RAINaDiv - Robust Artificial Intelligence to handle Natural Diversity
Developing a deep learning computer vision algorithm capable of sorting food (various types of nuts and vegetables) in real-time with unprecedented accuracy to identify complex cases of defective or seasonally varying products.
The algorithm uses multispectral images recorded directly from the sorting machine. Multispectral images contain more information than standard RGB optical channels. Recordings are streamed into a deep learning model which performs the multi-class object segmentation task to detect products or defects of different types.
Testing the algorithm on a larger variety of products and integrating it in production into the machine software.