AI Model for Checking Rice Quality
Challenges
Our customer is a leading Japanese corporation that runs a chain of restaurants in eight countries with over 4000 outlets. Their goal is to become the world’s top company in the food and beverage industry. To achieve this, they need to evaluate the quality of the rice they purchase from different suppliers at various stages of the procurement process.
HBLAB's Solutions
We used the Cascade Mask R-CNN model with general data to detect and classify rice grains into four categories: OK, error (having abnormal color), broken, and cracked.
After one month of development, we achieved the following accuracy results for each category:
Project details
- Used Technologies Instant Segmentation (Cascade mask RCNN), Synthetic data generation
- Development Team 1 AI Engineer
- Duration 1 month
Results
Rice classifying AP is 97% (IOU 0,75).
Rice classification into 4 types result:
■ OK: 96,6% (AP_75)
■ Error: 86,9% (AP_75)
■ Broken: 71,1% (AP_75)
■ Cracked: 61,2% (AP_75)
Related cases
