Scholarlink

<p>Automated ship detection and recognition have been a challenging and interesting issue in computer vision. Convolutional Neural Networks (CNNs) have become an efficient technique in ship detection. Because CNNs are highly complex, they provide an internal good representation of learned features. In the same way, they are prone to overfitting. Because of that, regularization techniques are needed for the minimization of overfitting and the improvement of the performance. We present in this paper an efficiently new approach, called ShipNET, which solves the task of ship detection in adverse conditions. As a matter of fact, this proposed approach is able to greatly impact the CNN performance and reduce its validation error. The experimental results show a clear amelioration when comparing it with different literature methods. In addition to that, the processing time is computationally effective as well, hence its use in real-time applications. The proposed approach yields a 99% precision and 98% specificity.</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

Deep Learning-Based Algorithm for Ship Detection

  • 期刊论文

  • 发表日期：

    2026年3月30日

查看全文
  • Taher Khadhraoui,
  • Jamin Naghmouchi,
  • Sami Ktata,
  • Hamid Amiri

摘要

Automated ship detection and recognition have been a challenging and interesting issue in computer vision. Convolutional Neural Networks (CNNs) have become an efficient technique in ship detection. Because CNNs are highly complex, they provide an internal good representation of learned features. In the same way, they are prone to overfitting. Because of that, regularization techniques are needed for the minimization of overfitting and the improvement of the performance. We present in this paper an efficiently new approach, called ShipNET, which solves the task of ship detection in adverse conditions. As a matter of fact, this proposed approach is able to greatly impact the CNN performance and reduce its validation error. The experimental results show a clear amelioration when comparing it with different literature methods. In addition to that, the processing time is computationally effective as well, hence its use in real-time applications. The proposed approach yields a 99% precision and 98% specificity.