Analysis of Effectiveness of Augmentation in Plant Disease Prediction using Deep Learning

Abstract

Crop diseases pose a significant threat to food production. Because of the widespread adoption of smartphone technology, it is now technically feasible to use various image processing techniques to identify the type of plant disease from a single picture. Detecting illness early will lead to more effective interventions to reduce the impact of crop diseases on the food supply. Image classification is the most important step required for disease prediction in plants and deep learning techniques are the most optimal techniques used for image classification in the current scenario. This paper analyzes three major transfer learning techniques namely InceptionV3, DenseNet169 and ResNet50 using augmentation and without augmentation for image classification and thereby plant disease detection. After applying the above mentioned techniques we analyzed the efficiency of the algorithm with the help of various quality metrics: precision, recall, accuracy, F1-score. The best model with highest accuracy is ResNet50 with 98.2 percent accuracy with augmentation and 97.3 percent accuracy without augmentation.