MDISL: Experimental Evaluation of Learning Based Potato Plant Disease Detection Using Modified Deep Image Segmentation Logic

Using a combination of AutoEncoder and the Xception model, this research proposes a unique method for the identification and classification of potato plant diseases, notably Early Blight and Late Blight. To effectively diagnose illnesses from potato leaf photos, a sophisticated deep learning (DL) technology is incorporated into the suggested methodology. Three thousand photos of healthy, Early Blight, and Late Blight leaves were included in the collection, which was taken from the Plant Village database. To guarantee consistent input data, preprocessing methods including picture scaling, normalization, and data augmentation were used. The AutoEncoder was utilized for feature extraction and dimensionality reduction, while the Xception model performed the classification task. The AutoEncoder-Xception model achieved a testing accuracy of 97.2%, outperforming traditional machine learning (ML) models such as Random Forest (RF) (91.5%), SVM (89.3%), and KNN (87.1%). The model demonstrated superior performance in terms of precision, recall, and F1-score, with values consistently above 95% across all classes. Confusion matrix analysis further confirmed the model's robustness, with minimal misclassifications observed.