Aircraft Detection Based On Feature Classification In PolSAR Image

Polarimetric synthetic aperture radar(PolSAR)can acquire abundant ground polarimetric information by actively transmitting microwave image to the ground,which is less affected by factors such as day and night and climate.It is widely used in the fields of target detection and strike effect evaluation.This thesis studies the aircraft target detection problem in PolSAR images under complex scenes.Firstly,the scattering characteristics of the aircraft represented in the PolSAR image are analyzed.By using these characteristics and the prior knowledge,an aircraft target detection algorithm fusing with environmental feature and scattering characteristics of PolSAR image and a multi-feature classification PolSAR image aircraft target detection algorithm are proposed respectively.Based on the research idea of using aircraft features to detect aircraft targets from PolSAR images directly,a PolSAR image aircraft target detection algorithm that fuses with environmental feature and scattering characteristics is proposed.The algorithm extracts suspected aircraft targets from the polarimetric whitening filter(PWF)map,and then converts suspected aircraft targets into super-pixels to obtain a mixed image.After polarimetric decomposition and classification,a new environmental feature is constructed to fuse with scattering feature to obtain the aircraft target detection feature,and binary classification(divided into aircraft targets and interfering objects)of suspected aircraft targets is performed through threshold discrimination to obtain the final detection results.Using the relationship between the aircraft target and the tarmac and runway,the tarmac and runway area can be used to filter out interference outside the area during the detection process.Based on this idea,a multi-feature classification PolSAR image aircraft target detection algorithm is proposed.The algorithm is divided into two parts: 1.The offline classifier training part.In this section,firstly this thesis use a hybrid feature selection method(a combination of Filter and Wrapper feature selection methods)to screen out aircraft target polarimetric features with better classification performance,and use these features to train a support vector machine(SVM)classifier.2.Aircraft target detection.In this part,dissimilation scattering power is constructed at first,and based on the dissimilation scattering power design algorithm,the tarmac,runway area,and suspected aircraft targets are successively extracted.Finally,the offline trained SVM is used to classify the suspected aircraft targets to obtain the aircraft targets.Both algorithms use features to classify suspected aircraft targets to obtain detection results.The first algorithm does not require training samples,and directly extracts suspected aircraft targets from the full map and discriminates them.However,due to the large number of suspected aircraft targets,the volume of calculation of the detection features one by one is tremendous.The algorithm is suitable for aircraft target detection in small and medium complex scenes.The second algorithm is more complicated and requires offline training of the classifier.However,since a large number of false alarms have been filtered out using the tarmac and runway during detection,the volume of calculation is small.It is suitable for large complex scenes.Both algorithms are verified by multiple measured data collected by UAVSAR and AIRSAR systems in the United States.The results show that the two detection methods proposed in this paper can effectively detect aircraft targets.