Plant leaf disease detection using CNN with transfer learning and XGBoost

Abstract

The plant leaf disease identification at an early stage is an important step to prevent economic and agricultural losses. It is a challenging task because infected areas are difficult to extract from different images of leaves. Here, the plant leaf disease detection method is proposed using a convolutional neural network that combines knowledge learned from the PlantVillage dataset using ResNet50 and XGBoost classifier, termed as ConRXG. The ResNet50 serves as a pre-trained model towards feature extraction for transfer learning with batch normalisation to prevent overfitting. The ReLu activation function and Adam optimiser are used to improve the accuracy of the model. This method demonstrates robustness while identifying the plant leaf diseases in the chosen dataset. ConRXG has the highest training accuracy (0.9865), validation accuracy (0.9730), recall (0.9950), and F1-score (0.8971) among all the existing baseline models.