A Transfer Learning-Based Approach Allium Sativum Disease Recognition and Classification

Abstract

Allium Sativum is a widely used spice noted for its culinary, medical, and spiritual purposes, including its roles as an antibacterial, anti-thrombotic, and antineoplastic agent. It has been integral to human civilization for thousands of years. In Bangladesh, Allium Sativum farming is economically significant, delivering major money to farmers and creating employment opportunities. It also encourages agricultural variety, and soil health lowers the hazards of monoculture through crop rotation. Allium Sativum has nutritional and health benefits due to its antibacterial characteristics and is a staple in Bangladeshi cuisine, assuring stable local demand and value addition through processed products. This study focuses on the classification of eleven Allium Sativum disorders using powerful deep-learning methods. The diseases classified include Basal Rot (Fusarium culmorum), Bloat Nematode (Ditylenchus dipsaci), Botrytis Rot (Botrytis porri), Bulb Mites (Rhizoglyphus spp--Tyrophagus spp), Bulb and Stem Nematode (Ditylenchus dipsaci), Disease Free, Leek Yellow Stripe (genus Potyvirus), Purple Blotch (Alternaria porri), Rust (Puccinia porri), and White Rot (Sclerotium cepivorum). Various convolutional neural network (CNN) architectures were employed, with Xception achieving the best test accuracy of 91.99%, exhibiting superior ability in diagnosing certain Allium Sativum disorders. Other models employed include MobileNetV2, InceptionV3, VGG16, and DenseNet121, each contributing to the robust identification of Allium Sativum diseases. This research emphasizes the potential of deep learning in strengthening agricultural practices by enabling accurate and fast disease identification, eventually helping farmers and the agricultural economy.