Aircraft Detection in SAR Images Based on Peak Feature Fusion and Adaptive Deformable Network

Due to the unique imaging mechanism of synthetic aperture radar (SAR), targets in SAR images often shows complex scattering characteristics, including unclear contours, incomplete spots, attitude sensitivity, etc. Automatic aircraft detection is still a great challenge images. To cope with these problems, novel approach called adaptive deformable network (ADN) combined peak feature fusion (PFF) proposed for detection. The PFF designed taking full advantage strong features aircraft, which consists extraction and fusion. fully exploit images, are extracted via Harris detector eight-domain pixel local maxima. Then, saliency under multiple conditions enhanced by multi-channel blending. All PFF-preprocessed fed into ADN training testing. core components contain an spatial (ASFF) module convolution (DCM). ASFF utilized reconcile inconsistency across different scales, raising characterization capabilities pyramid improving performance multi-scale further. DCM introduced determine 2-D offsets maps adaptively, geometric modeling abilities various shapes. well-designed established combining two modules alleviate problems sensitivity. Extensive experiments conducted on GaoFen-3 (GF3) dataset demonstrate effectiveness PFF-ADN average precision 89.34%, as well F1-score 91.11%. Compared other mainstream algorithms, achieves state-of-the-art performance.