Case Study: Cerulean achieves 97% defect detection accuracy with ScienceSoft's AI computer vision solution

Cerulean boosts defect detection to 97% with ScienceSoft

ScienceSoft was engaged by Cerulean, a UK-based company specializing in precision testing equipment for the tobacco industry, to address a challenge with an unreliable AI model for automated visual inspection. The previous model failed to accurately locate components and calculate key parameters on nicotine pouches, especially under real-world production conditions with poor seam visibility or rotated items, which risked undetected defects and regulatory issues.

ScienceSoft designed and implemented a production-ready computer vision solution utilizing a dual-model setup with YOLO and SAM. This new pipeline achieved 97% inspection accuracy and processed images six times faster, reducing the time from 12 seconds to just 1.5-2 seconds per image. The solution provided Cerulean with a reliable, high-speed system for quality control, strengthening their reputation for innovation and precision.

Cerulean