Abstract:
Diseases impact a larger proportion of crops, which makes agricultural output difficult. Early illness detection and prediction can lead to increased production. The fruit guava is cultivated in tropical and subtropical regions, including South American countries, India, Pakistan, and Chad. Numerous diseases, such as Canker, Dot, Mummification, and Rust, can affect guava plants. A diagnosis made only by eye inspection is laborious and unreliable. Guava disease detection is vital in maintaining crop health and ensuring optimal yields. Traditional methods of disease identification, which rely on manual inspection, are often time-consuming and prone to human error. An automated diagnosis and prediction system is required to assist farmers in detecting plant disease early on. As a result, we created a deep-learning technique for guava leaf disease classification and prediction. A dataset of guava leaf images is processed and trained using these convolutional neural network (CNN) architectures to classify common guava diseases such as anthracnose, rust, and powdery mildew. The performance of both models is evaluated based on accuracy, precision, recall, and F1-score. We looked at a dataset that had 5430 leaf instances that were divided into six classes. Two distinct and widely favored pre-trained CNN architectures were used to train the dataset. With an accuracy of 95.83% on the test data, the VGG-16 architecture fared better than the InceptionV3 model. The outcomes demonstrate that deep learning techniques outperform conventional techniques in terms of success and dependability
