Scholarlink

<p>X-band maritime radar can deliver high spatial resolution and rapid revisit rates, which is suitable for maritime monitoring of small extended objects spanning multiple resolution cells. However, there are a lot of clutter interference in the scene for maritime multiple extended object tracking (MEOT). Especially, the echo signal of maritime object is very weak with severe deformation and occlusion, which makes it difficult to track accurately. To address these issues, we proposes an adaptive MEOT algorithm based on an improved Siamese network with template-search matching method, which combines an optimised Kalman filter for state prediction and an adaptive feature modeling for candidate object enumeration. Moreover, the proposed width-height difference intersection over union (WH-IOU) and watershed-based template update strategy can facilitate shape adaptation. The experimental results show that the proposed algorithm has a better performance than those of the traditional SiamFC and multi-kernel correlation filter (MKCF) methods. Tested on real X-band radar data, the proposed algorithm reduces the mean centre-location error (CLE) to 3.95px, 50.6% below MKCF and 98.9% below SiamFC, and attains high overall precision (OP) of 0.9 and distance precision (DP) of 0.92 under heavy clutter.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multiple extended object tracking of X-band radar based on improved Siamese networks

  • 期刊论文

  • 发表日期：

    2025年9月5日

查看全文
  • Jinlong Yang,
  • Ao Liu,
  • Yong Cheng,
  • Xiaojia Wu

摘要

X-band maritime radar can deliver high spatial resolution and rapid revisit rates, which is suitable for maritime monitoring of small extended objects spanning multiple resolution cells. However, there are a lot of clutter interference in the scene for maritime multiple extended object tracking (MEOT). Especially, the echo signal of maritime object is very weak with severe deformation and occlusion, which makes it difficult to track accurately. To address these issues, we proposes an adaptive MEOT algorithm based on an improved Siamese network with template-search matching method, which combines an optimised Kalman filter for state prediction and an adaptive feature modeling for candidate object enumeration. Moreover, the proposed width-height difference intersection over union (WH-IOU) and watershed-based template update strategy can facilitate shape adaptation. The experimental results show that the proposed algorithm has a better performance than those of the traditional SiamFC and multi-kernel correlation filter (MKCF) methods. Tested on real X-band radar data, the proposed algorithm reduces the mean centre-location error (CLE) to 3.95px, 50.6% below MKCF and 98.9% below SiamFC, and attains high overall precision (OP) of 0.9 and distance precision (DP) of 0.92 under heavy clutter.