Research on integrated LiDAR and deep learning-based biomass estimation method for monitoring tomato stem and leaf biomass by using drones

Accurate and robust estimation of crop biomass is fundamental for precision agriculture, yield assessment, and intelligent crop management, and remains challenging under complex canopy structures and organ-specific growth dynamics. This study proposes a knowledge-guided biomass estimation framework for tomato plants by integrating physically based radiative transfer modeling, deep learning and LiDAR-derived structural information. For leaf biomass estimation, a Knowledge-Guided Convolutional Neural Network(KGCNN) was developed by coupling the LESS three-dimensional radiative transfer model with multi-layer convolutional feature learning, which enables the deep fusion of spectral and structural information while alleviating overfitting and improving regression stability. To address the limited capability of canopy spectral signals to capture stem biomass variations, terrestrial LiDAR data were utilized to extract the stem height and diameter. A hierarchical regression strategy was designed to model stem biomass by accounting for pronounced allometric growth characteristics along different internode segments. By jointly modeling leaf and stem biomass within a physically informed framework, the LiDAR-BESLM model achieves high accuracy, strong robustness and enhanced interpretability across the growth stages under real field conditions. The results demonstrate that this method provides an effective and scalable solution for fine-grained biomass estimation in tomato production systems, with potential applicability to other crops and cultivation environments.