Scholarlink

<p>With the growing demand for marine resource development, the detection of marine benthic organisms has become increasingly important for ecological protection, biodiversity, and resource management. However, traditional detection algorithms face challenges in complex marine environments, and advancements in deep learning technology provide new solutions. To enhance the precision and performance of detecting benthos in marine settings, this paper introduces an improved version of the Faster R-CNN model. The key enhancements include using ResNet50 for feature extraction, incorporating a multi-scale channel attention mechanism to highlight critical features and accommodate multi-scale information, replacing RoIPooling with RoIAlign to reduce quantization errors, and substituting the fully connected layer with an extreme learning machine classifier to enhance generalization and accuracy. Through extensive ablation experiments and comparative analyses, this paper demonstrates that the proposed improved model achieves higher detection accuracy compared to current mainstream detection models. Specifically, on the URPC2020 dataset, the model achieved approximately a 5% improvement in detection accuracy, and on the RUOD dataset, it achieved about a 10% improvement. These results validate the efficacy and excellence of the proposed model in marine benthos detection.</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

Faster R-CNN Marine Benthos Detection with Multi-scale Channel Attention Mechanism and RoIAlign

  • 期刊论文

  • 发表日期：

    2025年12月16日

查看全文
  • Yufeng Qiu,
  • Zhiyu Zhou,
  • Junyi Yang

摘要

With the growing demand for marine resource development, the detection of marine benthic organisms has become increasingly important for ecological protection, biodiversity, and resource management. However, traditional detection algorithms face challenges in complex marine environments, and advancements in deep learning technology provide new solutions. To enhance the precision and performance of detecting benthos in marine settings, this paper introduces an improved version of the Faster R-CNN model. The key enhancements include using ResNet50 for feature extraction, incorporating a multi-scale channel attention mechanism to highlight critical features and accommodate multi-scale information, replacing RoIPooling with RoIAlign to reduce quantization errors, and substituting the fully connected layer with an extreme learning machine classifier to enhance generalization and accuracy. Through extensive ablation experiments and comparative analyses, this paper demonstrates that the proposed improved model achieves higher detection accuracy compared to current mainstream detection models. Specifically, on the URPC2020 dataset, the model achieved approximately a 5% improvement in detection accuracy, and on the RUOD dataset, it achieved about a 10% improvement. These results validate the efficacy and excellence of the proposed model in marine benthos detection.