Back

Automated harvesting of fruits

Project Description:

For automated harvesting of fruits several tasks need to be performed. One is the classification of fruits and the semantic segmentation of the point cloud data for each object instance. For this task different AI models showed good performance. However the vision data is streamed in realtime and the AI image segmentation must also be performed in realtime to enable the automated lag free harvesting of the fruits.

Vision transformers have recently become established for this type of task. This type of segmentation algorithm will therefore be considered in particular.

Algorithms:

  • - MASK-RCNN
  • - DeepLab
  • - UNet
  • - Vision transformer (ViT)

Milestones:

  • - Literature Review for datasets and open source AI models
  • - Selection of model and dataset
  • - Training of model and benchmark with different hyperparameter
  • - Programming of evaluation software
  • - Programming of evaluation software

Challenges:

  • - -

Materials:

Datasets:

  1. MinneApple - Dataset
  2. MinneApple - Paper

Sources:

  1. [1] Vision transformer survey
  2. [2] Overview of segmentation algorithms
  3. [3] Semantic segmentation models and papers
  4. [4] Vision transformer: