Ensembled AttenNet: A Novel Deep Learning Approach for Mango Leaf Disease Detection

Abstract

Agriculture is severely threatened by plant diseases, which can result in poor quality and quantity of food. Because of the rapid spread of plant diseases and the vast areas affected, it is difficult to control them effectively. Several leaf diseases affect the mango industry particularly severely, reducing fruit yields and quality. In order to prevent and treat these diseases effectively, it is crucial to identify them quickly and correctly. In this study, attention methods and multi-scale feature fusion will be used to improve automatic detection of mango leaf diseases. To extract the features, we use a pretrained VGG16 model. A number of common diseases were evaluated, including anthracnose, bacterial canker, weevils, dieback, gall flies, powdery mildew, sooty mold, and healthy leaves. Data used for classification comes from Kaggle's MangoleafBD dataset. In order to train and test the model, 4000 images were used. In addition to cross-validation methods, precision, recall, and F1 metrics are used to measure model resilience. An accuracy of up to 99% was achieved with the model. This model provides better results than state-of-the-art models.