Energesman

Quarterly manual sampling offered a starting point, but it couldn’t capture waste composition changes between measurements. Small sample sizes meant the data often missed significant variations in what was actually coming through the facility.

Energesman was making significant infrastructure investments based on quarterly snapshots. When projects underperformed, the team suspected the underlying data wasn’t telling the whole story.

Industrial sorting lines present a difficult computer vision problem. Materials move constantly, lighting varies, and waste items overlap and obscure each other. Any AI solution would need to handle these real-world conditions reliably.

Softeta installed high-resolution industrial cameras across Energesman’s sorting lines, capturing diverse visual data under varying lighting and motion conditions. The team collected 10,000 video samples from livestream cameras in the actual factory setting.

Our team built a dataset of approximately 75,000 labeled images. Every sample required precise bounding boxes and object type classification. Categories included plastics (PP, LDPE, other), containers (alcohol bottles, glass jars, Tetrapaks, champagne bottles), and municipal collection bags.

Softeta trained deep learning models optimized for waste detection, then validated performance through F1-Confidence curves for threshold tuning, normalized confusion matrices to identify misclassifications, and visual inspections on real plant data.

The system deployed to cloud infrastructure, integrated directly into Energesman’s facility. It provides real-time detection feedback, live object counting, and anomaly alerts for non-conforming items – all accessible to operations staff as materials move through the plant.

Energesman now understands what’s actually coming through their facility continuously instead of quarterly. They can allocate investments more strategically and track whether changes produce expected results.

Automated classification handles the repetitive categorization work. Staff focus on exceptions and system oversight rather than manual sampling.

Transparent, accurate, continuous data produces more defensible compliance documentation than point-in-time samples.

The AI solution is ready for integration into Energesman’s upcoming robotic waste sorting system. This will further reduce manual labor and minimize errors caused by human limitations and workforce shortages.

With continuous data collection, Energesman plans to work more closely with the private sector, providing valuable market insights about waste composition trends.