A recently published paper by Reji J and Rama Rao Nidamanuri (2024) in Scientific Reports explores the use of deep learning to predict plant height and crown area for vegetable crops based on LiDAR point cloud data. This method aligns with the growing trend of integrating remote sensing technologies into precision agriculture, aiming to provide detailed insights into within-farm variations.
The primary goal was to develop and apply a deep learning framework to predict key structural parameters—plant height and crown area—at various growth stages of vegetable crops. LiDAR data was collected using a terrestrial laser scanner across different dates during the growth cycles of tomato, eggplant, and cabbage at the University of Agricultural Sciences in Bengaluru, India.
The researchers employed a hybrid deep learning framework that combined features of Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks. These models were chosen for their capability to handle temporal sequences and make predictions based on the evolving patterns of the crops.
The study demonstrated that this hybrid deep learning approach could predict plant structural parameters effectively, with accuracy levels around 80%. Specifically, the model achieved error rates for height predictions between 5 to 12%, with a balanced distribution of overestimations and underestimations. The hybrid model notably outperformed individual LSTM and GRU models in predicting the crown area, suggesting a significant improvement in capturing the spatial variability and complexity of plant growth.
However, the predictive accuracy diminished at the advanced growth stages, closer to harvest time. Despite this limitation, the prediction quality remained relatively consistent across the different crops tested.
Find the complete research here.