Leaf area index and plant height can be used as important indicators to measure crop growth and yield. Accurately, quickly, and non-destructive acquisition of soybean leaf area index LAI and plant height H is of great significance for soybean production management. This article utilizes a multi rotor drone equipped with multispectral sensors to obtain multispectral images of soybean flowering, pod setting, and bulging stages, and simultaneously collects ground data. Based on DSM extraction of soybean plant height H in the study area, the results showed that the measured plant height H was fitted with the height extracted based on DSM (R2=0.8246; RMSE=0.047). Then, Pearson correlation analysis was performed on the extracted multispectral vegetation index, plant height, and texture features. A soybean LAI estimation model was constructed using univariate linear regression, support vector machine regression (SVR), random forest regression (RF), and BP neural network regression models, respectively. The results show that after adding texture features, the accuracy of all three algorithm models can be improved, with R2 increased by 0.222, 0.202, and 0.178, respectively; In addition, the accuracy of the BP neural network model in inversion modeling at various growth stages is superior to the SVR model and RF model. The BP neural network model has the best accuracy at the soybean bulging stage, with R2 of 0.856 and RMSE of 0.143. Texture features can effectively improve the saturation problem of single use vegetation index estimation under high-density canopy during soybean podding, and can extract more information for estimating soybean LAI, thereby improving the accuracy of the estimation model and providing guidance for soybean field management.
|