Scholarlink

<p>Underwater object detection technology plays an important role in the areas of marine resource exploration, environmental protection, and species surveys. However, the quality of underwater image is degraded by uneven illumination and fuzzy water interference. As a result, it is challenging to accurately extract underwater object features, which reduces the detection accuracy of existing approaches. To solve this problem, a two-stage feature enhanced hybrid network is proposed. Firstly, a hybrid feature enhancement network&#xa0;(HFEN) is constructed. Both long range global information and richer local detailed information is extracted, hence the multi-scale feature extraction power of network has been enhanced. Furthermore, an edge feature enhanced module&#xa0;(EFEM) is designed. More boundary information of object is extracted from shallow layers, thus the discriminative feature representation of object in low contrast underwater environments is obtained. Finally, a two-stage object perception enhanced&#xa0;(TOPE) detection head is devised. Both semantic and prior information of object is extracted while the interference from background noise is suppressed, which results in superior localization and classification results. The qualitative and quantitative experimental results on two publicly available datasets demonstrate that the proposed network could obtain favorable detection results compared with other state-of-the-art approaches.</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

TFEH-Net: Two-stage feature enhanced hybrid network for underwater object detection

  • 期刊论文

  • 发表日期：

    2025年6月18日

查看全文
  • Tingting Yao,
  • Ning Li,
  • Xiangzheng Zhao,
  • Chenyi Lu,
  • Qing Hu

摘要

Underwater object detection technology plays an important role in the areas of marine resource exploration, environmental protection, and species surveys. However, the quality of underwater image is degraded by uneven illumination and fuzzy water interference. As a result, it is challenging to accurately extract underwater object features, which reduces the detection accuracy of existing approaches. To solve this problem, a two-stage feature enhanced hybrid network is proposed. Firstly, a hybrid feature enhancement network (HFEN) is constructed. Both long range global information and richer local detailed information is extracted, hence the multi-scale feature extraction power of network has been enhanced. Furthermore, an edge feature enhanced module (EFEM) is designed. More boundary information of object is extracted from shallow layers, thus the discriminative feature representation of object in low contrast underwater environments is obtained. Finally, a two-stage object perception enhanced (TOPE) detection head is devised. Both semantic and prior information of object is extracted while the interference from background noise is suppressed, which results in superior localization and classification results. The qualitative and quantitative experimental results on two publicly available datasets demonstrate that the proposed network could obtain favorable detection results compared with other state-of-the-art approaches.