Aircraft Detection Based on Multi-feature Classification for PolSAR Image

Aiming at the problem of high false alarm and poor flexibility in aircraft detection by PolSAR image, this paper presents a new aircraft detection algorithm for PolSAR image with large complex scenes. The performance of the detection process consists of SVM classifier training and detection of aircraft target. In the procedure of SVM training, multiple optimized polarization features of the aircrafts are selected through the combination of Filter feature selection and exhaustive method to train the SVM classifier. In the detection process, suspected aircraft targets are selected from the large PolSAR image by threshold method with dissimilation scattering power, then multiple polarization features extracted from the suspected aircraft targets will be input to the SVM classifier to obtain the detection result. The multiple testing data collected by UAVSAR system and the comparison with the existing algorithm of PolSAR image indicate that the method presented in this paper could effectively detect aircraft targets with less false and missed alarm as well as improved flexibility.