Embedded software solution for visual recognition

Context

Software solution deployed on a smartphone in offline mode to help technicians recognize and manage the inventory of over 575,000 electronic boards.

The Challenge

Within the French Ministry of the Armed Forces, SIMMT manages the maintenance of 4.3 million complete systems (essentially vehicles). This represents an annual volume of 575,000 electronic boards that technicians must identify with very few references indicated on the board.

The proposed solution must adapt on the one hand to the military context in operation and on the other hand to a data catalog for which there is only one photo available for certain parts.

The Solution

We have built an electronic board automatic recognition application that runs on a smartphone in real time and even without a connection.

We used Few-Shot Learning algorithms to allow us to achieve excellent performance despite a training corpus that can be restricted to a single photo.

The Results