• Login
    View Item 
    •   NM-AIST Home
    • Computational and Communication Science Engineering
    • Masters Theses and Dissertations [CoCSE]
    • View Item
    •   NM-AIST Home
    • Computational and Communication Science Engineering
    • Masters Theses and Dissertations [CoCSE]
    • View Item
    JavaScript is disabled for your browser. Some features of this site may not work without it.

    A deep learning model for early detection of maize diseases in Tanzania

    Thumbnail
    View/Open
    Full text (2.435Mb)
    Date
    2024-08
    Author
    Mayo, Flavia
    Metadata
    Show full item record
    Abstract
    Agriculture is considered the backbone of Tanzania’s economy, with more than 60% of residents depending on it for their livelihood. Maize is the main and dominating food crop in the country, accounting for 45% of all farmland produce. Inevitably, its production is affected by diseases that, if detected early, could be easily treated. Maize Streak Virus (MSV) and Maize Lethal Necrosis (MLN) are commonly reported diseases that are often detected too late by farmers. This raised the need for a sophisticated method for early detection of these diseases to enable timely treatment. This study investigated the potential of developing a deep-learning model for the early detection of maize diseases in Tanzania. Imagery datasets were used for model training, they were collected through physical observation with the help of a plant pathologist from three regions of the country; Arusha, Kilimanjaro, and Manyara. Two models, a Convolutional Neural Network (CNN) and a Vision Transformer (ViT), were developed from scratch and classified into four classes: Healthy, MLN, MSV, and WRONG. The results revealed that the ViT model outperformed the CNN model, achieving validation accuracies of 0.931 and 0.9096 respectively. Despite the superior performance of the ViT model, the CNN model was selected for deployment in a mobile-based application due to its smaller size, lower memory, and small computational requirements. A quantitative research method using a survey questionnaire was employed to gather requirements for system development and validation. Validation of the model's performance through questionnaires administered to farmers and agricultural experts, yielding highly positive feedback. This enthusiastic reception highlights the potential impact of the developed application on improving early disease detection and enhancing maize productivity in Tanzania. Empowering stakeholders of agriculture to identify more effective methods of managing them before serious harm is done. Consequently, custodians of agriculture, such as the Ministry of Agriculture, organizations, and other companies, will have the potential to detect maize diseases early, henceforth adding up to the increase in the Gross Domestic Product (GDP) and guaranteeing the country’s food security.
    URI
    http://doi.org/10.58694/20.500.12479/2917
    Collections
    • Masters Theses and Dissertations [CoCSE]

    Nelson Mandela-AIST copyright © 2021  DuraSpace
    Theme by 
    Atmire NV
     

     

    Browse

    All PublicationsCommunities & CollectionsBy Issue DateAuthorsTitlesSubjectsThis CollectionBy Issue DateAuthorsTitlesSubjects

    My Account

    LoginRegister

    Nelson Mandela-AIST copyright © 2021  DuraSpace
    Theme by 
    Atmire NV