Scholarlink

Tapioca is an essential food crop widely cultivated by millions of producers in the tropical regions due to the grain’s feature of being drought hardy and grow in marginal soils. But the diseases which affect the cassava leaves include the Cassava Mosaic Disease (CMD), Cassava Brown Streak Disease (CBSD), Cassava Bacterial Blight (CBB), and Cassava Green Mite (CGM) pose a great threat to the productivity of cassava. In this study, I look at the various effects caused by dataset size on cassava disease identification using deep learning. An experiment was performed on a low image dataset of 9,430 images, or 5 classes, and a high image dataset of 21,397 images, or 5 classes. An experiment of these included six architectures (DenseNet121, MobileNetV2, ResNet50, Xception, InceptionV3, and VGG16) for the small set, and three (MobileNetV2, InceptionV3, and VGG19-BN) for the large set. DenseNet121 presented the best result for the small dataset, having 90.5% of accuracy and a compact model size of 8.1M parameters. In the case of the large dataset, the specifications overall predicted well with an 83.9% accuracy rate, and F1 of 0.8484 clearly shows the effect of large dataset size on the model performance. This analysis illustrates pro and cons of a high accuracy and low computational complexity, and the differences in model behaviours when deployed on different datasets are described in detail. Thus, the results contribute usable knowledge of how to implement precision agriculture solutions according to available resources and farm scale.

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

Comparative Analysis of Deep Learning Models for Cassava Disease on Small and Large Datasets

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Seema Bohra,
  • Mamta Rani,
  • Pawan Singh

摘要

Tapioca is an essential food crop widely cultivated by millions of producers in the tropical regions due to the grain’s feature of being drought hardy and grow in marginal soils. But the diseases which affect the cassava leaves include the Cassava Mosaic Disease (CMD), Cassava Brown Streak Disease (CBSD), Cassava Bacterial Blight (CBB), and Cassava Green Mite (CGM) pose a great threat to the productivity of cassava. In this study, I look at the various effects caused by dataset size on cassava disease identification using deep learning. An experiment was performed on a low image dataset of 9,430 images, or 5 classes, and a high image dataset of 21,397 images, or 5 classes. An experiment of these included six architectures (DenseNet121, MobileNetV2, ResNet50, Xception, InceptionV3, and VGG16) for the small set, and three (MobileNetV2, InceptionV3, and VGG19-BN) for the large set. DenseNet121 presented the best result for the small dataset, having 90.5% of accuracy and a compact model size of 8.1M parameters. In the case of the large dataset, the specifications overall predicted well with an 83.9% accuracy rate, and F1 of 0.8484 clearly shows the effect of large dataset size on the model performance. This analysis illustrates pro and cons of a high accuracy and low computational complexity, and the differences in model behaviours when deployed on different datasets are described in detail. Thus, the results contribute usable knowledge of how to implement precision agriculture solutions according to available resources and farm scale.