Qualifying Lemon Scab Severity: Autoencoder and XGBoost in a Hybrid Approach

The proposed research creates an innovative technique for the accurate severity detection that has been pioneered in the case of Mexican lemons and therefore becomes essential when evaluating citrus quality within agricultural procedures. The study focuses on developing a deep learning model capable of accurately categorizing Mexican lemons into four severity levels: low severity, medium-level severity, and high severity. With this impressive overall accuracy rate of 97.19%, the autoencoder and XGBoost classification model showed a very good performance in distinguishing the different respective severity classes considered for cleanly defining these levels toward citrus diseases, demonstrating its capability to separate such extreme differences. The performance of the model’s classification capabilities was meticulously analyzed using precision, recall, F1-score, and also accuracy metrics at various levels of severity. Similarly, performance insights of the classification from the confusion matrix served to model some areas that required improvement. The ability of the model to precisely classify severity is a very central decision aid for citrus growers, offering an in-depth evaluation of orchard health status. Finally, this study underscores the necessity of accurate classification severity in citrus farming and incorporates it into the disease management approaches as well as quality control cultures as a fundamental resource. Finally, this model promises to revolutionize the citrus production by giving growers the ability to act beforehand in disease defense and crop quality improvement.