Scholarlink

<p>One of the important tools used by Natural Disaster Mitigation and Management (NDMM) in recent times is drone surveillance, particularly effective in disaster-affected areas. Drones equipped with modern imaging capabilities offer enhanced coverage and accessibility in disaster-affected areas, making them particularly effective in locating individuals trapped during floods and earthquakes. Traditional human-operated video analysis has limitations such as low processing speed, insufficient dataset resulting in poor performance, higher implementation cost, longer computation time, and so on. To address the traditional approach's complications posed in complex search and rescue scenarios, this paper suggest a Double Transfer Bidirectional Feature Pyramid-based YOLOv8. This innovative model leverages the Bidirectional Feature Pyramid to improve the extraction of image features through the pyramid with features at multiple scales, and uses a Double Transfer Strategy to transfer the pre-trained models to different operational environments. YOLOv8 architecture is employed due to its proven effectiveness in achieving high-speed and precise object detection under real-time constraints. These models bring a considerable enhancement of the accuracy and stability of the person detection in aerial images. Performance analysis proved that the proposed model achieves 99.68% accuracy on the 4&#xa0;K Drone, which is more efficient than the current methods. Moreover, the mean inference time of the model is lower to only 0.30&#xa0;ms of the total time per image, which may be considered a great improvement compared to the time-consuming traditional approaches. The evaluation process included cross-validation and testing across diverse disaster scenarios to ensure the robustness and reliability of the results. The proposed model is more accurate and with less execution time in real-time SAR applications, making it an essential tool for quick detection and leads to individuals during natural disasters. The model also enhances emergency response efforts, demonstrating its potential to save lives in critical situations.</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

Enhancing spatial accuracy in disaster response: a DTBiFP-YOLOv8 model for drone-based search and rescue operations

  • 期刊论文

  • 发表日期：

    2025年2月7日

查看全文
  • Siva Priya M S,
  • M. Vidhyalakshmi,
  • Kirankumar Manivannan

摘要

One of the important tools used by Natural Disaster Mitigation and Management (NDMM) in recent times is drone surveillance, particularly effective in disaster-affected areas. Drones equipped with modern imaging capabilities offer enhanced coverage and accessibility in disaster-affected areas, making them particularly effective in locating individuals trapped during floods and earthquakes. Traditional human-operated video analysis has limitations such as low processing speed, insufficient dataset resulting in poor performance, higher implementation cost, longer computation time, and so on. To address the traditional approach's complications posed in complex search and rescue scenarios, this paper suggest a Double Transfer Bidirectional Feature Pyramid-based YOLOv8. This innovative model leverages the Bidirectional Feature Pyramid to improve the extraction of image features through the pyramid with features at multiple scales, and uses a Double Transfer Strategy to transfer the pre-trained models to different operational environments. YOLOv8 architecture is employed due to its proven effectiveness in achieving high-speed and precise object detection under real-time constraints. These models bring a considerable enhancement of the accuracy and stability of the person detection in aerial images. Performance analysis proved that the proposed model achieves 99.68% accuracy on the 4 K Drone, which is more efficient than the current methods. Moreover, the mean inference time of the model is lower to only 0.30 ms of the total time per image, which may be considered a great improvement compared to the time-consuming traditional approaches. The evaluation process included cross-validation and testing across diverse disaster scenarios to ensure the robustness and reliability of the results. The proposed model is more accurate and with less execution time in real-time SAR applications, making it an essential tool for quick detection and leads to individuals during natural disasters. The model also enhances emergency response efforts, demonstrating its potential to save lives in critical situations.