Non-Destructive Monitoring of Rice Growth Key Indicators Based on Fixed-Wing UAV Multispectral Images

In recent years, with the rapid development of remote sensing technology, real-time and non-destructive monitoring of crop growth status has become a research hotspot. Remote sensing-derived agricultural information will provide guidance for the precise management of large-scale crops. Among various remote sensing monitoring platforms, unmanned aerial vehicles (UAVs) have attracted wide attention due to their simple operation and low cost. UAVs equipped with multispectral cameras can quickly obtain crop growth conditions.

This study attempted to combine texture information and spectral information of multispectral images of fixed-wing UAVs to explore the monitoring effect of “atlas” information on rice growth indicators.

A two-year rice field experiment involving different sowing dates, varieties, planting methods and nitrogen levels was conducted. During the key growth stages of rice, remote sensing images of the rice canopy were obtained using a Sequoia multispectral camera mounted on a fixed-wing UAV. Shoot destructive sampling was conducted simultaneously to obtain leaf area index (LAI), aboveground biomass (AGB), plant nitrogen content (PNC) and other agronomic indexes of rice. Simple regression, partial least squares regression and artificial neural network algorithms were used to construct rice growth index monitoring model based on multispectral images of fixed-wing UAV. The monitoring effects of spectral texture information in different models were compared and analyzed.

The quantitative relationship between vegetation index (VI), single-band texture features and rice LAI, AGB, and PNC was explored using simple linear regression. The results showed that vegetation indexes had strong correlations with LAI and AGB, with the best-performing indexes being CIRE and NDRE, with R² values of 0.80 and 0.76, respectively. However, for PNC monitoring, vegetation indexes did not achieve ideal results, with the best-performing RESAVI and NDRE having R² values of only 0.13 with PNC. Further analysis using simple linear regression revealed that single-band texture features did not perform well in monitoring rice growth indicators. In order to further analyze the monitoring effect of image texture on the above three indexes, normalized texture indexes (NDTI), ratio texture indexes (RTI), and difference texture indexes (DTI) were constructed by referring to the construction method of VI. Correlation analysis showed that the newly constructed texture index (TI) improved the monitoring accuracy of rice growth indicators compared to single-band texture feature but did not perform better than vegetation indexes. To combine spectral and texture information, partial least squares and artificial neural network modeling methods were adopted in this paper. VI and VI+TI were used as different input parameter combinations to construct rice LAI, AGB and PNC monitoring models. The results showed that both partial least squares and artificial neural network modeling methods significantly improved the monitoring accuracy compared to simple linear regression. The best performance was achieved using VI+TI as input variables and an artificial neural network model for validation, with validation R² values for LAI, AGB, and PNC models increasing from 0.75, 0.72, and 0.26 to 0.86, 0.92, and 0.86, respectively, while RMSE values were significantly reduced.

The monitoring accuracy of rice LAI, AGB and PNC can be effectively improved by using the fixed-wing UAV to collect multispectral images of rice canopy and using the texture features and reflectance information as input parameters of the model through the model construction method of artificial neural network. The research results will provide a theoretical basis for rapid monitoring of large area crop growth.