Scholarlink

<p>Agriculture holds significant importance in various countries, particularly in India. India ranks as the largest guava producer globally and the fifth largest apple producer. However, the crops, including apple and guava leaves, face persistent threats from pests like cutworms. These pests primarily target the leaves, stalks, and stems, causing severe damage and potential destruc- tion of the entire plant. Detecting pest infestations early on enables farmers to take necessary precautions and limit the spread of the infestation. In this study, we present an end-to-end pipeline for assessing disease severity in in- fected leaves. We employ a pretrained ResNet-50 model to identify infected leaves and introduce the SegLearner model, which accurately detects and delineates the infected areas in the leaf images. Subsequently, the severity of damage in an infected leaf is determined by pixel counting method. The model is tested on the prepared Mask-LeafDataset and ADE20K dataset. Experimental results show that our method has significantly improve the mIoU as compared with other state-of-the-art. We achieve 91.4% as highest mIoU on Mask-LeafDataset and 45.8% mIoU on the ADE20K dataset.</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

SegLearner: A segmentation based approach for predicting disease severity in infected leaves

  • 期刊论文

  • 发表日期：

    2025年4月16日

查看全文
  • Nidhi Upadhyay,
  • Neeraj Gupta

摘要

Agriculture holds significant importance in various countries, particularly in India. India ranks as the largest guava producer globally and the fifth largest apple producer. However, the crops, including apple and guava leaves, face persistent threats from pests like cutworms. These pests primarily target the leaves, stalks, and stems, causing severe damage and potential destruc- tion of the entire plant. Detecting pest infestations early on enables farmers to take necessary precautions and limit the spread of the infestation. In this study, we present an end-to-end pipeline for assessing disease severity in in- fected leaves. We employ a pretrained ResNet-50 model to identify infected leaves and introduce the SegLearner model, which accurately detects and delineates the infected areas in the leaf images. Subsequently, the severity of damage in an infected leaf is determined by pixel counting method. The model is tested on the prepared Mask-LeafDataset and ADE20K dataset. Experimental results show that our method has significantly improve the mIoU as compared with other state-of-the-art. We achieve 91.4% as highest mIoU on Mask-LeafDataset and 45.8% mIoU on the ADE20K dataset.