Early detection of diseases in kitchen garden plants is critical for optimising yields and ensuring plant health. This research investigates the efficacy of state-of-the-art transfer learning models for accurate disease detection across five common plant species: Tomato, Beans, Bell Pepper, Ladyfinger, and Cauliflower. Five Transfer Learning models—DenseNet121, InceptionV3, MobileNetV2, ResNet50V2, and InceptionResNetV2—were evaluated using a curated dataset comprising images of healthy and diseased leaves. The Tomato dataset was sourced from the PlantVillage database, containing 25,336 images across ten classes, including nine disease categories and one healthy class. The Beans dataset, obtained from the Makerere University Beans Image Dataset, consists of 14,129 images with diverse disease manifestations. The Ladyfinger dataset comprises 1949 images from a publicly available source on Kaggle, depicting healthy leaves and those affected by Yellow Vein Mosaic Disease. The Cauliflower dataset includes 7360 images collected from VegNet, featuring images of healthy leaves and those affected by Downy Mildew, Black Rot, and Bacterial Spot Rot. Lastly, the Bell Pepper dataset consists of 2475 images from the PlantVillage database, with images classified as diseased or healthy. The findings indicate that MobileNetV2 outperformed the other models for Beans (96.75%) and Bell Pepper (99.60%), achieving an accuracy of 99.80% on the Tomato dataset. Furthermore, four models, including MobileNetV2, achieved 100% accuracy on the Cauliflower dataset. DenseNet121 demonstrated significant performance on Ladyfinger, tying with ResNet50V2 at 99.49%. However, InceptionResNetV2 exhibited inconsistent results, particularly with a low accuracy of 93.07% for the Beans dataset. This research, thus, focuses on collecting and preprocessing high-quality datasets for the selected plant species, demonstrating effective transfer learning models for leaf disease detection, and providing insights into the suitability of each architecture for specific datasets. Thus, by leveraging transfer learning techniques, this research contributes to the growing knowledge of agricultural technology, addressing a critical need for efficient disease management in kitchen gardens. The implications of this study suggest that integrating accurate plant disease detection models into mobile applications can provide timely information to kitchen gardeners.
Chauhan et al. (Sun,) studied this question.