Aircraft Detection and Classification Based on Joint Probability Detector Integrated with Scattering Attention
Xiayang Xiao,Hecheng Jia,Qing Wang,Penghao Xiao,Chunzhuo Fan,Zhou Li,Haipeng Wang
DOI: https://doi.org/10.1109/taes.2023.3342798
IF: 3.491
2024-01-01
IEEE Transactions on Aerospace and Electronic Systems
Abstract:Synthetic aperture radar (SAR) image intelligence interpretation is always a challenging task due to its special imaging mechanism. This article focuses on the aircraft detection and recognition in large-scale SAR images. The objects in large-size SAR images are usually sparsely distributed and tend to be highly concentrated in certain areas. Therefore, the detection and classification in SAR images are affected easily by the imaging environment, leading to ineffective and instability. It suffers from time-consuming, high false-alarm rate, and low recognition rate. To alleviate these problems, a joint probability detector integrated with scattering attention (SA) is designed for aircraft detection and classification. Based on the fact that all aircraft are parked in the airport area, an airport detection method with mixed strategy is proposed to obtain the valid region (airport), narrowing down the scope of detection area, increasing the detection efficiency, as well as suppressing a lot of false alarms outside the airport. Considering that it is easy to generate false alarms within the airport on account of clutter scattering interference, a two-stage detector is introduced to improve the accuracy by replacing a stronger region proposal network and optimizing a lower bound to a joint probabilistic objective over two stages. Furthermore, the classification encoder module incorporates SA into a convolution neural network to leverage the distribution relation among strong scattered points to enhance the classification performance. Finally, extensive experiments on the Zhong Ke Xing Tu (ZKXT) and GaoFen-3 datasets verify the effectiveness and superiority of the proposed approach, and surpass runner up by 5.3% with the mF1 indicator. On a large GaoFen3 image (16 550 x 11 945 pixels), the proposed method improves the detection speed by 187.52%, without sacrificing the accuracy.