Mango leaf disease detection using image processing

Abstract

Bangladesh is an agricultural country and mango cultivation plays a significant role in the economy of Bangladesh. Mango trees are at risk of different kinds of leaf disease. As a result, it can be the reason for hindering food production and quality substantially. So, it is very much important for the farmers to timely detection of these diseases. As a result, farmers can ensure stable production and supply. So, in this thesis, we have provided a custom convolutional neural network (CNN) architecture that was designed especially for mango leaf disease detection in Bangladesh. Our dataset consists of over 7,535 images that show both affected and healthy mango leaves, exposing nine different leaf classifications. We have trained our custom CNN model through both healthy and sick images so that it can easily distinguish between affected and non-affected mango leaves. We have compared our custom CNN model with a few pre-trained models which are MobileNetV2, VGG16, DenseNet169, and InceptionV3 to evaluate our model’s performance and accuracy. So, the main motive of our thesis is to overcome the limitations of the previous research. Therefore, our suggested work is very much determined to be very accurate and to solve critical issues earlier researchers might have faced.