Abstract:
Pests and diseases remain a major constraint to agricultural productivity in Zimbabwe, causing significant
losses and limiting farmers’ income. In greenhouse farming, early detection of infestations is critical to ensure
healthy crops and reduce pesticide overuse. This study presents a computer-vision system powered by artificial
intelligence (AI) for early identification of common pests and diseases in greenhouse crops. A convolutional neural
network (CNN) trained on a regionally curated dataset of healthy and diseased plant images automatically classifies
visual symptoms from digital photographs. The system is implemented through a MATLAB desktop application,
enabling offline classification for users with limited internet access. The CNN achieved 83 % training accuracy
and 82 % validation accuracy, with high precision and recall across multiple crop categories. Testing
confirmed reliable detection of leaf curl, septoria leaf spot, and related infections in tomatoes and peppers. This
work demonstrates that locally trained deep-learning models can effectively support greenhouse farmers in Zimbabwe,
enhancing early response and minimizing losses due to pest and disease outbreaks.
