The overall purpose of this project is to incorporate and adapt novel AI methods to improve the analysis of images of grains, and to increase the understanding of when, why and how a trained classification model can be reliably used. This will be approached via two specific aims: 1) Establish a suitable deep learning architecture and training strategy and perform root-cause-analysis of reproducibility and generalization (transferability to new data) issues. 2) Explore and adapt assisted training approaches (user/expert-in-the-loop) to increase reliability and allow for sustainable and continuous improvement of grain image analyses.