Scholarlink

<p>Cotton, popularly known as “White-Gold” in India, accounts for about 23% of global production and significantly contributes to the Indian economy. However, it is highly susceptible to the infestation of various harmful insect-pests, which can cause substantial crop damage and yield reductions of up to 40–50%. In this context, we developed a lightweighted densely connected deep learning model for identifying insect-pests of cotton crop using RGB images. We collected 5,559 images of insect-pest infested cotton crops under natural field conditions across Indian agricultural farms. However, for enhancing the training images and to minimize the risk of overfitting the model, we applied a variety of image augmentation methods, including flipping, rotation, zooming, etc. The proposed model, with 83 layers, including four dense blocks and three transition layers, achieved around 99.24% of classification accuracy having 15&#xa0;s/epoch of training time, outperforming the other pretrained models. Furthermore, Grad-CAM visualization technique was used for demonstrating the model’s effectiveness and efficiency in multiclass classification of cotton insect-pest images. This model offers a practical tool for farmers to manage pest infestations and improve cotton crop yield.</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

A lightweighted densely connected network for insect-pest identification in cotton crop

  • 期刊论文

  • 发表日期：

    2025年8月9日

查看全文
  • Shalini Kumari,
  • Sudeep Marwaha,
  • Md. Ashraful Haque,
  • Harsh Sachan,
  • Chandan Kumar Deb,
  • Shashi Dahiya,
  • Alka Arora,
  • P. R. Shashank

摘要

Cotton, popularly known as “White-Gold” in India, accounts for about 23% of global production and significantly contributes to the Indian economy. However, it is highly susceptible to the infestation of various harmful insect-pests, which can cause substantial crop damage and yield reductions of up to 40–50%. In this context, we developed a lightweighted densely connected deep learning model for identifying insect-pests of cotton crop using RGB images. We collected 5,559 images of insect-pest infested cotton crops under natural field conditions across Indian agricultural farms. However, for enhancing the training images and to minimize the risk of overfitting the model, we applied a variety of image augmentation methods, including flipping, rotation, zooming, etc. The proposed model, with 83 layers, including four dense blocks and three transition layers, achieved around 99.24% of classification accuracy having 15 s/epoch of training time, outperforming the other pretrained models. Furthermore, Grad-CAM visualization technique was used for demonstrating the model’s effectiveness and efficiency in multiclass classification of cotton insect-pest images. This model offers a practical tool for farmers to manage pest infestations and improve cotton crop yield.