Esomonu, Nkechi Faustina2026-10-192026-10-192023-09Esomonu, N. F.(2023). An inelligent hybrid model for tomato leaf didease detection and control recommendations [Unpublishe Doctoral Thesis]. Federal University of Technology, Owerri, Nigeriahttps://repository.futo.edu.ng/handle/20.500.14562/2855This thesis is for the award of Doctor of Philosophy (PhD.) in Information Management TechnologyCrop diseases pose a significant threat to global food security, with the negative impact on food production stemming from a lack of early detection and effective management. Despite technological advancements, especially in artificial intelligence (AI), smallholder farmers face challenges in accessing and utilizing these tools. Tomato, a crucial vegetable crop in the global food chain, is particularly susceptible to various debilitating diseases, leading to annual losses for farmers. This research aims to address the impact of disease related losses on tomato farmers, specifically in Nigeria, by developing an AI-powered system for early detection, precise diagnosis, and continuous monitoring of tomato diseases. The motivation for this research stems from the recognition of the limited access smallholder farmers have to efficient technology tools for managing tomato diseases. The goal is to minimize the impact of these diseases on farmers' livelihoods and, by extension, society. Through early detection and effective management, the research seeks to empower farmers with the necessary tools to mitigate losses and enhance food production. The primary objective of this research is to develop a hybrid AI-powered system that aids smallholder farmers in early detection, effective management, and continuous monitoring of tomato plantation health. The approach involves the integration of deep and shallow learning models, with a focus on transfer learning. The functional unit comprises a convolutional neural network model (MobileNetv2), a support vector machine (SVM) model, and a natural language processing component for image processing, feature extraction, classification, and management counseling. To achieve the research objective, a hybrid architecture combining deep and shallow learning models was developed. The model utilized 11,000 tomato leaf images for training, ensuring a balanced representation of healthy and diseased types. The problem was formulated as a ten-class (multiclass) classification task. Evaluation of the proposed solution employed standard metrics, and the results demonstrated state-of-the-art performance. The hybridized MobileNetv2 and SVM achieved impressive metrics, including 93.68% precision, 93.84% F-1 score, 93.81% accuracy, and 92% recall. The evaluation of the proposed solution showcased its efficacy in achieving high precision and accuracy. The hybridized model, incorporating MobileNetv2 and SVM, demonstrated state-of-the-art performance in recognizing and diagnosing tomato diseases. The results indicate the potential of the developed AI-powered system for practical implementation in agriculture. In conclusion, this research contributes significantly to addressing the challenges faced by smallholder tomato farmers in Nigeria. The developed AI-powered system, combining deep and shallow learning models, demonstrates remarkable performance in early detection, precise diagnosis, and continuous monitoring of tomato diseases. The successful implementation of the model in an Android application suggests its potential for acceptability and adoption among the farming population. However, to enhance usability, further development of the natural language component, including speech processing capabilities, is recommended, particularly considering the literacy constraints of rural farmers. This research aligns with Sustainable Development Goals 1, 2, and 3, and it is recommended for further development and support by relevant government agencies to fulfill the promise of technology utilization in improving societal well-being through increased food availability.enAttribution-NonCommercial-ShareAlike 4.0 InternationalAICNNclassificationimagesdeep learningUSEDepartment of Information TechnologyDoctoral Thesis