Enhancing Model Accuracy of UAV-Based Biomass Estimation by Evaluating Effects of Image Resolution and Texture Feature Extraction Strategy

The unmanned aerial vehicle (UAV) remote sensing technology provides new opportunities to estimate crop aboveground biomass (AGB). However, the application of UAV remote sensing data in practical farms remains challenging because to obtain UAV image features, including spectral index and texture features (TFs), and to improve the accuracy of AGB estimation have not been determined. In response to this problem, spectral indices (SIs) were derived from UAV RGB images with four spatial resolutions, and eight gray-level co-occurrence matrix TFs were calculated using different calculation parameters (TF_CP) including six window sizes and four directions. Maize AGB estimation models were established based on SIs only and combination of SIs and TFs using machine learning algorithms. We explored the impacts of spatial resolution and TF_CP on the performance of AGB models and analyzed the potentials of combination of SIs and TFs for improving maize AGB estimation accuracy. Results reveal that the performance of maize AGB estimation can be enhanced by supplementing SIs with TFs. The optimal image resolution and TF_CP are 20 mm and window size of 11 × 11 and direction of 90°. Combining TFs calculated with multi-TF_CPs further enhances estimation performance with an adjusted coefficient of determination of 0.75 and 0.81, and root mean square error of 963.50 and 302.19 g/m 2 for fresh and dry AGB, respectively. These findings indicated the significance of choosing the optimal spatial resolution and TF extraction strategy in enhancing UAV-based maize AGB estimation performance, particularly in the practical farming scenario.