Abstract:
One of the most important economic activities in the global agricultural sector, groundnut
production is threatened by diseases involving leaves including; Leaf Spot, Alternaria Leaf
Spot, Rust, and Rosette. Old approaches to disease diagnosis are costly, cumbersome, and
slow, especially in intensive farming businesses. This work examines the use of transfer
learning in automating and improving the detection and classification of groundnut leaf
diseases. To address the class imbalance and augment the robustness of the model, men
and women from different ethnic backgrounds with a total of 1,720 images in five
categories were used to apply rescaling, rotation, and horizontal flipping. The following
deep learning base models: VGG19, InceptionV3, MobileNetV2, Exception, and
DenseNet201 models were fine-tuned and tested according to the accuracy, precision,
recall, and F1 score. Out of all the neural networks, Dense Net201 was the best since it got
the best test accuracy of 97.50%. The developed system offers a versatile, real-time disease
diagnosis solution for farmers to enhance their farming decisions, limiting pesticide
application, and increasing agricultural yield. This work contributes to both the
development of Al-assisted agricultural applications and the discussion of transfer
learning for solving practical problems.
