Scholarlink

The study focuses on the increasing complexity of automated target recognition (ATR) in surveillance and reconnaissance. It highlights the need for quick and accurate identification in crucial situations and emphasizes the role of technical improvements in addressing this problem. The study incorporates a self-attention mechanism into Convolutional Neural Networks (CNNs) architectures such as YOLOv8, DenseNet-121, InceptionV4, ResNet-152, and VGG16, with a specific focus on categorizing ground targets in Synthetic Aperture Radar (SAR) pictures. This systematic improvement incorporates self-attention modules carefully positioned at critical spots in the network architectures. The research utilizes the renowned Moving and Stationary Target Acquisition and Recognition (MSTAR) Mixed Targets dataset. YOLOv8 is the best classifier, with an outstanding total accuracy of 97.65% and it has high specificity of 98.86% and precision of 96.62% with the testing. This study emphasizes the need to thoroughly assess CNNs architectures to make educated decisions when choosing models designed explicitly for complicated target identification tasks. It offers valuable insights that may contribute to developments in ATR technology.

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

Incorporation of the Self-attention Mechanism into Convolutional Neural Networks for the Target Recognition

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Gangeshwar Mishra,
  • Prinima Gupta,
  • Rohit Tanwar

摘要

The study focuses on the increasing complexity of automated target recognition (ATR) in surveillance and reconnaissance. It highlights the need for quick and accurate identification in crucial situations and emphasizes the role of technical improvements in addressing this problem. The study incorporates a self-attention mechanism into Convolutional Neural Networks (CNNs) architectures such as YOLOv8, DenseNet-121, InceptionV4, ResNet-152, and VGG16, with a specific focus on categorizing ground targets in Synthetic Aperture Radar (SAR) pictures. This systematic improvement incorporates self-attention modules carefully positioned at critical spots in the network architectures. The research utilizes the renowned Moving and Stationary Target Acquisition and Recognition (MSTAR) Mixed Targets dataset. YOLOv8 is the best classifier, with an outstanding total accuracy of 97.65% and it has high specificity of 98.86% and precision of 96.62% with the testing. This study emphasizes the need to thoroughly assess CNNs architectures to make educated decisions when choosing models designed explicitly for complicated target identification tasks. It offers valuable insights that may contribute to developments in ATR technology.