With the rapid development of space technology, spacecraft play an important role in critical applications such as weather forecasting and map navigation. Therefore, spacecraft component recognition is crucial for monitoring spacecraft status and performing space missions. Semantic segmentation techniques are more suitable for spacecraft component recognition tasks because of their ability to perform pixel-level segmentation. However, due to the large size difference of spacecraft components, multiple spacecraft components of different sizes may appear in a single image at the same time. To address this problem, we propose a symmetric network called Multi Scale U-ConvNeXt (MSU-ConvNeXt), where the encoder is ConvNeXt and the decoder is Cascading Multiple-Scale Aware Pyramid Decoder (CM-SAPD). CM-SAPD consists of cascade operations and Scale Aware Pyramid (SAP) modules in four stages. Each stage is performed by cascading the encoder features of the same stage and all the outputs of the previous stage. The SAP module utilises a residual connection to enable better interaction between the original inputs and the fused features. In addition, SAP uses a different number of transposed convolutions according to each stage, allowing the network to recognise spacecraft components of different sizes at the same time. We verify the method on the UESD and URSO, and the experimental results show that MSU-ConvNeXt achieves better results in the simultaneous recognition of spacecraft components with large size differences.


    Zugriff

    Zugriff prüfen

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Titel :

    Multi-Scale U-ConvNeXt Network for Spacecraft Component Recognition


    Beteiligte:
    Shao, Xiaoxiao (Autor:in) / Gao, Shiwen (Autor:in) / Zhang, Wuxia (Autor:in) / Guo, Huinan (Autor:in) / Wang, Hua (Autor:in) / Lu, Xiaoqiang (Autor:in)


    Erscheinungsdatum :

    08.11.2024


    Format / Umfang :

    1162907 byte




    Medientyp :

    Aufsatz (Konferenz)


    Format :

    Elektronische Ressource


    Sprache :

    Englisch



    Traffic sign recognition system based on YOLOv8-ConvNeXt

    Qin, Lijuan / Tang, Xiaoyu / Fan, Chubin | SPIE | 2025


    Spacecraft Component Recognition using a Codebook of Texton Images

    Shi, Jian-Feng / Ulrich, Steve / Ruel, Stephane | AIAA | 2017




    Spacecraft Component Recognition Using a Codebook of Texton Images (AIAA 2017-5308)

    Shi, Jian-Feng / Ulrich, Steve / Ruel, Stephane | British Library Conference Proceedings | 2017