Optimizing Disease Detection Models in School Greenhouses: An AI and IoT-Based Approach for Smart Agriculture

In recent years, the integration of Internet of Things technologies and machine learning models in agriculture has significantly advanced smart farming practices. This paper presents our research on enhancing disease detection in tomato plants within a greenhouse environment using advanced object detection models. Collaborating with the University of Montreal greenhouse, we developed a realistic dataset comprising images from both the PlantVillage repository and over 1,900 manually labeled leaf images taken from the greenhouse. Using this dataset, we evaluated and compared three object detection models: Faster R-CNN, YOLOv10, and SSD, to accurately detect and classify tomato leaf diseases. Our approach enables us to train a model on a more realistic set of images, facilitating automatic and earlier disease detection for farmers. Our results show that the Faster R-CNN model with a ResNet-50 backbone and Feature Pyramid Network is the most effective for detecting diseased leaves, achieving a mAP50 score of 93.13%.