文献类型：期刊文献

英文题名：GSCDet: Geometric Semantic Contour Detection for Noncooperative Spacecraft

作者：Chen, Liwei[1];Su, Lin[1];Dai, Zhiyong[2];Wang, Zhuoran[1];Yi, Jianjun[1]

机构：[1]East China Univ Sci & Technol, Dept Mech Engn, Shanghai 200237, Peoples R China;[2]Shanghai Publishing & Printing Colleague, Dept Informat & Intelligent Engn, Shanghai 200093, Peoples R China

年份：2025

卷号：34

期号：10

外文期刊名：JOURNAL OF CIRCUITS SYSTEMS AND COMPUTERS

收录：;EI(收录号：20251818326366);WOS:【SCI-EXPANDED(收录号:WOS:001473355300001)】；

基金：This paper was supported by Shanghai Science and Technology Action Plan under Grant No. 21JM0010300, Shanghai Aerospace Science and Technology Innovation Fund (SAST) under Grant No. 2021-037, the Special Fund Technology Innovation Support Project of Shanghai under Grant Nos. 2021-cyxt-kj1, XTCX-KJ-2022-37,HCXBCY-2023-046 and the National Defense Basic Scientific Research Program of China (Grant No. JCKY2021606B002).

语种：英文

外文关键词：Edge detection; contour detection; U-Net; noncooperative spacecraft

摘要：Space tasks such as docking, capture and maintenance of noncooperative spacecraft increasingly rely on vision systems to extract the geometric features of the target spacecraft. This paper introduces a novel contour type named geometric semantic contour (GSC), which can succinctly describe the spacecraft's fundamental geometry, and proposes a GSCDet detector to extract GSC from monocular images. However, surface texture and spatial illumination have been shown to have a significant impact on traditional edge detection methods, resulting in difficulties in extracting the GSC. In this paper, an analysis of the characteristics of GSC is presented, and the GSC detection problem is decomposed into a two-stage model comprising coarse GSC map detection and further optimization of GSC. The proposed GSCDet employs the U-Net enhanced by a dual cross-attention module to fuse the coarse GSC map obtained by the instance segmentation method as a geometric semantic prior with the general edge map extracted by the traditional edge detection method, to achieve accurate and robust GSC recovery. Extensive experiments demonstrate the superiority of GSCDet to state-of-the-art edge detection methods in terms of the ODS, OIS and AP metrics. In comparison with the second-best performing method on the real image dataset, the ODS, OIS and AP have increased by 4.6%, 6.5% and 6.8%, respectively. Furthermore, the ablation experiments provide a more detailed evaluation of the contribution of each module in GSCDet.