Intelligent multi-class classification and diagnosis of citrus leaf diseases using deep convolutional network

Citrus plants, worthwhile to global agriculture, have their productivity drastically reduced because of diseases such as citrus canker, black spot, and greening. The actual diagnosis of these diseases requires a lot of technical expertise and takes considerable time; therefore it is impractical for extensive monitoring. This work proposes achieving an automated detection system using deep learning techniques for citrus leaf disease classification with four categories in the dataset, namely, canker, black spots, greening, and healthy. The dataset was augmented, thus improving model robustness by generating images. The system was developed using EfficientNetB0, which gives a good balance between accuracy and speed of the computational process. There was training and validation using k-fold cross-validation to ensure generalization. The model achieved test accuracy of 93%, supported by good precision, recall, and F1-scores across all classes. This study revealed that, for farmers, deep learning can be a trustworthy tool as far as fast and accurate disease recognition is concerned as required in precision agriculture.