Use of Trained Convolutional Neural Networks for Analysis of Symptoms Caused by Botrytis fabae Sard

Abstract

This study evaluated the use of convolutional neural networks (CNN) in agricultural disease recognition, specifically for Botrytis fabae symptoms. An experimental bean culture was used to capture images of healthy and affected leaflets, which were then used to perform binary classification and severity classification tests using several CNN models. The results showed that CNN models achieved high accuracy in binary classification, but performance decreased in severity classification due to the complexity of the task. InceptionResNet and ResNet101 were the models that performed best in this task. The study also utilized the Grad-CAM algorithm to identify the most significant B. fabae symptoms recognized by the CNNs. Overall, these findings can be used to develop a smart farming tool for crop production support and plant pathology research.

Keywords

• artificial intelligence
• deep learning
• botrys fabae sard
• severity scale
• neural networks
• agricultural disease
• inceptionresnet
• resnet101
• grad-cam
• agricultura inteligente
• investigación en patología vegetal