Predicting Vegetable Crop Height and Size Using Advanced LiDAR Technology

Jenn Hoskins
29th June, 2024

Predicting Vegetable Crop Height and Size Using Advanced LiDAR Technology

This experimental setup, featuring vegetable crop plots and a Terrestrial Laser Scanner (TLS), was used to acquire the critical LiDAR point cloud data necessary for training the deep learning model to predict plant height and crown area.

Image adapted from: J et al. / CC BY (Source)

Key Findings

  • The study by the Indian Institute of Space Science and Technology used LiDAR point clouds and a hybrid deep learning model to predict crop height and crown area for tomato, eggplant, and cabbage
  • The hybrid model, combining LSTM and GRU, achieved around 80% accuracy in predicting plant height and crown area, with error rates for height prediction ranging from 5 to 12%
  • The model was particularly effective during early and mid-growth stages but showed lower prediction quality at advanced growth stages, indicating a need for further refinement
Remote sensing has become a vital tool in precision agriculture, thanks to advancements in sensor miniaturization and platform technologies. By providing high-resolution data, remote sensing helps address within-farm variations, enabling better crop management. A recent study conducted by the Indian Institute of Space Science and Technology focuses on predicting plant-level crop height and crown area for vegetable crops using LiDAR point clouds and a hybrid deep learning framework[1]. The study aimed to develop a deep learning model to predict structural parameters of crops at various growth stages. LiDAR point clouds were collected using a terrestrial laser scanner on five dates during the growth cycles of tomato, eggplant, and cabbage. These data were then used to train a hybrid deep learning framework combining Long-Term Short Memory (LSTM) and Gated Recurrent Unit (GRU) models. The predictions were validated against ground truth measurements, showing that the model could predict plant height and crown area with around 80% accuracy. The hybrid model significantly improved predictions, particularly for crown area, compared to standalone LSTM and GRU models. The error rates for height prediction ranged from 5 to 12%, with a balanced distribution between overestimation and underestimation. This indicates that the hybrid approach effectively captures the temporal growth patterns of crops, making it a promising tool for precision agriculture. This study aligns with earlier research that utilized various remote sensing technologies for crop trait prediction. For instance, high-throughput field phenotyping using multispectral cameras has been shown to predict yield and classify wheat varieties based on texture features[2]. Similarly, hyperspectral imaging combined with deep learning has been used to monitor plant phenotypes under salinity stress, demonstrating the potential for non-destructive, rapid plant trait assessment[3]. The use of UAV-based multispectral imagery has also been explored for predicting cassava root yield, showing that integrating time-series image data with machine learning models can improve the accuracy of yield predictions[4]. These studies collectively highlight the growing importance of remote sensing and machine learning in modern agriculture. The current study expands on these previous findings by focusing on LiDAR point clouds and a hybrid deep learning framework for predicting structural parameters of vegetable crops. The study's results indicate a strong relationship between the features of the LiDAR point cloud and the auto-feature map generated by the deep learning models. This approach effectively captures the temporal growth patterns of crops, offering a robust method for early prediction of crop growth parameters. However, the study also acknowledges some limitations. The prediction quality was relatively low at advanced growth stages, closer to harvest. This suggests that while the hybrid model is effective during early and mid-growth stages, further refinement is needed for late-stage predictions. Despite these limitations, the stable prediction quality across different crops (tomato, eggplant, and cabbage) underscores the model's versatility. In conclusion, the study by the Indian Institute of Space Science and Technology demonstrates the potential of combining LiDAR point clouds with a hybrid deep learning framework for predicting plant-level structural parameters in vegetable crops. This approach offers a promising avenue for enhancing precision agriculture, building on the foundation laid by earlier research in remote sensing and machine learning[2][3][4].

VegetablesAgriculturePlant Science

References

Main Study

1) Deep learning-based prediction of plant height and crown area of vegetable crops using LiDAR point cloud.

Published 28th June, 2024

https://doi.org/10.1038/s41598-024-65322-8

Related Studies

2) Multi temporal multispectral UAV remote sensing allows for yield assessment across European wheat varieties already before flowering.

https://doi.org/10.3389/fpls.2023.1214931

3) Hyperspectral imaging combined with machine learning as a tool to obtain high-throughput plant salt-stress phenotyping.

https://doi.org/10.1111/tpj.14597

4) Machine learning for high-throughput field phenotyping and image processing provides insight into the association of above and below-ground traits in cassava (Manihot esculenta Crantz).

https://doi.org/10.1186/s13007-020-00625-1


Related Articles

An unhandled error has occurred. Reload 🗙