Convolutional neural network architecture for detection and classification of diseases in fruits

Artificial intelligence is now becoming a part of people’s everyday lives. It can help farmers detect any disease in the early stage and take pre-emptive actions to save their crops and control disease spread, thus preventing crop wastage as well as increasing their income. The present study uses a combination of 13 convolutional neural network (CNN) models to classify five types of fruits and their leaf images into 41 classes, including diseased and healthy. Results show that the average accuracy of this CNN architecture is above 90% for all 13 individual models. One of the CNN models has been compared with three pre-trained models, i.e. MobileNet, DenseNet121 and InceptionV3 trained using the same dataset. It shows that the CNN architecture used in this study has higher accuracy while also being simple and easy to train.