UAV Multi-Source Data Fusion with Super-Resolution for Accurate Soybean Leaf Area Index Estimation

Zhenqing Zhao¹Huabo Yao¹Depeng Zeng²Zhenfeng Jiang²Xihai Zhang^1*

• ¹College of Electrical Engineering and Information, Northeast Agricultural University,Harbin 150030, China, Harbin, China
• ²College of Agriculture, Northeast Agricultural University, Harbin 150030, China, Harbin, China

Leaf Area Index (LAI) is a key parameter that reflects the structure and growth status of crop canopies. Accurate measurement and assessment of LAI provide valuable insights into crop health. With the advancement of unmanned aerial vehicle (UAV) technology, multispectral sensors mounted on UAVs enable high-throughput estimation of LAI. However, UAV flight altitude directly affects both operational efficiency and image resolution, which in turn influences the accuracy of LAI estimation. In this study, we acquired RGB and multispectral images using UAVs flying at four altitudes: 15 m, 30 m, 45 m, and 60 m. We then conducted super-resolution (SR) image reconstruction and modeling analysis. The main findings are as follows: (1) SR algorithms varied significantly in their ability to reconstruct canopy images, and their effectiveness decreased as flight altitude increased. Among the tested methods, SwinIR provided the best performance in terms of PSNR and SSIM, surpassing Real-ESRGAN, SRCNN, and EDSR. (2) Texture features extracted from RGB images demonstrated strong sensitivity to LAI. The XGBoost algorithm, which integrated both RGB and multispectral data, achieved the highest accuracy, with a relative error of 4.16%, significantly outperforming models based solely on RGB data (5.25%) or only multispectral data (9.17%). (3) The incorporation of SR techniques significantly enhanced model accuracy at altitudes of 30 m and 45 m. At 30 m, models incorporating Real-ESRGAN and SwinIR achieved an average coefficient of determination (R²) of 0.86. At 45 m, the same SR methods yielded models with an average R² of 0.77. Overall, the results underscore the potential of integrating SR techniques with multi-sensor data to improve the accuracy of soybean LAI estimation. This approach provides a reliable and efficient solution for UAV-based crop monitoring and supports data-driven decision-making in precision agriculture.