Fig. 3. Principle and flow chart of plant multispectral reflectance correction. (IMAGE)
Caption
(A) The flow chart of generating plant multispectral point cloud. Raw images such as depth image and multispectral image were registered, and multispectral image was reshaped as a multichannel image at the beginning of the procedure. Then, follow the point cloud generation that relies on the transformation from depth image coordinate system to the world coordinate system under the constrains of the camera intrinsic parameters. Finally, with the fusion of multiview point clouds and the mapping of corrected multispectral textures, the 3D multispectral point cloud model was constructed. (B) The flow chart of calculating the spatial distribution of the DN values of the references and correcting the plant spectral reflectance using ANN. In the stage of model training, the 3D light field features of references were extracted from depth image as independent variables and the spectral DN values as dependent variables. In the stage of model application, the 3D light field features of plant were set as input to obtain the predictions of the corresponding DN values of the reference. Finally, the reflectance image is corrected pixel by pixel based on this method to generate a mappable texture.
Credit
Plant Phenomics
Usage Restrictions
Credit must be given to the creator.
License
CC BY