A field experiment was conducted in Bengaluru, India in 2017 to study the growth conditions of three different vegetable crops – tomato, eggplant, and cabbage. The experiment aimed to develop techniques using space technology for detecting crop types and ecosystem services. The experiment was conducted during the Kharif crop growing season, and different levels of nitrogen fertilizer were applied to the crops. LiDAR point cloud data and reference crop structural parameters were acquired for analysis.
The point cloud data was processed to generate a crop height model and estimate crown areas using advanced algorithms. Deep learning techniques, specifically a hybrid hierarchical model of stacked LSTM and GRU layers, were used to predict plant height and crown area based on the acquired data. The model, named ‘TemporalCropNet’, showed significant improvement in predicting crop structural parameters.
The training and validation of the deep learning model were done using a cross-validation strategy to ensure robustness. The performance of the model was evaluated using metrics like symmetric mean absolute percentage error and logarithmic deviation. The study did not involve human or animal subjects, and ethical approval was not required.
Overall, the experiment and analysis aimed to enhance understanding of crop growth conditions and develop predictive models using advanced technologies and deep learning algorithms.
Source link
Source link: https://www.nature.com/articles/s41598-024-65322-8