Analysis of Deep Learning Algorithms for Grape Leaf Disease Detection

Abstract

Plant leaf diseases are crucial in agriculture as they can affect food security. Grapes are one of the important fruits we consume for health considerations. This paper aims to investigate the various deep learning algorithms focused on plant disease detection. A systematic literature review was conducted to determine the top three deep learning algorithms to be utilized in this paper. It intends to compare the performance of the top three deep learning algorithms with appropriate performance metrics. The top three deep learning algorithms identified were Convolutional Neural Network, EfficientNet, and MobileNetV2. The grape leaf images from the PlantVillage dataset which comprise Black Rot, ESCA, and Leaf Blight were used in this paper. There are three thousand two hundred twenty-five (3225) images which were divided into 80% training, 10% testing, and 10% validation. Normalization, augmentation, and hyperparameters were implemented in the training. The results revealed that the EfficientNet model got 100% accuracy, while MobileNetV2 and CNN got 98% and 78% respectively. In terms of prediction, the EfficientNet and MobileNetV2 models also successfully predicted all three images while the Convolutional Neural Network model only predicted two images correctly. This finding suggests that the deep learning models EfficientNet and MobileNetV2 hold significant promise for enhancing image classification techniques in the detection of grape leaf diseases. Their application could lead to more accurate and efficient monitoring of grapevine health, potentially revolutionizing disease management practices in viticulture. This could result in earlier detection, improved yield management, and reduced reliance on chemical treatments, fostering sustainable agricultural practices and optimizing crop productivity.