Potato leaf disease classification using fusion of multiple color spaces with weighted majority voting on deep learning architectures

Early identification of potato leaf disease is challenging due to variations in crop species, disease symptoms, and environmental conditions. Existing methods for detecting crop species and diseases are limited, as they rely on models trained and evaluated solely on plant leaf images from specific regions. This study proposes a novel approach utilizing a Weighted Majority Voting strategy combined with multiple color space models to diagnose potato leaf diseases. The initial detection stage employs deep learning models such as AlexNet, ResNet50, and MobileNet. Our approach aims to identify Early Blight, Late Blight, and healthy potato leaf images. The proposed detection model is trained and tested on two datasets: the PlantVillage dataset and the PLD dataset. The novel fusion and ensemble method achieves an accuracy of 98.38% on the PlantVillage dataset and 98.27% on the PLD dataset with the MobileNet model. An ensemble of all models and color spaces using Weighted Majority Voting significantly increases classification accuracies to 98.61% on the PlantVillage dataset and 97.78% on the PLD dataset. Our contributions include a novel fusion method of color spaces and deep learning models, improving disease detection accuracy beyond the state-of-the-art.