Opto-Electronic Engineering, Volume. 52, Issue 2, 240270-1(2025)

Lightweight remote sensing military aircraft target detection in complex backgrounds

Hanlian Zhou... Qing Ye and Wenqi Liu |Show fewer author(s)
Author Affiliations
  • School of Computer Science, Yangtze University, Jingzhou, Hubei 434023, China
  • show less

    Aiming at the issues of low recognition accuracy, high computational cost, and large model size caused by the complex background and small target scale in remote sensing images of military aircraft, a lightweight military aircraft target detection algorithm, namely YOLOv8-MA, integrating reparameterization and detail enhancement is proposed. Firstly, a multi-branch gradient flow feature extraction module is designed through reparameterization to enhance the model's inference speed. Secondly, in combination with efficient RepGFPN, redundant model structures are discarded and the P2 layer is incorporated to construct a multi-scale feature fusion network, mitigating the problem of small target information loss due to excessive downsampling. On this basis, a lightweight detection head is proposed by integrating GN convolution and detail enhancement to reduce the number of model parameters and the amount of computation. Finally, a focus coefficient is introduced into the Shape-IoU to form a new loss function, thereby improving the detection performance of the model. On the public military aircraft dataset MAR20, the mAP50 of this algorithm is as high as 97.9%, and the model size is as low as 2.1 MB. Compared with YOLOv8n, the number of parameters decreases by 74.7%, the amount of computation reduces by 40.7%, and the FPS increases by 14 f/s, demonstrating that it can effectively enhance the detection effect of military aircraft in remote sensing images.

    Keywords
    Tools

    Get Citation

    Copy Citation Text

    Hanlian Zhou, Qing Ye, Wenqi Liu. Lightweight remote sensing military aircraft target detection in complex backgrounds[J]. Opto-Electronic Engineering, 2025, 52(2): 240270-1

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Article

    Received: Nov. 20, 2024

    Accepted: Dec. 30, 2024

    Published Online: Apr. 27, 2025

    The Author Email:

    DOI:10.12086/oee.2025.240270

    Topics