An Efficient Approach for the Classification of Leaf Diseases Using MobileNet Based CNN Model

On-time detection of abnormalities in the plants and taking appropriate measures to avoid the disease spreading leads to improvement of crop yields. Artificial intelligence and machine learning techniques overcome the challenges of other methods for disease detection with the ability of their high-speed computations. In this study, tomato leaf diseases were classified using transfer learning of deep convolutional neural networks (CNN) with the images obtained from PlantVillage dataset. The performance of CNN models was calculated using parameters such as accuracy, precision, recall and F1-score. Among the pre-trained networks, MobileNetV2 has the best classification accuracy of 98.14%. Also, proposed a new model to improve disease classification accuracy by adding GlobalAveragePooling and extra dense layers to MobileNetV2. Though the addition of these layers increases the model complexity, the accuracy of disease classification was improved to 99.04%. Also, the proposed model successfully classify 24 classes of plant leaves in PlantVillage dataset with an accuracy of 98.93%.