Detection of Plant Diseases in an Industrial Greenhouse: Development, Validation & Exploitation

The effective detection of plant diseases is crucial for the optimal management of agricultural systems. In this paper, we present our contributions in the context of detecting plant diseases in an industrial greenhouse [1], focusing specifically on tomatoes. Our main objectives are to develop and validate a detection system using the YOLOv8 model and to explore its potential for practical application in a real-world setting. To facilitate our research, we introduce a novel dataset comprising images of tomato leaves affected by various diseases. This dataset serves as a valuable resource for training and evaluating our detection model. We employ the YOLOv8 architecture, a state-of-the-art object detection framework, and experiment with different parameters to assess its performance in accurately detecting diseased areas on tomato leaves. Through extensive experimentation, we compare the performance of the YOLOv8 model using various parameters, such as different training strategies, data augmentation techniques, and hyperparameter configurations. The results provide insights into the optimal settings for achieving high detection accuracy and robustness. Furthermore, we demonstrate the practical utility of our developed model by conducting a real-life implementation within an industrial green-house. This exemplifies the integration of our detection system into an operational environment, showcasing its potential to assist greenhouse operators in early disease detection, monitoring, and decision-making processes. Our preliminary findings demonstrate promising disease detection capabilities on tomato leaves inside greenhouses, achieving an mAP50 score of 0.8 using our best model. Although there is room for improvement, these initial results indicate significant potential.