Convolutional Radon Transformation Method for Sparse Collective Recognizing
Bowen Cheng, Dan Wang, Jiaxiang Niu, LI Xiao-da, Shenshen Luan · 2023
The onboard processing technology based on remote sensing images has become the focus of current research area and received widespread attention. With the increasing amount of onboard data, how to extract effective information from massive data becomes increasingly important. In this paper, a novel sparse collective formation recognizing method is proposed to detect and analyze the formation of the surface ship formation. Firstly, a mobilenet is trained to detect the carrier and single ships to form a binary encoded map, which will be segmented by the DBSCAN method to extract the suspected formation. Then the binary encoded map is processed by the convolutional radon transformation and the result is compared with the standard confidential dataset (SCD) of three formations. Finally the formation type, confidential probability, direction of the surface ship formation was given by the comparison result which has the minimum root-mean-square-error (RMSE) with the SCD. The experimental results show that the proposed method can recognize the ‘Y’, ‘T’ and ‘I’ type surface ship formations with different offsets under more than 60% confidential probability and has a significant feasibility and robustness.