Use of satellite imagery and UAV to assess weed infesta-tion of soybean crops
Abstract
Estimating crop heterogeneities, including those caused by weed infestation, is an important task in precision agriculture. Sentinel-2 data from 2022 to 2024 and monthly DJI Mavic 3M imagery from 2024 were used to assess weed infestation in soybean fields in Khabarovsk Krai. Weekly NDVI time series were generated using Fourier series fitting. To identify weed-infested soybean fields, the characteristics of the NDVI seasonal curve were evaluated—specifically, the width of the peak at half (d1/2) and at three-quarters of the height (d3/4).
It was found that in Khabarovsk Krai, the average d1/2 ranged from 119.3 to 128.4 for weed-infested fields and from 81.1 to 92.2 for lightly weed-infested fields. Meanwhile, d3/4 ranged from 72.6 to 84.2 for weed-infested fields and from 50.1 to 57.8 for lightly infested fields. These differences were found to be statistically significant (p < 0.05).
Experimental results from weed-infested soybean plots in Primorsky Krai indicated that NDVI curve peaks were consistent with these specified ranges. Using monthly DJI Mavic 3M imagery of an experimental soybean site in 2024, located in Khabarovsk Krai, it was observed that heterogeneities associated with weed infestation could be identified in late June to early July, prior to crop row-joining. Fitted NDVI time series for the experimental field revealed differences in d1/2 and d3/4 between the control and herbicide-treated plots. Further development of this method will focus on the early prediction of NDVI curve parameters and their comparison with lightly weed-infested fields from previous years.
