Mangoleaf: Disease Detection Using Machine Learning

Abstract

Mango (Mangifera indica) is the most commonly cultivated fruit crop in Bangladesh and a supporting pillar of agriculture in the nation as an income source. Mango yield is typically affected by a chain of leaf diseases such as anthracnose, bacterial canker, cutting weevil, die back, gall midge, powdery mildew, scooty mould disease resulting in poor yields and quality. Manual identification following traditional techniques takes a long time, is not very accurate, and is not very helpful in surveying in large-scale. The study takes into account the use of machine learning algorithms, i.e., CNN model and a CNN model paired with pooling layers for the automatic classification and detection of mango leaf diseases from image data. A representative set of diseased and healthy mango leaves was collected and preprocessed for training and testing. All the algorithms were compared on the major metrics of accuracy, precision, recall, and F1-score. Maximum accuracy 95% was achieved with CNN model paired with pooling layers. The social, environmental, and ethical consequences of implementing such smart farm systems in rural Bangladesh are also included in this thesis. Through enabling the identification of diseases at an early stage and with accuracy, the system under consideration can prevent the use of pesticides that are harmful to the health of humans and enable sustainable production and enhance the livelihood of the local farmers. This research contributes to the application of AI in agriculture and establishes a foundation for further data-based management of plant diseases appropriate for Bangladesh's agricultural sector.