Case studies

The objective was to be able to detect all types of imperfections affecting the potatoes’ quality (more than 20) and to provide a solution that would give Patates Dolbec the flexibility of choosing the quality level of their product according to the varying needs of their customers (restaurants, grocery stores, etc.).

An additional challenge was to surpass the performance of the detection algorithm of their industrial sorting machine, the CELOX,  which is the market benchmark for this sort of task. This would allow Patates Dolbec to reallocate its quality assurance team to higher value-added activities.

Patates Dolbec was not satisfied with the performance of their older optical sorting machine, which had an error rate of 30%. This forced employees to manually sort part of the potatoes, resulting in the loss of a significant quantity of healthy potatoes and the associated revenue. In an industry with chronic labor shortage, a better solution was crucial. With the ever-changing nature of grown products, they needed an evolutive solution to be able to adapt to the changes in the base product.

We therefore retrofitted their optical sorting machine with high-definition cameras and a state-of-the-art deep neural network computer vision model, leveraging recent developments in AI and computer vision. To allow for periodic retraining of the model, a machine learning pipeline was developed in the AWS cloud. From data storage, labeling, and training to model registry and edge deployment, the AWS cloud serves as the foundational backbone of the upgraded sorting machine.

In the end, the implementation of AI in the quality control process had allowed Patates Dolbec to gain no less than 25% in efficiency as the error rate went from 30% to 5% in a matter of a few months.