synthetic data for food sorting.

Sorting of food is inherently complex because of the great variation in food and specifically in organic produce. Machine learning is a powerful tool and very fitting to this task. Reliable and highly consistent labeling of data is key to successfully developing robust algorithms. In this project, we were tasked with developing datasets to aid in the sorting of green beans. Specifically to help de-tangle beans: linking corresponding sections of bean, help grade the quality of the beans and a number of additional rich labels.

highlights

  • Enabling the development of novel algorithms through synthetic data
  • A high level of control over the simulated sensor data and the annotation
  • This pipeline opened new possibilities for future developments

the solution.

To generate a close synthetic dataset, the imaging setup – including lighting, environment and cameras – was reproduced in simulation. Inside this environment, a bean module was developed and employed. See our synthetic data section for more details on the architecture of these modules. Below you find a video to illustrate this bean module.

 

the results.

Our synthetic data pipeline enabled the generation of a varied dataset with highly consistent labeling. In the video Below, you find a selection of renders and annotations generated by the system.

synthetic data has the potential to unlock the use of machine learning where a lack of the right data is currently limiting.