Weizenährenerkennung mithilfe neuronaler Netze und synthetisch generierter Trainingsdaten

This paper investigates the usability of synthesized training data for the recognition of wheat ears using neural networks in the context of semantic image segmentation. For this purpose, detailed scenes of wheat fields consisting of 3D models with high-resolution textures and defined material properties are modeled. Afterwards, photo realistic color images are synthesized, which also contain a binary image mask with the locations of the ear models. The resulting image pairs are then used as a training data for two neural networks (U-Net and DeepLab-V3+). To determine whether these data allows domain adaptation, the trained networks are evaluate dusing real wheat field images. The IoU value of about 69.96 shows that information transfer from the synthesized images to real images is possible.