DEEP NEURAL NETWORK BASED IDENTIFICATION AND CLASSIFICATION OF SWEET ORANGE LEAF DISEASES

Abstract

Historically and currently, more than 60% of the world's population relies on agriculture to earn a living, making it an important source of livelihood for them. Plant diseases are still a major problem, however, and are responsible for up to 25% loss in agricultural production annually. In this study, a novel approach was taken using Convolutional Neural Network (CNN) and transfer learning for the detection and classification of sweet orange ailments. The structure of the proposed method is inspired by EfficienetNet-B0 pre-trained model with the addition of a new set of fully connected (FC) layers and a final layer for fine tuning the model for the new task. The model was trained under various sets of image classes, such as binary classification, six and eleven classifications, and each image class had a moderate number of images. Prior to training, preprocessing and augmentation techniques (rotation, flipping, etc.), were applied to enhance the dataset. The model trained was then tested with new, unseen images. The results of the evaluation reveal that the proposed method can accurately classify the sweet orange leaf diseases, with average accuracy of 96.20%, 97.45%, and 96.54% for binary, six-class and eleven-class classifications, respectively. These results demonstrate the model's strength and stability at various classification levels, and its 2.17% accuracy improvement over the latest state-of-the-art models. The proposed leaf ailment detection method can be an effective and secure way to reduce losses from leaf ailments and food security. The model is able to detect multiple types of plant diseases even at the initial stage of infection. The findings of this study may point to the use of a deep learning technique in the early detection of plant diseases, which will help to control the ailments in the crop sector effectively and automatically.