Abstract:
Wheat is a significant cereal crop for developing countries like Ethiopia. However, low
productivity persists due to factors such as weather conditions, diseases, and pests.
Traditional techniques of detecting wheat disease are laborious, consuming time, and
require solid knowledge of the fields. Therefore, many studies have attempted to
develop detection and classification models. Most of these studies are limited to a few
diseases, used a handcrafted feature extraction method, and took no remedy action for
the identified diseases. Therefore, the study’s aimed to construct an effective model for
wheat leaf diseases detection and classification by employing a transfer learning
approach. To achieve this, a total of 2,316 images of wheat leaf diseases, including five
different types of diseases like leaf rust, Septoria tritic blotchi, strip rust, powdery
mildow, and tanspot, are collected. Image preprocessing techniques, including resizing,
normalization, noise removal, and augmentation, are applied. A transfer learning based
pretrained models’ is utilized for extracting features and classifying the image. Four
pre-trained models’ extreme inception (Xception), Visual geometry group (Vgg16,
Vgg19), and Residual Network (ResNet50) are compared to determine best efficient
model that detects and classifies the diseases. Confusion matrices and classification
reports are used for evaluating effectiveness of the model. The suggested model attained
an impressive accuracy of 99.70% through fine-tuning and outperformed other models.
Additionally, a user-friendly web application is developed to assist experts and farmers
in detecting wheat leaf diseases and suggesting potential remedies to control the
diseases.
