Peanut Leaf Area Index Estimation Based on Fusion of Texture and Spectral Information from UAV Imagery

Unmanned aerial vehicles (UAV) has been increasingly popular in the fields of crop growth monitoring, with their practical advantages, such as their ability of low-cost, rapid and repetitive capture. LAI is a key indicator for evaluating crop population growth and canopy structures and its accurate measurements would be of significance in the modern precision agriculture research. Therefore, we develop a peanut leaf area index estimation method based on fusion of texture and spectral information from UAV multispectral imagery. In our work, we compared the performance of LAI estimation using spectral and textural characteristics, and explore the potential of integrating them for estimating peanut LAI. Following this, LAI estimation models are constructed using spectral, textural characteristics, and a combination of spectral and textural characteristics based on some frequently-used statistical models. The results indicate that compared with only spectral or textural characteristics, fusion of both achieves better accuracy, which suggests that the texture features offer the critical information for improving the peanut LAI estimation accuracy. In addition, in order to select the best model, we compared four machine learning models and found that Support Vector Regression (SVR) is an optimal approach for the peanut LAI estimation, with higher R
²
(=0.817) and lower RMSE (=0.639) values for different feature sets. What’s more, the key parameters for computing texture features would affect the estimation performance, where the fluctuation is slight from the grayscale. Our observations would have important inspirations for the statistical model-based science discovery in the field phenotyping inversion.