Application of Transfer Learning to Convolutional Neural Network Models for Mango Leaf Disease Recognition

Abstract

It is vital to first determine whether or not plant illnesses are there, and then to make steps to restrict the spread of those diseases in order to maximize both the quality and quantity of the harvest. First, it is important to determine whether or not plant illnesses are present. There are a number of advantages to mechanizing plant diseases, one of which is reducing the amount of time spent manually examining crops in a big agricultural area that produces a significant amount of mango. As a result of the fact that leaves are in charge of the majority of a plant’s nutrition absorption, it is very important to diagnose leaf diseases in a timely and precise manner. In this particular research, we classified and identified the several illnesses that may be harmful to mango leaf by using CNN. We employ multiple CNN models that have been trained via transfer learning in order to increase the quality of the results obtained from the training set. These CNN models include DenseNet201, InceptionResNetV2, InceptionV3, ResNet50, ResNet152V2, and Xception. Acquiring pictures, segmenting those images, and deriving features from them are all stages that are included in the process of sickness diagnosis. The collection contains approximately a thousand photographs, all of which depict either healthy mango leaves or mango leaves affected by illness. According to the findings of our investigation into the overall performance matrices, the DenseNet201 model earned the highest level of accuracy (98.00%) compared to all of the other models.