Based On Polarization Feature Classification In PolSAR Image

Polarimetric Synthetic Aperture Rader(PolSAR)is capable of acquiring polarization information of features by actively emitting microwave imaging,and is highly adaptable for all-weather and all-day operation,and also possesses the characteristics of strong penetration of feature targets,so it is applied in various fields such as target detection,area classification,geological exploration It has been used in various fields such as target detection,area classification,geological exploration,military strikes,etc.In the 21 st century,the detection of aircraft targets has a special military significance,and the research on aircraft target detection algorithms has gradually attracted the attention of scholars from various countries.In this thesis,the problem of aircraft target detection in PolSAR images is addressed,and after understanding the domestic and foreign methods for aircraft target detection,the feature of aircraft targets are further studied,and two innovative algorithms are given for the problems existing in the existing algorithms.Algorithm 1 gives a detection method based on region extraction and fusion of polarized scattering features for the characteristics of too much coherent spot noise in large and complex scenes of PolSAR images,inconspicuous shape features of aircraft targets,and too many false alarms easily generated by the results.The algorithm first extracts the runway and apron area of the airport,and after extracting the suspected aircraft target in this area,it combines Yamaguchi coherence decomposition and the data provided by references to construct a new feature quantity for thresholding to extract the real aircraft target;to further reduce the false alarms in the results,the extraction results are optimized by using the feature that the aircraft target has background homogeneity.To address the drawback that the threshold needs to be set in advance in Algorithm 1 and the adaptivity needs to be improved,the thesis gives a second algorithm: an aircraft target detection method based on feature processing and Pinball loss function support vector machine.In the Pin-SVM offline training part,the original multiple features are first filtered to extract the optimal subset of features with better classification performance,while the Pinball loss function is introduced into the original SVM in order to further weaken the effect of feature noise.In the aircraft target detection part,Algorithm 2 inherits the advantages of Algorithm 1.In order to improve the accuracy of the detection results,the runway and apron regions are first extracted and alternative targets within the regions are identified,and their features are extracted and fed into the already trained Pin-SVM to obtain the final detection results.Both algorithms in the thesis first extract the runway and apron areas and detect the suspected aircraft targets,and then use polarized features to detect the real aircraft targets,both of which are suitable for large and complex scenes.The first algorithm is simpler and does not require offline training in advance,and can directly perform feature judgments on the suspected aircraft targets in the images to obtain results,but requires setting thresholds in advance.The second algorithm first requires offline training,sample construction,and feature screening,but after training the Pin-SVM classifier,no manual threshold setting is required,and the detection of aircraft targets can be performed directly.Both algorithms are validated in the experimental stage using multiple sets of experimental data with large and complex scenes provided by AIRSAR and UAVSAR,and the results show that both Algorithms 1 and2 can effectively detect aircraft targets in PolSAR images.