Scholarlink

<p>Precise fruit recognition is important in today’s agriculture due to improved crop yield, quality measurement, and business performance. With artificial intelligence (AI) and computer vision (CV), farmers can track ripeness, identify early disease onset, and minimize post-harvest loss. With the ongoing increase in the demand for food worldwide, AI-driven solutions present new opportunities to beat traditional agricultural challenges. The intersection of intelligent automation, deep learning, and image processing has revolutionized conventional farming practices, increasing the use of smarter resources and sustainability. Technologies such as CV and AI enable automation, decreasing the eco-basis, and increasing monitoring of plant health, pests, crop yield estimation, and irrigation. This work compares the performance of best-of-breed YOLO (You Only Look Once) object detection models, YOLOv8 through to YOLOv12, for real-time guava fruit detection in natural outdoor farming environments. For additional model accuracy enhancement with dynamic environments like motion blur, non-uniform illumination, and sensor noise, a&#xa0;probabilistic fused Wiener filter (PFWF) was used during the image pre-processing stage. PFWF improves visual quality by denoising highly effective images without losing relevant features, providing cleaner input to the detection models. Models were tested with respect to accuracy, F1&#xa0;score, and intersection over union (IoU). The performances showed a&#xa0;common improvement across all versions, with the best results on YOLOv12 as follows: 89.9% accuracy, 88.2% F1&#xa0;score, and 84% IoU. The comparison presented here highlights the strength of integrating strong object detection models with sophisticated pre-processing methods such as PFWF and provides useful insights for the practical implementation of intelligent, robust, and scalable fruit detection systems in real agricultural scenarios.</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

Optimizing YOLO-Based Models for Real-Time Guava Detection with Probabilistic Fused Wiener Filter-Enhanced Feature Fusion

  • 期刊论文

  • 发表日期：

    2025年9月24日

查看全文
  • Siddharth Singh Chouhan,
  • Esha Saxena,
  • Anju Shukla,
  • Rajneesh Kumar Patel,
  • Uday Pratap Singh

摘要

Precise fruit recognition is important in today’s agriculture due to improved crop yield, quality measurement, and business performance. With artificial intelligence (AI) and computer vision (CV), farmers can track ripeness, identify early disease onset, and minimize post-harvest loss. With the ongoing increase in the demand for food worldwide, AI-driven solutions present new opportunities to beat traditional agricultural challenges. The intersection of intelligent automation, deep learning, and image processing has revolutionized conventional farming practices, increasing the use of smarter resources and sustainability. Technologies such as CV and AI enable automation, decreasing the eco-basis, and increasing monitoring of plant health, pests, crop yield estimation, and irrigation. This work compares the performance of best-of-breed YOLO (You Only Look Once) object detection models, YOLOv8 through to YOLOv12, for real-time guava fruit detection in natural outdoor farming environments. For additional model accuracy enhancement with dynamic environments like motion blur, non-uniform illumination, and sensor noise, a probabilistic fused Wiener filter (PFWF) was used during the image pre-processing stage. PFWF improves visual quality by denoising highly effective images without losing relevant features, providing cleaner input to the detection models. Models were tested with respect to accuracy, F1 score, and intersection over union (IoU). The performances showed a common improvement across all versions, with the best results on YOLOv12 as follows: 89.9% accuracy, 88.2% F1 score, and 84% IoU. The comparison presented here highlights the strength of integrating strong object detection models with sophisticated pre-processing methods such as PFWF and provides useful insights for the practical implementation of intelligent, robust, and scalable fruit detection systems in real agricultural scenarios.