Deep Clustering Techniques for Plant Disease Discovery
| Author | AlKameli, Alya Sayed Husain Ahmed Husain | |
|---|---|---|
| metadata.dc.contributor.department | University of Bahrain ,College of Information Technology | |
| metadata.dc.date.accessioned | 2026-09-20T15:52:29Z | |
| Publication date | 2024 | |
| Degree level | Doctoral | |
| Abstract | Abstract : Plant disease detection is essential for maintaining food security and agricultural output. The complexity and diversity of plant disease datasets are often too much for traditional methods to handle, so new ideas are needed to improve disease discovery processes. In order to better control and monitor plant diseases, this thesis looks at the use of deep clustering approaches for disease discovery. It does this by utilizing the capabilities of cutting-edge machine learning techniques. Through a comprehensive exploration of deep clustering algorithms, including dynamic autoencoders (DynAEs), autoencoder with K-Means, First Integer Neighbor Clustering Hierarchy (FINCH), Variational Deep Embedding (VADE), deep embedded clustering (DEC), and Improved Deep Embedded Clustering (IDEC), this thesis examines their efficacy in accurately categorizing plant diseases. Our examinations illustrate that when compared to all five other deep clustering models, DynAE continues to yield better results in terms of ACC and NMI. A comparative study sheds light on the advantages and disadvantages of each of the aforementioned methods, offering insightful information on how well they function in actual situations. The key findings of this thesis reveal that DynAE significantly outperformed other state-of-the-art deep clustering methods, such as autoencoder with K-Means, FINCH, VADE, DEC, and IDEC, across four distinct plant disease datasets: potato, apple, corn, and grape The performance was evaluated using accuracy (ACC) and normalized mutual information (NMI) as metrics. Remarkably, DynAE achieved the highest accuracy of 72.21% on the potato dataset, indicating its superior ability to correctly cluster the data points compared to other methods. Additionally, DynAE recorded the highest NMI of 63.37% on the corn dataset, demonstrating its effectiveness in maintaining the mutual information between the predicted and true clusters. These results underscore DynAE's robustness and potential as a leading approach for clustering tasks in plant disease detection. Following the analysis of DynAE's overall performance, an important enhancement in this work was the replacement of the K-means clustering algorithm with the Gaussian Mixture Model (GMM) in the DynAE framework. The original DynAE method relied on K-means for clustering, but this thesis demonstrates that GMM significantly improves clustering accuracy and robustness. DynAE with GMM achieved an even higher accuracy of 78.99% on the corn dataset, along with a slightly improved NMI of 63.88% on the same dataset. This modification leads to superior clustering performance, with DynAE using GMM outperforming not only the original DynAE model with K-means, but also other deep clustering methods. | |
| metadata.dc.identifier.other | Drupal node 3718 | |
| metadata.dc.identifier.other | https://digitalrepository.uob.edu.bh/en/dar/deep-clustering-techniques-plant-disease-discovery | |
| metadata.dc.identifier.uri | http://165.227.56.122/handle/123456789/7286 | |
| Language | en | |
| Link | https://digitalrepository.uob.edu.bh/en/_flysystem/fedora/2025-07/00872_deep_clustering_techniquesfor_plant_disease_discovery.pdf | |
| Title | Deep Clustering Techniques for Plant Disease Discovery | |
| Type | Doctoral Dissertation |
Files
Original bundle
1 - 1 of 1
Loading...
- Name:
- node-3718-1-00872_deep_clustering_techniquesfor_plant_disease_discovery.pdf
- Size:
- 4.1 MB
- Format:
- Adobe Portable Document Format