Scholarlink

Agriculture is the backbone of a country’s economy and food production. Crop diseases cause economic instability and a decrease in agricultural production. Maize, one of the most cultivated crops in the world, is prone to various diseases that adversely affect its yield and quality. Most farmers face challenges in controlling and detecting these diseases. Thus, early detection of diseases is essential for farmers to avoid further losses. To overcome this challenge, this study focuses on deep learning techniques such as EfficientNetV2B2, ResNet50, InceptionV3, VGG16, and Xception for maize crop disease detection. It uses a maize crop image dataset from the Nelson Mandela African Institution of Science and Technology and the Tanzania Agricultural Research Institute. The dataset consists of 17,277 images divided into three classes, namely healthy, Maize Lethal Necrosis (MLN), and Maize Streak Virus (MSV). Although all the models achieved promising results, EfficientNetV2B2 showed the overall highest accuracy, reaching 92%. Finally, for transparent decision-making, Gradient Weighted Class Activation Mapping (Grad-CAM), an Explainable Artificial Intelligence (XAI) technique, was integrated with EfficientNetV2B2 to enhance model interpretation. The results of this research are poised to develop AI applications in agriculture, enabling timely and transparent diagnosis of maize diseases.

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

Unveiling Agricultural Insights: Optimizing Transfer Learning Models with Grad-CAM to Improve Maize Disease Detection

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Basit Hussain,
  • Malika Muradi,
  • Christian Boateng,
  • Eliya Christopher Nandi,
  • Imenagitero Ulysse Tresor,
  • Annajiat Alim Rasel

摘要

Agriculture is the backbone of a country’s economy and food production. Crop diseases cause economic instability and a decrease in agricultural production. Maize, one of the most cultivated crops in the world, is prone to various diseases that adversely affect its yield and quality. Most farmers face challenges in controlling and detecting these diseases. Thus, early detection of diseases is essential for farmers to avoid further losses. To overcome this challenge, this study focuses on deep learning techniques such as EfficientNetV2B2, ResNet50, InceptionV3, VGG16, and Xception for maize crop disease detection. It uses a maize crop image dataset from the Nelson Mandela African Institution of Science and Technology and the Tanzania Agricultural Research Institute. The dataset consists of 17,277 images divided into three classes, namely healthy, Maize Lethal Necrosis (MLN), and Maize Streak Virus (MSV). Although all the models achieved promising results, EfficientNetV2B2 showed the overall highest accuracy, reaching 92%. Finally, for transparent decision-making, Gradient Weighted Class Activation Mapping (Grad-CAM), an Explainable Artificial Intelligence (XAI) technique, was integrated with EfficientNetV2B2 to enhance model interpretation. The results of this research are poised to develop AI applications in agriculture, enabling timely and transparent diagnosis of maize diseases.