Scholarlink

Landmines remain a legacy of long histories of armed conflicts and malicious terrorist activities. Despite initiatives undertaken by the governments of these countries and inter-humanitarian actions, they continue to be deployed. This paper focuses on the use of convolutional neural networks (CNN) to improve landmine detection. It aims to reduce false positive rates in the mine detection process. The proposed model is based on a CNN algorithm, known as Faster R-CNN, renowned for its good performance in detecting objects in images. The article further relies on the Scikit-learn framework to identify and isolate false alarms caused by some types of mines with sensitive and specific features. Such mines are frequently encountered on the battlefields of northern Chad. The results obtained using this approach show a strengthening of detection capabilities while minimizing classification anomalies. These results could contribute to securing potentially dangerous areas.

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 for Reliable Landmine Detection: A CNN Approach to Minimizing False Positives

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Mahamat Ismael Hassan,
  • Daouda Ahmat,
  • Samuel Ouya

摘要

Landmines remain a legacy of long histories of armed conflicts and malicious terrorist activities. Despite initiatives undertaken by the governments of these countries and inter-humanitarian actions, they continue to be deployed. This paper focuses on the use of convolutional neural networks (CNN) to improve landmine detection. It aims to reduce false positive rates in the mine detection process. The proposed model is based on a CNN algorithm, known as Faster R-CNN, renowned for its good performance in detecting objects in images. The article further relies on the Scikit-learn framework to identify and isolate false alarms caused by some types of mines with sensitive and specific features. Such mines are frequently encountered on the battlefields of northern Chad. The results obtained using this approach show a strengthening of detection capabilities while minimizing classification anomalies. These results could contribute to securing potentially dangerous areas.