Airport runway detection based on a combination of complex convolution and ResNet for PolSAR images

Aiming at the problem that the real-valued neural networks cannot fully use of the complex information of the Polarimetric Synthetic Aperture Radar (PolSAR) images and the information loss during the data transmission of traditional convolutional neural network(CNN), a Complex Cross-Residual Network (CS-ResNet) based on a combination of Complex-Cross Convolutional Network(CS-CNN) and Residual Network(ResNet) is given to improve the performance of airport runway detection. Firstly, the complex-valued network is used to classify the terrains in PolSAR images, then morphological processing is used to extract the region of interests(ROIs) from the classification results, and finally the real runway area is extracted from the ROIs by using the geometric characteristics of the airport runway. A group of comparative experiments using actual measured data are conducted, and the results show that the algorithm in this paper can improve the detection accuracy, such as the structural details of airport runway and the reduction of false alarms and missing alarms.