FranceRetail and e-commerce2023 – ongoingGroupe Dubreuil
Photo-based part identification in the Agrizone mobile app
Groupe Dubreuil's Agrizone app found the right farm part from a photo about 90% of the time
Problem
Customers of Agrizone, the group's online shop for agricultural parts, had to find one part among more than 300,000 catalog references. Many parts carry engraved reference numbers that are hard to read, and the parts are often dirty or damaged.
Approach
Groupe Dubreuil's data scientists built a photo search for the Agrizone mobile app with Google Cloud's Vision AI. Text recognition reads the reference numbers, and an image model turns each catalog picture into a vector so the app can find visually similar products. With Google Cloud, the team added image pre-processing and denoising to improve reading of engraved text. The tool runs on Compute Engine, and the team had set a target of under 3 seconds per search.
Results
Takeaway
Image clean-up before text recognition made the largest gain in this project. The group is looking at the same approach for automotive parts in other subsidiaries.
Categories
- Industry
- Retail and e-commerce
- Business function
- Marketing and sales
- Innovation type
- Product and service innovation
- AI technology
- Computer vision
- Adoption
- Built in-house
- Evidence
- Vendor case study
Sources
Source summaries
Google Cloud's customer story describes how Groupe Dubreuil built a photo-based product search for its Agrizone farm-parts shop using Vision AI text recognition and image embeddings. It reports a product match about 90% of the time, an average response of 2.13 seconds and a rise in detection from 63% to 92% on a test sample after image pre-processing. The page carries no date; the Internet Archive first saved it on June 27, 2023.
Full details are in the original sources.
