Wheat Leaf Disease Classification from Images Using Deep-Learning Techniques
| dc.contributor.advisor | Melkamu Beyene (PhD) | |
| dc.contributor.author | Tatek Chane | |
| dc.date.accessioned | 2025-08-17T21:55:49Z | |
| dc.date.available | 2025-08-17T21:55:49Z | |
| dc.date.issued | 2025-01 | |
| dc.description.abstract | Ethiopia, the largest wheat-producing country in sub-Saharan Africa, relies heavily on wheat as a staple crop essential for food security and economic stability. However, wheat cultivation faces significant challenges due to various biotic and abiotic stressors, including diseases caused by pathogenic fungi, viruses, and adverse environmental conditions. Traditional methods to detecting and classifying illnesses of wheat leaves are labor-intensive, time-consuming, and often inaccurate, necessitating the development of more efficient and reliable approaches. This study explores the application of deep learning techniques for the early detection and classification of wheat leaf diseases, focusing on healthy leaves, brown rust, orange rust, and yellow rust. A custom convolutional neural network (CNN) model, along with pre-trained models such as VGG16, VGG19, InceptionV3, ResNet152, DenseNet201, and EfficientNetB7, was developed and evaluated using a dataset of 4,397 labeled wheat leaf images. Data preprocessing techniques, including resizing, normalization, and augmentation, were employed to improve the model's robustness. The research compared the performance of custom CNN architectures and pre-trained models using accuracy, loss, precision, recall, and inference speed metrics. Fine-tuning the pre-trained models enhanced their performance, with DenseNet201 achieving the highest accuracy of 99.80%. The study found that selecting a model involves balancing accuracy and computational efficiency and addressing challenges such as class confusion through dataset diversity and architecture optimization. The findings demonstrate the effectiveness of transfer learning models for wheat disease detection and offer valuable insights for agricultural applications, suggesting that future research should focus on expanding the dataset, exploring additional architectures, and implementing advanced augmentation methods to further improve model performance and adaptability. The experimental results demonstrate that the developed model can accurately identify and classify wheat leaf diseases. These experiments suggest that deep learning, particularly through transfer learning, can significantly improve the efficiency and precision of wheat disease management. | |
| dc.identifier.uri | https://etd.aau.edu.et/handle/123456789/6881 | |
| dc.language.iso | en_US | |
| dc.publisher | Addis Ababa University | |
| dc.subject | Wheat Leaf Disease | |
| dc.subject | Transfer Learning | |
| dc.subject | VGG16/19 | |
| dc.subject | InceptionV3 | |
| dc.subject | ResNet152 | |
| dc.subject | EfficientNetB7 | |
| dc.subject | and DenseNet201Classification | |
| dc.subject | Agricultural Technology | |
| dc.title | Wheat Leaf Disease Classification from Images Using Deep-Learning Techniques | |
| dc.type | Thesis |