High-resolution reconstruction of images for estimation of plant height in wheat using RGB-D camera and machine learning approaches

In this study, a pipeline has been proposed where colour image and depth information of wheat plants are captured using an red green blue-depth (RGB-D) camera; later these two are combined to create a three-dimensional point cloud of the plant. The point clouds were processed to calculate the plant height. The results were then statistically analysed with the help of machine learning algorithms, viz. linear regression, support vector machine and artificial neural network (ANN). The comparison of the results shows that ANN performed better than the other two models with mean squared error 189.94, root mean squared error 13.70, mean absolute error 11.40 and mean absolute percentage error 18.73. The proposed study shows a high-precision and low-cost technology that can be widely used for non-destructive measurement of phenotyping parameters for wheat and other crops.