Document Details
|
Document Type |
: |
Thesis |
|
Document Title |
: |
Plant Disease Diagnoses Using Multi-level and Multi-scale Convolutional Neural Network تشخيص أمراض النبات باستخدام خصائص متعددة المستويات ومتعددة المقاييس للشبكة العصبية الترشيحية |
|
Subject |
: |
Faculty of Computing and Information Technology |
|
Document Language |
: |
Arabic |
|
Abstract |
: |
The impact of overlooked diseases on agricultural crops leads to substantial losses for farmers. However, manual field visits for plant disease diagnosis (PDD) prove to be expensive and time-consuming. Although several computerized PDD frameworks have been proposed, several challenges still persist, including early-stage leaf disease diagnosis, diverse disease classes, and computational complexity in diagnosis systems. This research presents a state-of-the-art Convolutional Neural Network (CNN)-based PDD framework named PDD-Net to overcome these challenges. The proposed framework incorporates data augmentation techniques and integrates multilevel and multiscale features to establish a class and scale-invariant architecture. To address gradient vanishing and exploding problems, PDD-Net employs the Flatten-T Swish (FTS) activation function. Additionally, the focal loss function is utilized to alleviate the impact of class imbalance during PDD-Net training. Experimental evaluation on the PlantVillage dataset shows superior performance of the PDD-Net method compared to baseline models, achieving an average precision of 92.06%, an average recall of 92.71%, an average F1 score of 92.36%, and an accuracy of 93.79%. Furthermore, on the cassava leaf disease dataset, the PDD-Net method demonstrates an average precision of 86.41%, an average recall of 85.77%, an average F1 score of 86.02%, and an accuracy of 86.98%. These results serve as compelling evidence for the efficiency and robustness of PDD-Net in plant disease diagnosis. By addressing critical challenges in PDD, the proposed framework offers a reliable and automated solution for accurately detecting plant diseases, enabling timely intervention and minimizing economic losses for farmers.
Key Word: Plant disease diagnosis, Convolutional neural network, Multi-level features, Multi-scale features, Classification. |
|
Supervisor |
: |
Dr. Turki Talal Turki |
|
Thesis Type |
: |
Master Thesis |
|
Publishing Year |
: |
1445 AH
2023 AD |
|
Added Date |
: |
Wednesday, December 20, 2023 |
|
Researchers
| حامد أحمد الغامدي | Alghamdi, Hamed Ahmed | Researcher | Master | |
|
Back To Researches Page
|