Fast Detection and Reconstruction of Tank Barrels Based on Component Prior and Deep Neural Network in the Terahertz Regime

Abstract

Terahertz (THz) regime has shown superior performance in terms of reflecting the details of target structures. However, the extended structures (ESs) may be discretized into endpoints in synthetic aperture radar (SAR) images if the observation aperture deviates from the specular orientation of ESs, which deteriorates subsequent image interpretation and intelligent imaging. Taking the tank barrels as the object, this paper proposes a novel solution to detect and reconstruct the critical ES by effectively exploiting the component prior and imaging characteristics (CPIC). The core of the proposed method lies in converting the phase matching of the multi-views methods into object detection based on CPIC and deep neural network. Firstly, the CPIC of tank barrels is analytically determined and regarded as the theoretical basis of datasets annotation. Thus, the modified multi-scale object detection network is constructed to enhance the detection performance and estimate the orientation of barrels. Finally, comprehensive simulation, anechoic chamber and field experiments are carried out to validate the effectiveness of the proposed methods. The results show that the proposed method can outperform the existing object detection networks in terms of the detection accuracy of multi-scale objects, and outperform the existing multi-view methods in terms of time need with comparable accuracy.