Citrus Leaf Disease Prediction: Deep Feature Fusion Perspective

Citrus disease is a major problem of economic and productive loss in agriculture worldwide. Crops are affected by uneven climatic conditions, which leads to lower agricultural yields. This will affect the world’s agricultural economy. This paper proposes an image recognition method based on machine learning (ML) and deep learning (DL) feature fusion techniques for citrus leaf disease prediction. The three most popular CNN models were considered. The CNN models were AlexNet, vgg16, and vgg19. The features of these CNN models were extracted and fed into the support vector machine (SVM) to classify citrus leaf diseases. These classification models were evaluated in terms of performance metrics calculation. Subsequently, the multilayer feature fusion technique was adapted to the best-performing CNN models. Here, vgg16 performed the best among AlexNet, vgg16, and vgg19. Then, the different combinations of feature fusion of vgg16 are adapted. These fused features were fed into the SVM for citrus leaf disease classification. The feature fusion of ‘fc6 plus fc7’ of vgg16 with SVM yielded the best results; that is, the maximum values of accuracy, sensitivity, specificity, precision, FPR, F1 score, MCC, and Kappa coefficient were 97%, 97%, 98.09%, 97.06%, 1.58%, 95.34%, 84.4%, and 86.79% respectively. This method is very much helpful to the farmer as well as to the customers.