Scholarlink

The crop diseases are main threat to food security due to the inadequate infrastructure this area remains unaddressed. The Internet of Things (IoT) devices like drones and sensors are used to collect data on different environmental conditions like crop health, humidity, temperature, and soil moisture. This research proposed a Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) for wheat disease detection and crop prediction. The dataset is collected from Azad Kashmir district in Pakistan using sensors which includes different environmental conditions such as soil moisture, temperature, and humidity. The collected data is preprocessed by data augmentation technique which enhances the size of data. Here, different image augmentations are used like shear, vertical and horizontal flip, and height and weight flip. WGAN is capable of learning intricate features and correlations within the data through its adversarial training. The gradient penalty is used to avoid the discriminators to reduce the overfitting and allow the model to generalize better in unseen data. The WGAN-GP attained recall 98.52%, precision 98.89%, f1-score 98.70%, and accuracy 99.06% on collected dataset when compared to convolutional neural network (CNN) and MobileNet.

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

Internet of Things-Based Wheat Disease Detection and Prediction Using Wasserstein Generative Adversarial Network with Gradient Penalty

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • V. M. Aparanji,
  • P. S. Abdul Lateef Haroon,
  • E. Manju More,
  • K. Sudhakar,
  • R. Rana Veer Samara Sihman Bharattej

摘要

The crop diseases are main threat to food security due to the inadequate infrastructure this area remains unaddressed. The Internet of Things (IoT) devices like drones and sensors are used to collect data on different environmental conditions like crop health, humidity, temperature, and soil moisture. This research proposed a Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) for wheat disease detection and crop prediction. The dataset is collected from Azad Kashmir district in Pakistan using sensors which includes different environmental conditions such as soil moisture, temperature, and humidity. The collected data is preprocessed by data augmentation technique which enhances the size of data. Here, different image augmentations are used like shear, vertical and horizontal flip, and height and weight flip. WGAN is capable of learning intricate features and correlations within the data through its adversarial training. The gradient penalty is used to avoid the discriminators to reduce the overfitting and allow the model to generalize better in unseen data. The WGAN-GP attained recall 98.52%, precision 98.89%, f1-score 98.70%, and accuracy 99.06% on collected dataset when compared to convolutional neural network (CNN) and MobileNet.