Innovative UAV LiDAR Generated Point-Cloud Processing Algorithm in Python for Unsupervised Detection and Analysis of Agricultural Field-Plots

Abstract

The estimation of plant growth is a challenging but key issue that may help us to understand crop vs. environment interactions. To perform precise and high-throughput analysis of plant growth in field conditions, remote sensing using LiDAR and unmanned aerial vehicles (UAV) has been developed, in addition to other approaches. Although there are software tools for the processing of LiDAR data in general, there are no specialized tools for the automatic extraction of experimental field blocks with crops that represent specific “points of interest”. Our tool aims to detect precisely individual field plots, small experimental plots (in our case 10 m2) which in agricultural research represent the treatment of a single plant or one genotype in a breeding trial. Cutting out points belonging to the specific field plots allows the user to measure automatically their growth characteristics, such as plant height or plot biomass. For this purpose, new method of edge detection was combined with Fourier transformation to find individual field plots. In our case study with winter wheat, two UAV flight levels (20 and 40 m above ground) and two canopy surface modelling methods (raw points and B-spline) were tested. At a flight level of 20 m, our algorithm reached a 0.78 to 0.79 correlation with LiDAR measurement with manual validation (RMSE = 0.19) for both methods. The algorithm, in the Python 3 programming language, is designed as open-source and is freely available publicly, including the latest updates.

Crop growth dynamics: Fast automatic analysis of LiDAR images in field-plot experiments by specialized software ALFA

“…Data for software testing and optimization were acquired using UAV LiDAR system (Ricopter VUX-SYS, Riegl GmBh, Austria) using same parameters as described previously [ 25 ]. Eight flights from March to June were performed to scan an array of barley field-plots organized into three experimental blocks (48 variants per each block, see the cyan experimental area in Fig 2B ).…”

“…Field plot experiments were performed on the spring barley variety Francin (Selgen, CZ) during the year 2021. Field-plots were scanned using UAV system Ricopter VUX-1UAV, Riegl GmBh, Austria) that was operated automatically using the UGCS software (SPH Engineering, Latvia) as described previously [ 25 ]. The flight altitude was set to 20 m AGL, horizontal speed of flight 4 m/s, scanning line distance 9 m, calculated side overlap (77%, 31 m).…”

“…To overcome low accuracy of analysis in cotton-canopy height, Xu et al [ 24 ] developed method implementing canopy laser interception compensation mechanism in their model. Our previous algorithm [ 25 ] was based on direct-point analysis strategy to segment trial field-plots of winter wheat. However, direct-point analysis is highly demanding on computing time.…”

Forestry Digital Twin With Machine Learning in Landsat 7 Data

“…We adopt some feasible approaches from a review of agricultural DT (Cor et al, 2021 ) because there is no forestry DT review article in our survey. Agricultural DT uses images and point-cloud data from agricultural machines and drones, necessitating the use of a powerful computer for data processing (Polák et al, 2021 ). Using remote sensing images to analyze forest land could be done on a personal computer.…”

Forestry digital twin with machine learning in Landsat 7 data

“…Because there is no forestry digital twin review article in our survey, we try to find some feasible approachs from a review of agricultural digital twin Cor et al (2021). Agricultural DT use images and point cloud data from agricultural machines and drones, therefore require powerful computer for data process Polák et al (2021). Forest land analysis using remote sensing images which could be done in personal computer.…”