Precision Agriculture: Guava Disease Diagnosis via CNN and Random Forest

Guava is an essential fruit in many areas, but production is hampered by common diseases. In order to accurately identify diseases in guava leaves, this research pioneers a novel approach combining Convolutional Neural Networks (CNN) and Random Forest. For training and validation, a dataset encompassing five main guava diseases is used. Three convolutional layers and max pooling are part of the model architecture, which is repeated for reliable feature extraction. A flattening layer follows this, and then the Random Forest classifier—which is set up with 12 decision trees—enters. With 98.9% in 5-fold cross validation and 99.2% in 10-fold cross validation, experimental results demonstrate exceptional accuracy. The model outperforms current methods, demonstrating its effectiveness in diagnosing diseases. Additional performance indicators include precision, recall, and F1-scores. The suggested methodology not only aids in early disease diagnosis but also assures guava farmers of significant financial gains by reducing yield losses. The GradCAM method also offers transparency, providing insightful information into the model's decision-making process. The dataset needs to be expanded, and real-time deployment options need to be investigated. This study represents an important advance in precision agriculture and may completely alter how guava disease is managed.