Screening of Citrus Diseases Using Deep Learning Embedders and Machine Learning Techniques

A tropical fruit with great economic significance, citrus is cultivated and farmed in almost 150 different nations. A total of about 157.98 million tons of citrus fruit are produced worldwide. Citrus plant diseases have a major negative impact on the citrus fruit market as well as production in terms of quantity as well as quality. Attention must be paid to citrus fruits and leaves in order to promptly diagnose and treat problems; failing to do so results in severe financial loss. Citrus plant diseases cause a yearly loss of about 50% of citrus fruits at citrus farming operations. In this study, to produce the vector representation of the images, we used three deep learning image embedders namely Inception V3, VGG16, and Squeeze net. We used five machine learning models, including Random Forest, KNN, Gradient Boosting, SGD, and Neural Network with 5-fold Cross Validation, to predict citrus diseases. Using the considered dataset, the classification performance of each independently tuned ML model across all of its ensembles has been obtained. The neural network with Inception V3 as image embedder has obtained maximum average accuracy of 96.6%, 96.6% recall, 96.5% precision, and 96.5% F1 Score. Finally, our prediction model offers a reliable technique for accurate detection of citrus diseases to assist farmers.