Computer vision-based non-invasive biomass estimation and water stress monitoring in tomato plants

  Full text PDF

Precision horticulture demands intelligent monitoring systems for automated crop management. While traditional biomass estimation relies on destructive sampling, modern ICT-driven approaches offer transformative solutions. This study presents an AI-powered methodology integrating YOLOv11 for autonomous biomass estimation and stress monitoring in tomato crops. High-resolution RGB imagery was captured across multiple phenological stages under two irrigation regimes. YOLOv11’s computer vision capabilities enabled automated canopy detection, segmentation, and digital biomass quantification, eliminating destructive sampling. Novel AI-driven metrics were introduced: convex hull area for stress-induced canopy alterations and compactness (digital biomass to convex hull ratio) for automated canopy assessment. This integrated approach achieved robust accuracy in stress detection and biomass estimation (R² =0.821), enabling real-time monitoring for precision horticulture. The model demonstrated exceptional performance with segmentation precision of 0.950, recall of 0.979, and mean average precision of 0.975 at IoU 0.5, with mAP50-95 of 0.826. The rapid inference time of 2.3ms per image enables high-throughput phenotyping and decision support for site-specific management. YOLOv11-derived digital biomass correlated strongly with fresh biomass (R² = 0.821). Image-derived features effectively differentiated control and stress conditions. Genotype analysis revealed variation in biomass accumulation: Arka Abhed and Arka Rakshak performed better under optimal irrigation, while Arka Vikas showed greater stress resilience. These results validate YOLOv11 as a scalable solution for intelligent crop monitoring and precision input application in next-generation digital horticulture.