Improved estimation of SPAD values in walnut leaves by combining spectral, texture, and structural information from UAV-based multispectral image

An important measure to describe the physiological status of vegetation is the chlorophyll concentration of the vegetation. For managing fields in precision agriculture, monitoring walnut growth, and estimating production, accurate determination of chlorophyll content is crucial. Spectral indices play a crucial role in the non-destructive and efficient monitoring of crop physiological parameters, especially in estimating chlorophyll content. However, spectral indices have low sensitivity to high chlorophyll levels and are susceptible to interference from background signals, which may result in decreased stability of the model. The unmanned aerial vehicle (UAV) captures high-resolution images that contain abundant spatial information, including texture and structural information. These spatial information can reflect crop canopy structure, may help to improve the estimation precision of crop chlorophyll content. However, research on utilizing drone-based spatial information for estimating crop chlorophyll content is relatively limited. The aim of this study is to explore the potential of integrating spectral, textural, and structural information to improve the accuracy of walnut leaf chlorophyll content estimation. This study used a drone equipped with a multispectral camera to capture images of walnut tree canopies. Based on these images, we extracted 17 spectral indices, 8 texture indices, and 5 structural indices. Then, we applied the Boruta algorithm to select the optimal spectral, texture, and structural indices, as well as their combinations. Finally, the SPAD (Soil and plant analyzer development) values estimation model for walnut leaves was established using the Decision Tree Regression (DTR), Random Forest Regression (RFR), and Extreme Gradient Boosting (XGBoost) methods. The research findings indicate that the accuracy of SPAD values estimation model constructed by combining remote sensing indices (spectral indices (SI), texture indices (TI) and structural indices (STI) are combined in pairwise or in full) is better than that of single remote sensing indices. In the combined remote sensing indices models, the accuracy of the model constructed by the pairwise combination of three remote sensing indices is relatively limited. However, when using the combination of spectral indices, texture indices, and structural indices, the estimation accuracy of SPAD values for walnut leaves can be effectively improved, and this combination is considered the best way to estimate SPAD values. Furthermore, among the trio of SPAD values estimation models considered (namely DTR, RFR, and XGBoost), the XGBoost model exhibited superior performance. Notably, when combining SI+TI+STI in the construction of the XGBoost model, it demonstrated the highest level of accuracy in estimating SPAD values (Training: R2T =0.95, RMSET=1.08; Validation: R2v = 0.72, RMSEv=2.13). The findings of this research elucidate that the incorporation of spatial information from UAV multispectral imagery facilitates the monitoring of physiological parameters in walnut trees. By integrating the spatial and spectral information of UAV multispectral imagery, a feasible and accurate estimation method has been provided for monitoring the chlorophyll content in walnut leaves.