Scholarlink

<p>With the current unprecedented interest in exploring the bounds of machine learning and computer vision systems in all domains, several areas of interest started resurfacing; among which is “Precision Agriculture”. The urge to find more efficient methods into preserving the crop yields and the ecological balance is what caused the articles discussing weed management to follow positive progressing trend. This review article provides a launching point for future experiments and projects in machine learning and computer vision systems tackling weed detection and classification. This review brings added value by consolidating in one place results reported in studies that used classical machine learning methods such as SVM, KNN, and ANN, as well as deep learning systems and customized models including VGG16, ResNet, Faster R-CNN, YOLO, DeepLabV3+, U-Net, WeedVision, YOLO-Weed Nano. Its scope covers weed detection, classification, and segmentation using datasets acquired from UAV imagery, field and greenhouse cameras, stereoscopic video, and public datasets such as DeepWeeds, Weed25, and CottonWeedDet12, in addition to crop-specific datasets from corn, soybean, cotton, rice, wheat, onion, and turfgrass environments. In conclusion, a positive trend is observed between the dataset size and model accuracy. Moreover, VGGNet scored the highest accuracy among all reviewed models within all of the discussed papers. The acquired results pave the way for numerous future works including autonomous weed management vehicles and the creation of dataset accommodating diverse agriculture fields applications.</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

Systematic review of autonomous weed detection and identification using deep learning and computer vision

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月17日

查看全文
  • Bahga Abdul-Hakeem,
  • Omar Fayez ElMahdy,
  • Sherif Mehanny

摘要

With the current unprecedented interest in exploring the bounds of machine learning and computer vision systems in all domains, several areas of interest started resurfacing; among which is “Precision Agriculture”. The urge to find more efficient methods into preserving the crop yields and the ecological balance is what caused the articles discussing weed management to follow positive progressing trend. This review article provides a launching point for future experiments and projects in machine learning and computer vision systems tackling weed detection and classification. This review brings added value by consolidating in one place results reported in studies that used classical machine learning methods such as SVM, KNN, and ANN, as well as deep learning systems and customized models including VGG16, ResNet, Faster R-CNN, YOLO, DeepLabV3+, U-Net, WeedVision, YOLO-Weed Nano. Its scope covers weed detection, classification, and segmentation using datasets acquired from UAV imagery, field and greenhouse cameras, stereoscopic video, and public datasets such as DeepWeeds, Weed25, and CottonWeedDet12, in addition to crop-specific datasets from corn, soybean, cotton, rice, wheat, onion, and turfgrass environments. In conclusion, a positive trend is observed between the dataset size and model accuracy. Moreover, VGGNet scored the highest accuracy among all reviewed models within all of the discussed papers. The acquired results pave the way for numerous future works including autonomous weed management vehicles and the creation of dataset accommodating diverse agriculture fields applications.