A Fully Automatic Algorithm for Reflector Detection in Radargrams Based on Continuous Wavelet Transform and Minimum Spanning Tree

Xiaofeng Liu,Wenzhe Fa

DOI: https://doi.org/10.1109/tgrs.2022.3170036

IF: 8.2

2022-01-01

IEEE Transactions on Geoscience and Remote Sensing

Abstract:Spaceborne radar sounding is a powerful technique in planetary subsurface exploration due to its large penetration ability into loose desiccated materials. Three radar sounders, Mars Advanced Radar for Subsurface and Ionosphere Sounding (MARSIS), shallow radar (SHARAD), and Mars Orbiter Subsurface Investigation Radar (MOSIR), are orbiting Mars and have obtained a large quantity of radargrams. These direct observations provide valuable information for deciphering subsurface structure and composition of Mars, which are critical for understanding internal stratigraphy, climate change, and orbital variation in the planet. For this reason, automatic, effective, and precise extraction of subsurface reflectors in radargrams is a prerequisite and remains as a challenge though recent studies have made great progresses. In this study, we propose a fully automatic method for reflector detection in radargrams using transform domain techniques and graph theory. In the preprocessing step, the spatial spectrum information is used to automatically extract region with reflectors in a radargram, which can significantly reduce the computational cost. Then, continuous wavelet transform (CWT) with double threshold is used to detect peaks in each frame, and minimum spanning tree (MST) is further used to connect the peaks based on global information. Our proposed method can quickly reconstruct the reflectors in radargrams with a relatively low tracking error rate, which is better than previous methods based on local information. To testify the effectiveness of the algorithm, we applied our algorithm to SHARAD radargrams in the two polar regions. The results show that more than 90% (80%) of the reflectors can be well-detected in the strong (weak) echo regions, and their subsurface information can be extracted efficiently. Compared with previous methods, our method has a higher degree of automation and a wider applicability. The proposed algorithm can be applied to analyze radargrams over large regions of Mars and other bodies.