MangoLeafDTR: ResCANet for mango leaf disease detection with rule-based treatment recommendation

Abstract

Mango farming faces seasonal leaf diseases that affect the quality and quantity of production. Usually, manual examination is used to diagnose various kinds of disease. Sometimes, due to a lack of early prediction of these diseases causes a lot of harm to the mango farmers. We introduce an AIdriven system to detect Bangladeshi mango leaf diseases and recommend treatments. Our model ResCANet is built on a custom convolutional neural network with residual block and channel attention mechanisms. It is trained with a primary dataset to detect mango leaf diseases such as Anthracnose, Sooty Mold, White Scale, and Healthy leaves. A rule-based system recommends treatments for farmers. While collecting the dataset, we mainly focused on the pre-farming time when the mango trees are flowering, as many kinds of diseases occur at that time. We collected the dataset throughout the whole months of March and April. At this time farmer manually inspects each disease and sprays medicine so that no diseases can harm the flowers. All recommendations are made under the close supervision of an agricultural expert and farmers. Our system is efficient for real-world use in farms. It achieved an accuracy of 96%. This framework detects the problem and recommends treatment.