Classification of leaf spot diseases in banana using pre-trained convolutional neural networks

Banana is a leading fruit crop in the global market and is grown all over the world. However, its production and trade are severely affected by the diseases caused by fungi, bacte- ria and viruses. Early diagnosis and management of such diseases are essential to avoid the yield loss. This paper demonstrates a deep learning based automated algorithm for the classification of three important leaf spot diseases in banana namely, Sigatoka, Cordana and Deightoneilla. Images of banana leaves infected with these diseases have been applied to four augmented pre-trained convolutional neural networks and their performance in disease classification is compared. An accuracy of 91.7% is achieved on the model with DenseNet 121 backbone.