This paper presents a deep learning-based image features extraction and compression for multi-tasks, which can be applied to various intelligent tasks. We explore the multilevel features of the source by designing different information extraction networks, which contain text semantics, image segmentation, and pixel information. We propose a coarse-to-fine architecture to excavate the plentiful semantic information received from the encoder. The coarse module recovers the multi-granularity image according to the receiving symbols, and the fine module fuse different quality image to improve the reconstruction performance. In particular, we use multi-attention networks to extract and recover the image features at pixel levels. To overcome the artifact blocks phenomenon during the image reconstruction process that lacks necessary information, we design a dual features block that can mitigate the problem. Meanwhile, the system can accomplish different tasks by changing the last layers of the model.


    Zugriff

    Zugriff prüfen

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Titel :

    Semantic Communication Approach for Multi-Task Image Transmission


    Beteiligte:
    Zhang, Zhenguo (Autor:in) / Yang, Qianqian (Autor:in) / He, Shibo (Autor:in) / Shi, Zhiguo (Autor:in)


    Erscheinungsdatum :

    01.09.2022


    Format / Umfang :

    494242 byte





    Medientyp :

    Aufsatz (Konferenz)


    Format :

    Elektronische Ressource


    Sprache :

    Englisch



    A Multi-Task Semantic Communication System for Natural Language Processing

    Sheng, Yucheng / Li, Fang / Liang, Le et al. | IEEE | 2022


    Task-Oriented Semantic Communications for Speech Transmission

    Weng, Zhenzi / Qin, Zhijin / Tao, Xiaoming | IEEE | 2023


    Contrastive Learning based Semantic Communication for Wireless Image Transmission

    Tang, Shunpu / Yang, Qianqian / Fan, Lisheng et al. | IEEE | 2023



    Universal Weighted-Knowledge Bases for Task-Unaware Semantic Communication Systems

    Jiang, Shiyao / Jiao, Jian / Zhang, Ke et al. | IEEE | 2024