Scholarlink

Weed growth in crops represents a challenge for farmers as this can affect plant thriving due to nutrient stealing. This paper presents a tool for weed detection in a sunflower field using computer vision techniques. To this end, regions of interest were extracted from 25 multispectral images; 20 vegetation indices were used to characterize the classes Background, Weed and Sunflower. Afterwards, Correlation Analysis (CA), Principal Component Analysis (PCA), AutoEncoder (AE), CA-PCA and CA-AE techniques were applied to create 5 datasets to train the Support Vector Machine (SVM), K-Nearest Neighbors (KNN) and Naive Bayes (NB) classifiers. The classifier that obtained the best separation between the classes Background, Weed and Sunflower was the SVM classifier applied on the PCA set based on 3 principal components, with an Accuracy of 81.5%, Precision of 81%, Recall of 81%, F1 of 81% and Cohen Kappa of 72%.

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

Weed Detection in a Sunflower Field Using Supervised Learning Techniques

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • A. Mexicano,
  • J. C. Carmona,
  • S. Cervantes,
  • K. Bee,
  • P. N. Montes

摘要

Weed growth in crops represents a challenge for farmers as this can affect plant thriving due to nutrient stealing. This paper presents a tool for weed detection in a sunflower field using computer vision techniques. To this end, regions of interest were extracted from 25 multispectral images; 20 vegetation indices were used to characterize the classes Background, Weed and Sunflower. Afterwards, Correlation Analysis (CA), Principal Component Analysis (PCA), AutoEncoder (AE), CA-PCA and CA-AE techniques were applied to create 5 datasets to train the Support Vector Machine (SVM), K-Nearest Neighbors (KNN) and Naive Bayes (NB) classifiers. The classifier that obtained the best separation between the classes Background, Weed and Sunflower was the SVM classifier applied on the PCA set based on 3 principal components, with an Accuracy of 81.5%, Precision of 81%, Recall of 81%, F1 of 81% and Cohen Kappa of 72%.