Intelligent Early Warning System for Basil Disease Detection in Vertical Farming Using Deep Learning and Morphological Analysis

Abstract

This paper presents the results of the development and validation of a prototype intelligent early warning system for plant diseases in vertical farming, using basil (Ocimum basilicum L.) as a model organism. The system integrates computer vision (deep learning) methods with morphological analysis. Experiments were conducted under controlled conditions in a vertical farm with LED lighting using the cultivars "Queen Sheba", "Siamese Queen", and "Kapriz". A YOLOv8x model was trained for trichome detection on leaves (mAP@0.5 = 0.833), revealing a dependence of trichome count on the spectral composition of light (maximum under increased blue-light fraction) and statistically significant differences between adaxial and abaxial leaf surfaces (Mann–Whitney test, p < 0.05). A YOLOv11 model was trained for disease identification (fusarium wilt, bacterial spot) on a dataset of 214 annotated images, achieving a precision of 74.7% at a recall of 69.3%. A morphological analysis module based on the PlantCV library was developed, enabling automatic determination of leaf geometric dimensions with an error of 5–20% depending on imaging conditions; a calibration method using a reference object (pot) was proposed, ensuring independence from camera technical specifications. The technical design of an early warning system integrating visual symptom detection and morphometric monitoring was formulated. The results demonstrate the promising application of neural network technologies for precision monitoring of phytopathologies in vertical greenhouses.