Abstract:
Accurate and timely detection of plant diseases is crucial for sustainable agriculture
and food security. This research presents a real-time monitoring system utilizing deep
learning techniques to detect diseases in plant leaves with high accuracy. We combined
several plant datasets, including the PlantVillage Dataset, resulting in a comprehensive
dataset of 30,945 images across eight plant types (potato, tomato, pepper bell, apple,
corn, grape, peach, and rice) and 35 disease classes. Initially, a custom Convolutional
Neural Network (CNN) model was developed, achieving a leaf classification accuracy
of 95.62%. Subsequently, the dataset was partitioned for individual plant disease
detection, applying nine different CNN models (custom CNN, VGG16, VGG19,
InceptionV3, MobileNet, DenseNet121, Xception, and two hybrid models) to each
plant type. The highest accuracy rates for disease detection were: 100% for potato
(custom CNN), 98% for tomato (InceptionV3, custom CNN, VGG16), 100% for
pepper bell (MobileNet, custom CNN), 100% for apple (MobileNet, Xception), 98%
for corn (custom CNN), 99% for grape (custom CNN, VGG19, DenseNet121), 100%
for peach (VGG16, custom CNN), and 98% for rice (DenseNet121). A web and mobile
application were developed based on the best-performing models, allowing users to
insert or capture images of plant leaves, detect diseases, and receive treatment
suggestions with high confidence levels. The results demonstrate the effectiveness of
deep learning models in accurately identifying plant diseases, offering a valuable tool
for enhancing disease management and crop yields.
