Monitoring Leaf Area Index of the Sown Mixture Pasture Through UAV Multispectral Image and Texture Characteristics
Xiaoxue Wang,Shicheng Yan,Wenting Wang,Liubing Yin,Meng Li,Zhe Yu,Shenghua Chang,Fujiang Hou
DOI: https://doi.org/10.1016/j.compag.2023.108333
IF: 8.3
2023-01-01
Computers and Electronics in Agriculture
Abstract:Leaf area index (LAI) is an important phenotypic trait closely related to photosynthesis, respiration, and water utilization. In recent years, unmanned aerial vehicles (UAVs) multispectral capabilities enable the acquisition of spectral information from visible light to near infrared, facilitating vegetation growth monitoring. This study aims to explore the methodology of combining vegetation indices, color indices, texture information, and ecological factors based on UAV multispectral images to enhance the accuracy of the sown mixture pasture LAI estimation. A field experiment involving 13 mixed sowing combinations of alfalfa (Medicago sativa L.), tall fescue (Festuca elata Keng ex E. Alexeev) and plantain (Plantaga lanceolata L.) was conducted out. Multiple linear regression, Bagging algorithm, support vector machine (SVM), random forest algorithm (RF), KNN algorithm, and back propagation neural network (BP) were used to construct the LAI prediction model. The results showed that combining vegetation index (VI) + color index (CI) + normalized difference texture index (NDTI), and ecological factors (EF) could enhance the accuracy of LAI estimation. The sensitive characteristic combinations for alfalfa, tall fescue, and plantain were found to be NDRE (Normalized difference red-edge index) + NGBDI (Normalized green-blue difference index) + (B) MEA-(G) HOM (Blue band Mean - Green band Homogeneity) + DTR (Daily temperature difference), MSR (Modified simple ratio) + NGRDI (Normalized green-red difference index) + (G) COR-(R) VAR (Green band Correlation - Red band Variance) + DTR (Daily temperature difference)), and MSR (Modified simple ratio) + ExG (Excess green) + (G) SEM-(G) MEA (Green band Second-order moment - Green band Mean) + DTR (Daily temperature difference), respectively. RF exhibited superior prediction capability, further enhancing the accuracy of forage LAI prediction. The alfalfa, tall fescue, and plantain obtained coefficient of determination (R2) of 0.83, 0.79 and 0.79, root mean squared error (RMSE) of 0.50, 0.58 and 0.70, and mean absolute error (MAE) of 0.36, 0.45 and 0.55, respectively. These findings provide valuable insights for the estimation of leaf area index of the sown mixture pasture through UAV multispectral images and texture characteristics.