Scholarlink

<p>This paper offers an extensive detection architecture centered upon an AI-based compiler, using an approach based on YOLO. The proposed system seeks to deliver esoteric efficiency and accuracy with power consumption, productively allowing for real-time execution. Substantial analysis was conducted with the Pascal VOC dataset alongside running the standard YOLOv8 and YOLOv11 deep-learning models using the nature-inspired optimizer, hybrid PSO-GWO on the Pascal dataset, and juxtaposing for deep understanding. The proposed YOLO-World compiler achieved an mAP@0.5 of 91.5%, surpassing the corresponding YOLOv8 performance of 83.3% mAP@0.5 and the YOLOv11 performance of 84.42% mAP@0.5; it also demonstrated exceptional performance on other quantitative metrics, including an F1-score of 0.8307, a Recall of 90.47%, and an IoU of 92%. Optimized analyses further underscored the total feasibility of the model, achieving a speed of 185.18 FPS, a processing time of 0.0054&#xa0;s, and a run-time complexity reduction of 98.44%, which highlighted consistency and produced clear output. These increases are assessed based on the overall performance of the proposed architectural system, which includes error-free detection, real-time spatial visioning, and gains from resource leasing. The work also discusses the next steps and implementations related to studies on metaheuristic optimizations aimed at finding possible avenues to enhance the performance of computational infrastructures.</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

Real-time object detection using YOLO variants with nature-inspired algorithm

  • 期刊论文

  • 发表日期：

    2026年1月6日

查看全文
  • Jappanjot Kaur,
  • Law Kumar Singh

摘要

This paper offers an extensive detection architecture centered upon an AI-based compiler, using an approach based on YOLO. The proposed system seeks to deliver esoteric efficiency and accuracy with power consumption, productively allowing for real-time execution. Substantial analysis was conducted with the Pascal VOC dataset alongside running the standard YOLOv8 and YOLOv11 deep-learning models using the nature-inspired optimizer, hybrid PSO-GWO on the Pascal dataset, and juxtaposing for deep understanding. The proposed YOLO-World compiler achieved an mAP@0.5 of 91.5%, surpassing the corresponding YOLOv8 performance of 83.3% mAP@0.5 and the YOLOv11 performance of 84.42% mAP@0.5; it also demonstrated exceptional performance on other quantitative metrics, including an F1-score of 0.8307, a Recall of 90.47%, and an IoU of 92%. Optimized analyses further underscored the total feasibility of the model, achieving a speed of 185.18 FPS, a processing time of 0.0054 s, and a run-time complexity reduction of 98.44%, which highlighted consistency and produced clear output. These increases are assessed based on the overall performance of the proposed architectural system, which includes error-free detection, real-time spatial visioning, and gains from resource leasing. The work also discusses the next steps and implementations related to studies on metaheuristic optimizations aimed at finding possible avenues to enhance the performance of computational infrastructures.